Accessibility of Innovative Services in Radiation Oncology.

Chapter 1: Introduction to the Study

Introduction

The field of radiation oncology (RO) involves using high-energy radiation

to target cancer lesions while sparing the surrounding healthy tissues (American

Cancer Society, 2018). The field experienced rapid growth in the early 2000s,

with the development of innovative techniques, such as intensity-modulated

radiation therapy (IMRT). This technique allowed clinicians to “bend” radiation

and customize each patient treatment by creating concave dose distributions that

spare healthy tissues (Chen, 2014). This rapid growth paved the way for an array

of other innovative techniques, such as advances in imaging and oncology

informatics (Chetty et al., 2015).

Many fields in health care struggle to implement innovative techniques

such as these in the clinical workflow. The adaptation of innovation within

discrete organizations is well understood, yet how the entire health care system

transforms to accept health care innovations and thus improve population health is

abstruse (Parston et al., 2015). Morris, Wooding, and Grant (2011) estimated that

it takes 17 years for an innovation to spread throughout the health care industry.

This time lag is more prominent in low-resource settings, leading to inequalities

(Keown et al., 2014). In RO, differences in treatment modalities have a notable

impact on cancer survival, yet centers offering innovative treatment techniques

are inaccessible to part of the population, leading to health care disparities (Chen,

2014).

Jacobs et al. (2016) investigated the innovations adopted in Dutch RO

centers. The authors found that in the Netherlands, RO centers implement on

average 12 innovations per year. In reviewing the literature, there are no current

studies quantifying the diffusion of innovation in RO across the United States.

The purpose of this study was to fill this gap in the literature. University

affiliation and urbanicity were investigated as possible predictors. The role of

medical physicists was also investigated by using an array of demographic

parameters and context-specific antecedents to develop a predictive model for

accessibility of innovative services in RO in the United States.

There are numerous examples of innovative practices, policies, and

products that improve health care provision (Darzi & Parston, 2013). The

methods by which an innovation gains momentum and diffuses (or spreads)

throughout a specific population or social system over time is typically referred to

as diffusion of innovation (Rogers, 2003). Diffusion is a passive spread, as

opposed to dissemination, which is active (Rogers, 2003). Public health

researchers often overlook the possible lack of diffusion of innovation in efforts to

improve population health (Darzi & Parston, 2013). Developing a tool to measure

innovation in RO can be the first step in developing a predictive model for

improving the diffusion of innovation in the field. This quantitative assessment

may contribute to the improvement of accessibility and quality of RO services

across the United States, improve cancer outcomes, and thus contribute to positive

social change.

Background

Innovation and its implementation are actively studied across many

disciplines. The implementation of innovative methods in health care and how

these methods reach the general population is an active field of research. For

example, the researchers who undertook the Global Diffusion of Health Care

Innovation study investigated the effects of rapid adoption of innovations on

public health worldwide (Darzi & Parston, 2013). Parston et al. (2015) identified

three phases of change management: (a) creating a climate for change by having

vision, strategy, and a specific agency to promote diffusion, transparency, and

communication channels; (b) engaging and enabling the whole organization by

having incentives and rewards; and (c) implementing and sustaining by making

time and space for learning. The authors emphasize that implementing change in

health care is both an art and a science (Parston et al., 2015). Parston et al.

explored quantitative and qualitative parameters that affect innovative technique

implementation in a variety of socio-demographic environments. In health care,

the adoption of innovative techniques is not an academic exercise, it can save

lives.

Clinical effectiveness in RO is by necessity retrospective, and there have

been no long-term clinical trials for the majority of innovative techniques used in

health care. The adoption of new techniques is primarily based on reasonable

theoretical long-term benefits (Chen, 2014). Assessing cancer survivorship as a

whole using big-data analyses and defined value is still ongoing (Nardi et al.,

2016). Smith et al. (2017) discussed the definition of the value of innovation

implementation from multiple stakeholders and how scientific evidence enters the

marketplace of RO. Value can mean (a) societal value, such as lost wages, lost

productivity, and willingness to pay; (b) care value, such as disease control,

quality of life and long-term theoretical benefits; or (c) system value, such as

incremental costs, budget impact, and affordability. Nardi et al. (2016)

investigated the value of comprehensive cancer centers designated by the National

Cancer Institute (NCICCC). The authors found that for specific cancers, such as

hepatobiliary, lung, pancreas, gastric, breast, cervical, oral, and colorectal cancer,

patients had a 20-30% higher 5-year survival when treated at an NCICCC (Nardi

et al., 2016). NCICCCs are disproportionally located in the Eastern United States,

which means that for many patients, it is simply not an option to receive treatment

at these facilities (Nardi et al., 2016). Additionally, many health care plans do not

cover services at university hospitals or NCICCCs at all (Nardi et al., 2016). The

majority of Americans, therefore, do not have access to centers that provide better

cancer outcomes, either because of geographical limitations or because their

insurance does not cover treatment at institutions with proven better outcomes

(Nardi et al., 2016). Lack of access to care is one of the many challenges in the

contemporary United States, and there are no existing studies addressing this issue

in the field of RO, according to the review of the literature. This study measured

the accessibility of innovative services in RO in the United States and attempted

to fill this literature gap.

Problem Statement

The field of RO has experienced rapid growth in the past 25 years, with

technological advancement as the driving force. The need for innovation in health

care is broadly accepted as necessary, with current innovation considered

insufficient at this point (Parston et al., 2015). Implementing innovative

techniques requires significant resources, and community hospitals are trailing

behind (Nardi et al., 2016; Pfister et al., 2015). Fragmented data on innovation are

available from vendors and professional societies such as the American College

of Radiology (ACR) and the American Society for Radiation Oncology

(ASTRO), but there has been limited analysis of the barriers to developing and

implementing new technology in RO (Mayo et al., 2016).

The study of the mechanism by which innovations spread geographically

and across time is of particular interest in health care and is an active field of

research. These diffusion patterns are explained using various theoretical models,

one of which is the diffusion of innovation theory, as described by Rogers (2003).

Diffusion patterns can be assessed by sampling a variety of stakeholders. In this

study, the problem of diffusion of innovation was addressed from the point of

view of medical physicists (MPs), who are responsible for the acceptance,

commissioning, and implementation of innovative techniques in RO. MPs are

emerging into leadership positions in RO facilities across the United States, not

only as technical experts but as the leaders in quality management (Delis et al.,

2017). In the study, MPs were evaluated as the adopters of innovations in RO.

MPs are not passive recipients of innovation: They seek innovations, assess them,

find meaning in them, develop positive or negative opinions about them, gain

experience with them, complain about them, work around them, modify them, and

circle back to the beginning, beginning the process once again (Rogers, 2003).

The main adopter categories may be broadly defined as innovators, early

adopters, early majority, late majority, or laggards (Rogers, 2003). There is

extensive literature on how cognitive and social psychological antecedents affect

the individual’s adoption (Greenhalgh, Robert, Macfarlane, Bate, & Kyriakidou,

2004). The psychological antecedents included in this study were motivation,

appreciation, self-efficacy and leadership style.

As discussed in the “Background” section, innovative techniques are not

accessible equally by all patients, as they are typically available at higher rates in

university hospitals or urban centers (Pfister et al., 2015). This lack of access is

directly related to inferior survival outcomes (Nardi et al., 2016). Jacobs et al.

(2016) studied the implementation frequency of innovative techniques in the

Netherlands and found a wide range across all centers. They found differences

between university and community hospitals and emphasized the lack of national

recommendations for prioritizing innovations, setting goals, and societal

interventions as long-term solutions (Jacobs et al., 2016). There is a literature gap,

however, on the extent of accessibility of innovations in RO in the United States

and its impact on public health. The accessibility of innovative services in RO

across the United States as an indicator of public health outcomes was further

assessed in this study by exploring the role and involvement of MPs.

Purpose of the Study

The purpose of this study was to investigate the accessibility of innovative

services in RO in the United States and assess possible diffusion patterns. These

patterns were used to create a predictive model of the factors that may affect MPs’

role in the diffusion of innovation. An Internet-based survey was conducted with

a convenience sample of MPs practicing in RO in the United States. A

quantitative cross-sectional analysis of survey data was conducted to gain insight

on the accessibility of Food and Drug Administration (FDA)-approved innovative

techniques in RO. The dependent variable was innovation score, defined as the

number of techniques available to or purchased by a department divided by the

number of techniques used clinically, weighed by partial implementation factors.

Independent variables were geographic location; practice details (university

affiliation, reporting structure, size of physics group); demographics (age, gender,

DABR status, residency status, meeting attendance, and education level); and

context-specific psychological antecedents (motivation, appreciation, and

leadership).

Research Questions and Hypotheses

The research questions (RQs) and hypotheses were as follows:

RQ 1: What are the differences in accessibility to innovation in RO based on

location and type of practice in the United States?

RQ 2: What are the statistically significant factors (demographics, practice details,

context-specific psychological antecedents) that predict the accessibility to

innovation in the RO clinic?

RQ 3: What are the statistically significant barriers that MPs practicing in RO in

the United States face in implementing innovations?

Hypothesis 1a

H1a0: There is no statistically significant difference in accessibility to

innovation in RO between university and non-university hospitals.

H1a1: There is a statistically significant difference in accessibility to

innovation in RO between university and non-university hospitals.

Dependent variable: innovation score (continuous).

Independent variable: university affiliation (categorical, two levels: has

university affiliation, does not have university affiliation).

Hypothesis 1b

H1b0: There is no statistically significant difference in the accessibility to

innovation in RO between metropolitan and nonmetropolitan hospitals.

H1b1: There is a statistically significant difference in accessibility to

innovation in RO between metropolitan and nonmetropolitan hospitals.

Dependent variable: innovation score (continuous).

Independent variable: urbanicity (categorical: metropolitan,

nonmetropolitan).

Hypothesis 2

H20: There are no statistically significant factors predicting innovation

score.

H21: There are statistically significant factors predicting innovation score.

Dependent variable: innovation score (continuous).

Independent variables: reporting structure, size of physics group, age,

gender, DABR status, residency status, meeting attendance, education

level, leadership (categorical), appreciation, and motivation (continuous).

Hypothesis 3

H30: There are no statistically significant barriers affecting innovation

score.

H31: There are statistically significant barriers affecting innovation score.

Dependent variable: innovation score (continuous).

Independent variables: barriers (continuous).

Theoretical Framework for the Study

“Evidenced-based decision making” is a frequently used phrase in health

care fields, yet research shows a disconnect between strong scientific support and

diffuse clinical usage. Pedersen (2015) discussed that health care fields are

different in their adoption of innovations due to contextual sense-making:

engagement, materialization and scientification of innovations provides meaning

to users. Wisdom, Chor, Hoagwood, and Horwitz (2014) defined innovation

adoption as a complex quality improvement intervention and categorized the

mechanistic constructs for the individual to accept innovation and implement it in

their organization; adoption of innovations is a dynamic process. The unit in this

study was MPs. The MPs, as individuals, accept or reject innovations and in turn

facilitate their RO departments in fully utilizing innovations. The assessment of

the role the MPs play in the diffusion of new products and practices across the

modern RO clinic health care system in the United States was studied by

measuring diffusion patterns of these products and practices. Diffusion is defined

as “the process in which an innovation is communicated through certain channels

over time among the members of a social system” (Rogers, 2003, p. 5). Rogers

developed the diffusion of innovation theory in 1962 (Rogers, 2003). The theory

is based on principles of communication of participants in a social system

(Rogers, 2003). There are four elements that influence diffusion: the innovation

itself, the communication channels, time, and the social system (Rogers, 2003).

In health care, clinical practices are often labeled evidenced-based, yet

there is a significant gap between clinical practice and recent scientific

developments. Health care innovations are not simple to implement in most

clinical settings and require differential resource allocation across multiple levels

of the organization. Omachonu and Einspruch (2010) developed a health care-

specific conceptual framework involving various stakeholders such as physicians

and caregivers, patients, organizations, and regulatory agencies. Physician

acceptance, the complexity of innovation, partnerships and collaboration,

organizational culture, regulatory acceptance, and organizational leadership all

contribute to the probability of acceptance of an innovation (Omachonu &

Einspruch, 2010). In the field of RO, there has been a broad discussion on how

technological advances can lead to improved patient outcomes, and stakeholders

are encouraged to embrace innovation (Chetty et al., 2015). Yet there is no

framework to quantify or improve the accessibility of innovative techniques. For

this study, accessibility of innovative RO services was measured using the

innovation score, as described in the “Nature of the Study” and “Definitions”

sections that follow.

Nature of the Study

This study was a quantitative numerical assessment of the extent of

accessibility of innovative techniques across the United States. The dependent

variable was the accessibility of innovative techniques: how many end-users have

each technique, how many of them use it, and to what extent. It is common to

have partial implementations of innovative technologies (Smith et al., 2017). For

participants who did not use techniques that were available to them, influencing

variables were investigated. Correlating the accessibility with the type of hospital

revealed if community hospitals were trailing behind. Reviewing the zip code

entries relative to the innovation score revealed if rural areas were indeed less

likely to utilize innovative techniques. Age, gender, DABR status, residency

status, meeting attendance, and education level of the MPs and correlation with

the availability of innovative techniques revealed how these factors affect

innovation. This information is pertinent to overcoming real and perceived

barriers in technology implementation in RO.

In 2018, the AAPM directory listed approximately 7,500 members in total,

and approximately 5,500 members are listed as active in the United States (not

student, retired, associate, or corporate members) (American Association of

Physicists in Medicine, 2018). Seventy-seven percent of the membership reported

being active in RO (American Association of Physicists in Medicine, 2018); thus,

there were about 4,200 eligible MPs. The survey was posted on LinkedIn, the

MedPhysUSA user forum, and the AAPM Blackboard forum. The MedPhysUSA

is a listserv hosted by Wayne University. It is a highly active online community

for MPs, where surveys are frequently posted, with over 4,000 members. Users

must subscribe, and there are volunteer moderators. Unofficial surveys typically

garner an approximate 5-10% response rate and official surveys an approximate

30-35% response rate. (American Association of Physicists in Medicine, 2017)

Considering this response rate and the estimated number of AAPM members,

approximately 200 responses were expected for this survey.

Definitions

American Association of Physicists in Medicine (AAPM): A scientific and

professional organization composed of more than 8,000 scientists whose clinical

practice is dedicated to ensuring accuracy, safety, and quality in the use of

radiation in medical procedures such as medical imaging and radiation therapy

(American Association of Physicists in Medicine, 2019) .

Context-specific psychological antecedents: For the purposes of this study,

motivation (the internal process that makes a person move toward a goal),

appreciation (recognition), and leadership.

Diffusion of innovation: The study of the spread of new ideas and

technologies among individuals and groups (Rogers, 2003).

Diagnostic radiology (DR): The field of medicine that specializes in the

diagnosis of disease using ionizing and non-ionizing radiation (American College

of Radiology, 2011).

Intensity-modulated radiation therapy (IMRT): A subset of advanced

methods used in RO to manipulate the area of treatment and conform to the shape

of individualized tumors and spare critical organs (National Cancer Institute,

2011).

Innovation: For the purposes of this study, innovation will be broadly

defined to cover products (for example, new technology, inventions, drugs, etc.);

practices (ways of working, clinical protocols, workforce changes, etc.); and

policies (those things that regulate/influence the use of products and practices).

The idea that an innovation is “new” is relative: It is defined as previously

unknown to the relevant unit, not new on an absolute time scale (Rogers, 2003).

Innovation Score: The total score of measured innovation indicators for

product, technology, and organizational innovation. This is a continuous variable.

Innovation Utilization Score: The total score of measured innovation

utilization indicators for product, technology, and organizational innovation. This

is a continuous variable.

Medical oncology (MO): The field of medicine that specializes in the

treatment of malignant cancers using pharmaceutical agents (National Cancer

Institute, 2011).

Radiation oncology (RO): The field of medicine that specializes in the

treatment of malignant cancers using ionizing radiation (National Cancer Institute,

2011).

Assumptions

There are inherent assumptions for this study. As primary data were

collected, it was assumed that enough MPs will participate in the study to achieve

statistical significance or at least make the results as representative as possible. It

was also assumed that participants answered truthfully to all the questions of the

study. Further, it was assumed that MPs play a vital role in the diffusion process.

Scope and Delimitations

It is important to note the difference between availability and accessibility.

Availability of innovative techniques is undoubtedly essential. After all, if

something is not available at all, it cannot be used and implemented. Yet the

availability of innovative techniques in RO would be more of interest in

marketing studies, not public health. The focus of this study is on accessibility, in

the context of equitable distribution when taking into account demographic

composition, urbanicity and under-served populations (World Health

Organization, 2018).

The survey was sent to the entire database of the MedPhysUSA listserv,

the AAPM Blackboard, and also posted on LinkedIn. Only MPs practicing in RO

were included. Participants practicing in U.S. territories or internationally were

excluded. Participants with both Master’s and Doctoral degrees were included.

Participants were assumed to be fluent in English.

Limitations

Selection Bias

A possible source of selection bias is the exclusion of physicists who are

not participating in the online forums mentioned in the “Nature of the study”

section. This possible selection bias was minimized by using LinkedIn in a

snowball fashion, to attempt to reach MPs who do not participate at the

MedPhysUSA or AAPM Blackboard. Similarly, MPs in rural community

hospitals may have been too embarrassed to participate, if they feel the

technology they are utilizing is not adequate. Both of these could be a serious

source of error, as the MPs in the areas with the most need for new technology

may not participate. To minimize selection bias, results were reviewed in total and

in a weighted average from university and community centers, based on known

proportions, when needed.

Information Bias

One of the possible sources of information bias in this study was recall

bias. Szklo and Nieto (2014) define recall bias as inaccurate recall of past

exposure. In the context of this study, the respondents may have inaccurately

recalled the exact details of the implementation of new techniques. The recall bias

may be entangled with social desirability bias. MPs are highly skilled and highly

paid professionals, and admitting that they have something available but not using

it clinically may be ignominious. This may have influenced them to report that

they are using a technique, or that they have been using it more than what is

actually the reality. Questions were posed on a sliding scale in an attempt to make

it more likely for the respondents to answer more truthfully.

Confounding

University affiliation and urbanicity were assessed as possible

confounders for the study. All these factors may affect both the exposure and the

outcome separately, and they are not directly in the causal pathway.

Statistical Limitations

Based on the expected participant number, 5% chance of Type I error and

20% chance of type II error is reasonable. The statistical power of 80% is a

reasonable initial goal as well Type II errors, not detecting an effect that is

actually there, or false negative, have more significant social implications than

Type I errors in this study. Not detecting an under-performance in community and

rural RO centers would lead to lack of supplemental efforts from AAPM to boost

their utilization of new techniques. This would be a disservice to the patients as

discussed in the section below.

Significance

There is a limited ability for hospitals to purchase innovative techniques,

as RO medical devices are extremely expensive. Even though there is insurance-

based reimbursement, the capital investment is beyond reach for many hospitals.

For example, a linear accelerator costs $1.5 million to $6 million, and the

innovative techniques under consideration cost $50,000 to $500,000 (Modern

Technology, n.d.). These capital investment amounts require years of strategic

planning and significant budget amounts. Since RO treatments are delivered daily

over several weeks, patients commonly select facilities close to their area of

residence (Pfister et al., 2015). Innovative techniques provide an improvement in

survival; thus disparities in accessibility to these innovative techniques is a public

health problem.

Survival rate numbers have significantly improved over the past ten to

twenty years on the global level, and the reason is largely attributed to improved

targeting methods, through the use of innovative techniques such as IMRT

(Cancer Research UK, 2015). Providing RO care of the highest quality to all

patients by improving the utilization of innovative FDA-approved techniques can

contribute to the improvement of accessibility and quality of RO services across

the United States and subsequently improve cancer outcomes. Better health

outcomes are directly related to a gain in hours worked and an increase in

individual income up to 30% (Marquez & Farrington, 2013). These results are

even more pronounced in low-income settings, when health costs may be out of

pocket. Declining health costs is often a reason quoted for crossing the poverty

line (World Health Organization, 2010). Healthier people are more efficient in the

workplace, and this affects government and commercial sector alike (General

Electric, 2014). Innovative techniques are available, but may not be accessible to

all patients. There are enormous inequalities within and between nations in health

care in general and the public health effects of this at the global level are being

investigated by the Global Diffusion of Health care Innovation study (Darzi &

Parston, 2013). This public health problem is in alignment with the ninth essential

public health service to "evaluate effectiveness, accessibility, and quality of

personal and population-based health services" (Centers for Disease Control and

Prevention, 2017). There are currently no studies measuring the accessibility of

innovative services in RO in the United States. Measuring the accessibility of

innovative services in RO and developing a predictive model to offer a foundation

to improve it is a contribution to positive social change.

Summary

The National Cancer Institute (2018a) estimated that in 2018 alone, there

were 1.7 million new cancers and approximately 609,000 deaths due to this

disease in the United States. Cancer incidence in the United States is 439.2 per

100,000 people, and mortality is 163.5 per 100,000, with higher mortality in men

(196.8 per 100,000) than women (139.6 per 100,000). Mortality is highest in

African American men (239.9 per 100,000). Throughout their lifetimes, 38.4% of

Americans will be diagnosed with cancer, leading to $147.3 billion in national

expenditures for cancer care. As the population ages, the number of new cancer

cases per year is predicted to be 23.6 million by the year 2030. There are

quantitative demands, such as the number of cancer centers and relative staff

(Aneja & Yu, 2012; Yang et al., 2014), and the challenges of rural cancer care in

the United States is well documented, with multiple studies correlating driving

distance to RO center with treatment decisions and outcomes (Charlton,

Schlichting, Chioreso, Ward, & Vikas, 2015).

There are approximately 2,500 RO centers in the United States, 5,500

radiation oncologists, and 4,000 MPs specializing in RO (Ballas, Elkin, Schrag,

Minsky, & Bach, 2006). Little is known, however, about the different techniques

used in every RO center. While the majority of patients are treated using national

standards (National Comprehensive Cancer Network, n.d.), the guidelines provide

a wide range of options, allowing customized clinical decisions.

Even though products may receive FDA approval, little is known about

their implementation in the clinical setting. It is well demonstrated that the

majority of health care spending is due to the cost of technological innovations

(Dybczak & Przywara, 2010). In turn, there are many publications on how

specific innovations improve output, such as survival, toxicity, safety, service,

efficiency, or cost-effectiveness. In the next chapter, a literature review was

conducted on the diffusion of innovation in health care in general, and RO

specifically. Systematic reviews and cross-sectional studies were reviewed.

Results were thematically synthesized to the diffusion of innovations in health

care in general, diffusion of specific innovations in RO, effects of innovations on

value, effects of innovations in quality and patient outcomes, and social impact.

Chapter 2: Literature Review

Introduction

The purpose of this study was to investigate the accessibility of innovative

services in RO in the United States and assess possible diffusion patterns. These

patterns were used to create a predictive model of the factors that may affect MPs’

role in the diffusion of innovation in RO. From the literature review presented in

this chapter, it is evident that RO has a push-pull relationship with innovation.

Innovation is readily available, yet not extensively utilized, at least not in all

settings. For innovative technologies that are purchased as part of a package, as

well as those with no upfront costs, health care professionals appear hesitant to

undertake implementation. To some extent, the hesitancy to embrace innovation is

understandable, as outcomes have been improving using traditional techniques

(Chen, 2014; Jagsi et al., 2012; Smith et al., 2017). For practitioners in suburban

or rural areas, who are not in direct contact with opinion leaders and research

findings, it may indeed appear this way that outcomes are improving with

continuous use of traditional techniques. Even though errors leading to patient

death exist in all medical fields, in RO there are many catastrophic errors that can

occur on any day of the long treatment path, typically lasting weeks. Despite the

importance of this topic, there are currently no assessments of the diffusion of

innovation in RO in the United States. This study will contribute to filling this

literature gap.

The chapter begins with an overview of the literature search strategy,

theoretical foundation, and methodology used in the study. In the literature review

that follows, the theoretical foundation of the study, diffusion of innovation, and

how it applies to health care and RO are presented. Previously published

systematic reviews and cross-sectional studies were used to develop a thematic

synthesis of key literature for the review. Topics include the diffusion of

innovations in health care in general, diffusion of specific innovations in RO,

effects on value, effects on quality and outcomes, social impact, and predictive

parameters.

Literature Search Strategy

Walden University Library and Google Scholar were used to conduct

searches of the literature. Search terms included radiation oncology, oncology,

diffusion of innovation, early adopters, medicine, health care, radiology, equity,

and public health. Key words were based on key concepts in previously identified

published reviews. Diffusion of innovation is sometimes referred to by other

terms, and thus, similar key word items were also used, such as innovation,

adoption rate, new technology, and new technology accessibility. The key words

were combined in various ways using Boolean operators in context--for example,

early adopters in radiation oncology, new technology in radiation oncology,

diffusion of innovation in healthcare, new technology in radiation oncology

innovation, health equity, and accessibility to new technology. After the initial

review of the results, further parameters were identified that were considered to

possibly be influential. Additional searches were conducted using the following

terms: academic, academic cancer centers, university hospital, community

hospital, rural/rurality, city/urban/urbanicity, demographics, reporting structure,

employment type, appreciation, motivation, and leadership. These were combined

with innovation, adoption rate, new technology, and new technology accessibility

using Boolean operators as described. Only articles published in English were

included. The date range was limited to articles published since 2013, with the

exception of seminal works on the theory of diffusion and well-established

psychological context-specific antecedents.

The various search parameters produced over 200 results. These were

screened manually and articles were excluded when not directly pertinent.

Additional studies were identified by following the references. This step was

performed by using the “cited by” function in Google Scholar and by manually

looking up references within the selected articles, when applicable. Due to the

narrow focus on RO, the search was expanded to other fields of medicine and

health care. Studies were included when they addressed diffusion of innovation

patterns of clinical innovations in other medical fields, especially DR and MO,

which are the closest fields to RO. Results were thematically analyzed as

described in the following sections.

Theoretical Foundation

Health care practitioners have a long and beguiling history of accepting or

rejecting innovations. The basic principles of hygiene were developed from

Hippocratic medicine in ancient Greece. These principles may be considered

elementary today, but they were innovative at the time. They received

considerable criticism and skepticism and, to this day, are not accepted in certain

parts of the world (Tountas, 2009). More recently, the resolution of scurvy was

much delayed in the 18th century, as the British Royal Navy stubbornly refused to

accept and diffuse the practice of vitamin C supplementation. The fable may be

well known to public health practitioners, but the details of the delayed

implementation circa 1793-1800 as described by Vale (2008) paint a picture full

of political intrigue. The influence of a few forward-thinking physicians was the

catalyst needed for the widespread adoption of vitamin C against scurvy (Tountas,

2009). Public health history is full of similar examples, from the distant past until

present day.

Schumpeter (1989) described the continuous generation of innovations as

“creative destruction” (p. 83) and concluded that adaptation to innovation is the

rule, not the exception. Increasing health care costs, aging populations, and more

demanding consumers are compelling organizations to offer innovative solutions

(Herzlinger, 2006; Varkey, Horne, & Bennet, 2008). However, less than 50% of

all evidence-based practices are effectively implemented in the health care system

(Alexander & Hearld, 2011; M. Jacobs et al., 2017). Innovation implementation is

an area that needs improvement in health care in general and in RO as well.

Diffusion of Innovation

Rogers first introduced diffusion of innovation theory in 1962, while

studying rural sociology in the Midwestern United States and the adoption rates of

hybrid seeds in the area (Rogers, 2003). He combined over 500 studies from a

variety of fields, including anthropology, early sociology, rural sociology,

education, industrial sociology, and medical sociology, and the synthesis was his

seminal diffusion of innovation theory (Rogers, 2003). The theory has expanded

into social network analysis and is currently frequently used in communications,

marketing, development studies, health promotion, organizational studies,

knowledge management, conservation biology, and complexity studies,

particularly in the medical field and health communication studies (Rogers, 2003).

There are five stages in implementing an innovation, as described by Rogers

(2003):

1. Knowledge. The individual is exposed to an innovation for the first time

but does not have significant knowledge about the specifics of the

innovation. There is not yet any motivation from the individual to pursue

the innovation.

2. Persuasion. The individual is actively interested in the innovation and is

beginning to consciously explore related information/details.

3. Decision. The individual begins to study the advantages and disadvantages

of using the innovation and decides whether to either adopt or reject the

innovation. This stage is the most individualistic and thus the most

difficult to collect empirical evidence on.

4. Implementation. The individual begins to use the innovation to a smaller

or larger degree, depending on the situation. The individual also

determines their personal opinion on the usefulness of the innovation and

may research additional information on the innovation.

5. Confirmation. The individual finalizes the decision to continue using the

innovation. This stage is both intrapersonal and interpersonal,

confirmation the group has made the right decision.

The degree by which innovation is accepted by individuals depends on

both the individual’s characteristics and on the characteristics of the organization

in which the innovation is going to take place and typically follows an S-curve

(Rogers, 2003). Rogers classified individuals within a social system as being in

one of five categories:

1. Innovators. This group is willing to take risks, has the highest social

status, has financial liquidity, is social, is in touch with scientific

sources, and has frequent interaction with other innovators. They have

high-risk tolerance for adopting new technologies.

2. Early adopters. These are typically the opinion leaders among all the

categories. They may have higher social status, financial freedom,

higher education and may be more socially open than late adopters.

They are more thoughtful in adoption choices than innovators.

3. Early majority. This group adopts an innovation after significantly

larger amount of time than the innovators and early adopters. They

may have above average social status, contact with early adopters and

are less likely to hold positions of opinion leadership in a system.

4. Late majority. This group adopts an innovation much later than the

average participant. They view innovation with cautiously and with a

high degree of skepticism and even after the majority of society has

adopted the innovation. The Late Majority group are below average

social status, have little financial liquidity, in contact with others in

late majority and the early majority and little opinion leadership.

5. Laggards. This group is the last to adopt an innovation. Contrary to the

other categories, Laggards show little to no opinion leadership.

Individuals belonging to this group typically have a strong dislike for

change-agents. Laggards typically tend to be focused on "traditional

ways," have the lowest social status, lowest financial liquidity may

typically be older than adopters, and are in contact with only family

and close friends.

Rogers (2003) recognizes five qualities that determine the success of an

innovation.

1. Relative advantage. This is the degree to which an innovation is

perceived as better than what is currently available, relative to the

actual users, for example in terms of economic advantage, social

prestige, convenience, or satisfaction. If the innovation has a high

perceived relative advantage, it will get adopted at a more rapid pace.

Relative advantage is subjective and may carry a different meaning for

different individuals or groups. It depends on the particular needs and

dynamics of the user group.

2. Compatibility with existing values and practices. This is the degree to

which an innovation is perceived as being consistent with the values,

past experiences, and needs of potential adopters. If innovation is

comparable with the group's values, norms or practices, it will be

adopted more rapidly.

3. Simplicity and ease of use. This is the degree to which an innovation is

perceived as difficult to understand and use. If an innovation is simple

and easy to understand it will be adopted more rapidly than

innovations requiring increased understanding and new skills.

4. Trialability. This is the degree to which an innovation can be

experimented with on a limited basis. If innovation can be trialed out

by the individual considering it, without being bound in it, it will be

more likely that the individual will consider it.

5. Observable results. If the results of an innovation are easily visible to

users, they are more likely they are to adopt it. When the results are

clearly visible, there is lower uncertainty and increased peer discussion

of a new idea, as more people seek information about the innovation.

According to Rogers (2003), these five qualities determine between 49 and 87%

of the variation in the adoption of new products.

Reinvention is one of the leading principles in the diffusion of innovation.

The success of an innovation is highly dependent on how well it evolves to meet

the needs of the more risk-averse individuals in a population. The innovation does

not have to be new in an absolute sense; it just has to be new to the individual.

Organizational and Personal Antecedents

Organizational and personal characteristics dynamically influence the

outcome of successful innovative behavior. The organizational and personal

constructs related to innovation are increasingly investigated across many

research fields such as information systems, psychology, organizational and

management science and multidisciplinary science (Najaftorkaman, Ghapanchi,

Talaei-Khoei, & Ray, 2015). The organizational constructs affecting innovation

were summarized by Wisdom et al. (2014) to be absorptive capacity, leadership

style, networking, culture, size and structure, social climate, social network,

training readiness and effort, and traits and readiness for change. The individual

characteristics affecting innovation were summarized by Wisdom et al. (2014) to

be affiliation with organizational culture, attitudes, motivation, readiness for

improvement and reward, feedback on execution and fidelity, awareness, social

networking, knowledge/skill, competence, and demographic factors. Context-

specific psychological antecedents, such as appreciation, motivation, and

leadership are also well-studied to influence the individual’s response to

innovation (Greenhalgh et al., 2004; Pedersen, 2015; Wisdom et al., 2014).

Adoption and Diffusion of Innovation in Health Care

The adoption and diffusion of innovative services in health care is distinct

from other fields. The lack of effective implementation of innovations in health

care affects the lives of patients. Reinhardt, Hietschold, and Spyridonidis (2015)

analyzed Roger’s classical diffusion of innovation framework and how it pertains

to health care. The authors investigated the five qualities that determine the

success of an innovation mentioned before and found them to be valid in the

health care setting: relative advantage translates to focus on “hard” numbers,

compatibility translates to leaving non-core processes unchanged, complexity

translates to using easy-to-use innovations and shifting the point of care,

trialability translates to making parts of the innovation trialable in a small scale,

and observability translates to successful communication of outcomes to all

relevant stakeholders. The authors also pointed out the distinction between

individuals and organizations as highly important in the health care setting.

Individuals may be innovative in one or many domains, thus making them more

likely to adopt an innovation in their health care role, but they can also be

resistant when confronted with barriers. On the other hand, organizations are

motivated by gains and resistance on the organizational level may become evident

when the implementation of an innovation is misinterpreted to coincide with

individual or group loss of power or status. The authors concluded that the various

stakeholders in health care form a complex network, and a wide array of different

professions needs to collaborate to ensure widespread use of an innovation. They

also emphasized that formal and informal partnerships are often needed and that

champions or “celebrity hospitals” are influential in innovation adoption and

diffusion.

Methodology

A broad review of the selected articles was conducted. The methods used

can be categorized into two main groups: systematic reviews, cross-sectional

studies, and surveys.

Systematic Reviews

In systematic reviews, authors collect and summarize previously published

studies and present it in a concise format. Even though there is no clear definition

of what constitutes a systematic review, the main characteristics are typically (a)

clear objectives with specific methodology that can be reproduced by others; (b)

an exploration of existing literature including methodology to attempt to find all

possible studies fitting the eligibility criteria; (c) validity assessment of the studies

included, for example for bias etc; and (d) a presentation of the results as a

synthesis (Moher et al., 2015). The use of systematic reviews is increasingly

common, and the quality of these reviews is improving through reporting

standardization, especially after the adoption of the Preferred Reporting Items for

Systematic reviews and Meta-Analyses for Protocols (PRISMA-P). Strengths of

systematic reviews include more precise estimates of effects than those derived

from individual studies and a cohesive approach to the research problem. The

main limitations are possible erroneous conclusions if all studies were not

identified. For example, Parston et al. (2015) did a systematic review of

successful innovations in health care with a significant impact on public health

and assessed the factors affecting rapid adoption. In another example, M. Jacobs

et al. (2017) conducted a literature review on the frequency of output evaluation

of innovations in RO. Output was defined as survival, toxicity, safety, service,

efficiency or cost-effectiveness. The authors looked for publications in three

ways: innovations in general health care, radiotherapy-specific innovations, i.e.,

organizational innovations and general implementation of innovations, and

innovations per tumor group/radiotherapy technique. Smith et al. (2017) did a

review of the current appropriateness of the use of advanced technologies for

radiation therapy and surgery in oncology. They focused their study on the

definition of the value of innovation implementation from multiple stakeholders

and how scientific evidence enters the marketplace of radiation oncology. The

authors specifically studied the adoption rate of two distinct innovative

techniques, IMRT in the 1990s and early 2000s and proton therapy in the 2000s.

Some of the articles cited in these systematic reviews are not explicitly mentioned

in this section, yet will be used in the synthesis section below.

Cross-Sectional Studies

Cross-sectional studies are a type of observational study that analyzes data

from a population or population sample at a specific point in time and draws

conclusions on the state of affairs at that moment in time. Strengths of this type of

study include a relatively low cost, rapid completion, and relative ease to achieve

large sample sizes. Limitations of cross-sectional studies are the inability to

determine causation and follow through to develop changes in the observed

patterns (Cresswell, 2018). For example, Jacobs et al. (2016) studied the degree

of innovation routinely implemented in the Netherlands by doing a descriptive

cross-sectional study. The authors used semi-structured interviews to collect

information on product innovation (number of introductions of new or

significantly improved treatments, number of new positioning devices, percent of

patients on phase I-II randomized trials, percent of patients in phase III trials),

technological innovation (frequency of implementation of new medical devices,

number of products purchased), market innovation (percentage of patients from

outside the market area, percent of referring hospitals) and organizational

innovation (new practices for organizing procedures, new methods for organizing

work responsibilities, new methods for organizing external relationships with

organizations or public institutions). Pfister et al. (2015) used Surveillance,

Epidemiology, and End Results (SEER) Medicare data to study risk-adjusted

mortality by cancer category in the Medicare population and investigated the

difference in outcomes based on university affiliation status. Pollack, Soulos, and

Gross (2015) studied peer exposure the adoption of a new cancer treatment

modality (breast brachytherapy) by using SEER data and correlated the exposure

of early brachytherapy adopted among non-early adopters by investigating their

shared patients. Nardi et al. (2016) did a cross-sectional analysis of five-year

survival between specialty and non-specialty cancer hospitals. Keown et al.

(2014) did a two-phase qualitative and quantitative cross-sectional study in eight

countries (Australia, Brazil, England, India, Qatar, South Africa, and the United

States) and compared cultural factors enabling health care innovation.

Surveys

While there is no research done on validating web-based surveys

specifically on MPs, Dykema, Jones, Piché, and Stevenson (2013) analyzed the

use of web-based surveys in clinicians and other health care providers and

concluded to achieve good response rate, high incentives are needed. The authors

summarize their findings to equate web-based surveys to traditional ones,

assuming a rigorous methodology and clarity in result reporting.

Literature Review Related to Key Concepts and/or Variables

Results of the literature review were thematically synthesized. Diffusion

of innovations in science and health care in its entirety is presented. Subsequently,

RO-specific implementations are assessed. Each of the following categories is

then presented as separate entities: effects on value, quality and outcomes, social

impact and predictive parameters.

Diffusion of Innovation in Science

Innovation and science have a tight-knit relationship. The terms science,

technology, and innovation are frequently used interchangeably. For example, the

United Nations Educational, Scientific and Cultural Organization (UNESCO) has

an entire program devoted to science, technology, and innovation, as means for

economic development and social progress (UNESCO, 2018). In the classic linear

models discussed by Eizagirre, Rodríguez, and Ibarra (2017), there is an

expectation that funding science will lead to innovation, through technological

advancements. However, this narrow approach fails to include socio-economic

context and the role of government in allowing deviation from traditional models

so mutually responsive co-responsibility can emerge between social actors.

Eizagirre et al.(2017) concluded that the relationship between science and

innovation is converging to inertia and a new paradigm is needed. The blurring of

the lines between neoliberalism and economization in the United States is

blatantly apparent when reviewing the accessibility of innovative services in

health care (Berman, 2014).

Diffusion of Innovation in Health Care in General

Over the past twenty years, there has been an exponential increase in

innovative policies, products, and practices aiming to increase life expectancy,

quality of life, effectiveness of treatment, efficiency of treatment, and equity of

delivery (Frenk, 2013). How innovations are actualized to improve public health

is often referred to as translational research. The time lag between an introduction

of an innovation and its diffuse clinical practice was estimated to be on average

17 years, however the way this time lag is measured varies depending on content

and exact research questions (Morris et al., 2011). It is logical to attempt to

shorten this time so the benefits can be maximized; in fact there is an extensive

body of literature on the individual and organizational characteristics that promote

acceptance and proliferation of innovation in health care (Parston et al., 2015).

Different countries are adopting different strategies to do so. Darzi and Parston

(2013) studied the local context of spreading innovations for eight different

countries and found substantial differences in cultural dynamics. In the United

States, the authors found that standards and protocols along with incentives and

rewards were the most important enabling factors for innovation, followed by

funding for research development and diffusion, communications channels across

and beyond health care, transparency of findings and data demonstrating success

and informatics. Vision and strategy, along with specific resources to identify and

promote innovation were ranked as the least important parameter for the diffusion

of innovation among expert users on the panel (Darzi & Parston, 2013).

The need for innovation in health care is broadly accepted not to be

sufficient (Harris, Bhatti, Prime, del Castillo, & Parston, 2018; S. R. Jacobs et al.,

2015). There is general agreement that innovation implementation improves

patient outcomes (Daniels & Capouya, 2017; Parston et al., 2015). There are

numerous examples of adverse patient outcomes due to ineffective

implementation of innovation in the health care setting (Reinhardt et al., 2015).

The majority of criticism against innovation is related to high costs; innovation is

often described as being too costly, as research and development alone cost

billions. These simplistic approaches often underestimate the complexity of cost-

benefit and the lower downstream costs of improved health at the individual and

societal level (Dzau, Asch, Hannaford, Aggarwal, & Pugh, 2017). There needs to

be a distinction, however, with low-cost yet proven approaches, especially at low

resource settings; high-cost innovative approaches are only beneficial if they

outperform their pre-existing solutions and are affordable enough to lead to

improved population health. Innovations need to sustainably show benefits to be

meaningful (Dzau et al., 2017).

The health care setting provides an excellent field to emphasize the

difference of the different levels that innovation can take place in. In their seminal

work on health care diffusion of innovation, Greenhalgh, Robert, Macfarlane,

Bate, and Kyriakidou (2004) make a health care industry-specific distinction on

the adoption by individuals and adoption by organizations. Organizational

innovativeness is affected by the administrative intensity, centralization,

complexity, internal and external communications, formalization, functional

differentiation, managerial attributes toward change, managerial tenure,

professionalism, slack resources, specialization, technical capacity, and vertical

differentiation. The individual is influenced by general antecedents (tolerance of

ambiguity, intellectual ability, motivation, values, and learning style) and context-

specific psychological antecedents (values, goals, specific skills, etc.). The

majority of publications focus on the science of diffusion, and few on the art of

change (Pedersen, 2015). There are also significant differences in how innovation

is implemented in low-resource settings (Harris et al., 2018).

Diffusion of Innovation in RO

RO prides itself as a high technology field. Linear accelerators, atomic

particles, and algorithms work harmoniously to cure cancer. There is currently

extensive usage of many advanced cutting-edge technologies: (a) computational

advances using graphics processor units (GPUs), cloud-based methods, and

parallel processing have improved calculation speed and are being used for

automated knowledge-based treatment planning; (b) high performance imaging

systems such as positron emission tomography (PET) allow tumor identification

and radiotherapy response measurement, and onboard magnetic resonance

imaging (MRgRT) improves real-time monitoring; (c) heavy particle treatments

better spare healthy tissues, especially in pediatric populations. However, the field

has not universally accepted these innovations (Chetty et al., 2015) and little is

known about the factors that may influence groups and individuals (Pollack et al.,

2015). This may be partly affected by the fact that few publications include the

term "innovation," while in reality, they are actually describing an innovative

technique (M. Jacobs et al., 2017).

M. Jacobs et al. (2015) applied the Delphi method to determine indicators

for innovation in RO and derived 13 indicators in four categories:

 Product innovation: number of introductions of new or significantly

improved treatments, number of new positioning devices for patient

treatments, number of approved patents, percentage of patients in

phase III randomized trials approved by an IRB, percentage of patients

in phase I-II trials approved by an Institutional Review Board (IRB)

 Technological innovation: frequency of implementation of new

medical devices, number of products for which royalties have been

obtained or which have been sold to the industry, number of regulatory

agency approved marked products that have been produced by the

department (for example regulated by the Food and Drug

Administration (FDA), or the Conformité Européenne (CE))

 Market Innovation: percentage of patients from outside the market

area, number and percentage of new general hospitals that refer to the

desired patient population.

 Organizational innovation, new practices and for organizing

procedures, new methods of organizing work responsibilities and

decision making, new methods of organizing external relationships

with other organizations or public institutions

The authors suggest that these indicators are most useful when measured

in the past three calendar years. These indicators are useful in an attempt to study

innovation in RO. Yet, there is currently no study that systematically quantified

the diffusion of innovation in RO in the United States. Innovation implementation

has been more successfully quantified in the Netherlands with a small number of

centers. Jacobs et al. (2016) studied 15 Dutch RO cancer centers (75% of the total

in the Netherlands) and found that they implement on average12 innovations per

year (range 5-25). The authors will attempt to repeat at three-year increments. The

average number is sufficiently large and the authors concluded that Dutch

radiotherapy centers are highly accepting of innovation. The wide range,

however, suggests a non-homogenous distribution, which the authors did not find

to be attributed to academic status. The authors concluded that systematic

collaboration between centers and a national registry would be beneficial to

improving innovation implementation even further.

Value of Innovation

The definition of value in health care is not very clear. As mentioned

before, some of these innovations are extremely expensive and frequently out of

reach for standalone community hospitals. Even though one can argue that the

investment in innovative technologies translated to increased income for the

providers and hospitals and to increased quality-adjusted life-years, the math is

not straightforward in RO. In MO, the development of new pharmaceuticals may

increase the cost of treatment for new, promising chemotherapy agents directly

improves outcomes (Roubou & Alexopoulou, 2015). In RO, innovations need

long-term investments to sustain the appropriate use of the technology in the form

of upgrades, service contracts, staff education, etc (Smith et al., 2017). There is

also a broad discussion of reimbursement for innovative techniques. If the

innovative technique has an associated financial motive, then the decision to

implement it may be spurious at best. The higher reimbursement rates for IMRT

discussed previously were certainly a contributing factor to its fast and successful

diffusion across the United States. Conversely, some of the innovative techniques

available reduce the number of total treatments and the overall provider and

hospital reimbursement. A common example is the use of hypo-fractionated

versus conventional whole breast irradiation in the United States: despite

supportive 10-year data from the United Kingdom, only 11% of eligible patients

received this shorter and less costly regiment in 2008 and 35% in 2013 (Bekelman

et al., 2014). In the absence of financial incentives, providers are less likely to

adopt innovative treatments that may improve outcomes but reduce their direct

financial benefit. In health care in general, value can mean societal value, such as

lost wages, lost productivity, and willingness to pay, care value, such as disease

control, quality of life and theoretical long-term benefits, or system value, such as

incremental costs, budget impact and affordability (Smith et al., 2017). All these

three constitute a cost-effective ratio to consider an innovative technique to be of

better value than the established technique, and there is little agreement on the

calculation of this ratio. Usage of lower cost but better techniques does have

successful examples from DR. The Joint Task Force on Adult Radiation

Protection was created by the collaboration between the American College of

Radiology and the Radiological Society of North America with the purpose

addressing the public’s concern about exposure to ionizing radiation from medical

imaging. The Joint Task Force collaborated with the American Association of

Physicists in Medicine and the American Society of Radiologic Technologists to

develop standards and launch an extensive awareness campaign under the slogan

“Image Wisely” for adults and “Image Gently” for pediatric patients (American

College of Radiology, 2011). These efforts are focused on optimizing existing

techniques to limit unnecessary exposure, and often inadvertently highlight the

limitations of older equipment. Approaches such as this provide a clear statement

to the market and to practitioners that they need better and different innovations.

Effects on quality and outcomes. In the field of RO, innovative

techniques are often implemented based on the “as low as (is) reasonably

achievable” principle (ALARA). This principle means “making every reasonable

effort to maintain exposures to ionizing radiation as far below the dose limits as

practical, consistent with the purpose for which the licensed activity is

undertaken, taking into account the state of technology, the economics of

improvements in relation to state of technology, the economics of improvements

in relation to benefits to the public health and safety, and other societal and

socioeconomic considerations, and in relation to utilization of nuclear energy and

licensed materials in the public interest” (Nuclear Regulatory Commission, 2018).

In RO, innovations are frequently implemented using this principle, in the

absence of clinical trials, and there is broad agreement that this improves quality

and outcomes (M. Jacobs et al., 2017). While clinical trials are typically the gold

standard for evidence-based approaches (Daniels & Capouya, 2017), the

implementation of innovative techniques using ALARA is extensive in RO

(Chen, 2014). Since there are no clinical trials to drive nationwide

recommendations, users can choose their own methods on how to treat patients,

within the broad standards described by the National Comprehensive Cancer

Network (National Comprehensive Cancer Network, n.d.). This leads to

significant differences in outcomes for different populations and leads to the

conclusion that quality of cancer care in the United States is inconsistent, with

patients treated at specialty cancer hospitals having an adjusted 10% lower

mortality at one year than patients treated at community hospitals (Pfister et al.,

2015). This inequality to access to appropriate cancer care may not be solely due

to innovations, but it is certainly a parameter (Nardi et al., 2016). Additionally,

there is little innovation implementation in using big data in RO. Since the advent

of electronic medical records (EMR), there has been little or no standardization of

specific field entry. The most basic example is the radiation prescription: after ten

years of discussion, there is barely a draft in progress (personal communication,

RO-SSI group, July 16, 2019). This makes data extraction and comparison

extremely difficult and often meaningless. Data elements such as survival,

recurrence, diagnosis and staging, provider-reported toxicities, dosimetric data

from delivered plans, and use of innovative technologies (such as breath-hold,

image-guided radiation therapy, immobilization devices) are largely missing as

elements in most EMRs. Extracting this data automatically would lead to self-

evaluation and scoring relative to national standards, thus improving the adoption

of standards (Mayo et al., 2016).

Social impact. Different cancer care treatments especially in

hepatobiliary, lung, pancreas, gastric, breast, cervical, oral, and colorectal cancer,

where patients five-year survival is higher by 20-30% for patients treated at

comprehensive cancer centers designated by the National Cancer Institute

(NCICCC) (Nardi et al., 2016). These NCICCCs are disproportionally located in

the eastern United States and major cities across the country (National Cancer

Institute, 2018b). The average American living in a rural area has to travel an hour

to reach specialized oncology care, and an hour and a half to reach a university

hospital (Charlton et al., 2015). Many patients live so far away from an NCICCC,

they simply do not have the option to seek treatment there. Additionally, many

health care plans do not cover services at university hospitals or NCICCCs at all

(Nardi et al., 2016). This translates to the majority of Americans not having

access to centers that provide better cancer outcomes, either because of

geographical limitations or because their insurance does not cover treatment at

institutions with proven better outcomes. This is one of the many challenges in

modern America, especially in rural areas. One of the ten essential services of

public health is to “evaluate effectiveness, accessibility, and quality of personal

and population-based health services” (Centers for Disease Control and

Prevention, 2017). The possible lack of diffusion of innovation in RO is a factor

that is often overlooked in improving population health. Providing RO care of the

highest quality to all patients by improving the utilization of innovative FDA-

approved techniques wound contribute to the improvement of accessibility and

quality of RO services across the United States and subsequently improve cancer

outcomes. This would be a contribution to positive social change.

Organizational Predictive Parameters

Academic affiliation status has been shown to have a positive main effect

on innovation and accessibility of services, possibly through slack staffing

resources and increased financial resources (Wisdom et al., 2014). Urbanicity has

been shown to have a positive effect on innovation and accessibility of services,

possibly through competition (Angst, Agarwal, Sambamurthy, & Kelley, 2010;

Charlton et al., 2015). Employment type details and type of practice have an

unclear relationship with innovation. Organizational structure and climate have

been shown to influence innovation adoption, possibly through fecund

management tiers (Angst et al., 2010; Wisdom et al., 2014); the number of

physicists and reporting structure will be used to assess the organizational

structure and climate for MPs.

Individual Predictive Parameters

Gender has a complicated relationship with innovation and there are

conflicting results in the literature (Lee, 2016). Recent research focuses on

entrepreneurship and gender ratios on the executive level and its relationship to

organizational innovation and success (Belghiti-Mahut, Lafont, & Yousfi, 2016).

The possible influence of gender in MPs practicing in RO is not certain, and it

will thus be interesting to include this parameter in the analysis. On the contrary,

the effects of education and age are well studied. Age typically has a negative

main effect on innovation adoption (younger people are more likely to adopt

innovations) and education typically has a positive main effect on innovation

adoption (people with higher education levels are more likely to adopt

innovations) (Arts, Frambach, & Bijmolt, 2011). Social status typically has a

positive main effect on innovation adoption (people with higher status are more

likely to adopt innovations) (Rogers, 2003); affiliation with academic institutions

will be used to assess the social status in MPs. Interpersonal channels typically

have a positive main effect on innovation implementation (people with more

intrapersonal channels are more likely to adopt innovations) (Rogers, 2003); the

number attendances in national meetings will be used to assess interpersonal

channels in MPs. Organizational structure and climate have been shown to

influence innovation adoption (people in larger, well-organized groups are more

likely to adopt innovation) (Wisdom et al., 2014); the number of physicists and

reporting structure will be used to assess the organizational structure and climate

for MPs. Motivation and appreciation are two individual characteristics that are

frequently used to assess the position of the individual in the stages previously

described by Rogers (2003) (Wisdom et al., 2014); the sense of appreciation and

motivation will be directly assessed in the study participants as context-specific

psychological antecedents. Appreciation and motivation typically have a positive

main effect on innovation adoption, as they may increase the individual’s

tolerance of risk (people who feel appreciated and motivated are more likely to

adopt innovations). Similarly, opinion leadership typically has a positive main

effect on innovation adoption (people who consider themselves opinion leaders

are more likely to adopt innovations) (Arts et al., 2011); volunteering in AAPM

committees will be used to assess the opinion leadership of MPs.

Summary and Conclusions

Innovation implementation is an area that needs improvement in science,

health care in general and in RO specifically. The diffusion of innovation

theoretical framework is one of the oldest social science theories, and it is well

studied in health care, thus grounding the literature review. In this study, the

results of the literature review were thematically synthesized to diffusion of

innovations in health care in general, diffusion of specific innovations in RO,

effects on value, effects on quality and outcomes, social impact, organizational

and personal predictive parameters.

The state of diffusion of innovation in RO in the United States is currently

not measured, but it is suspected to follow the low rates experienced in other

health care fields. There are many underutilized innovations in RO that can

enhance patient and staff safety and improve patient outcomes simultaneously.

Developing a tool to measure innovation will be the first step in developing a

predictive model to improving its diffusion. There is currently no systematic

quantification of diffusion of innovation in RO in the United States, and no

existing secondary dataset that can be used to extract this information.

Understanding the individual and organizational real and perceived barriers to

implementing innovations would create the framework to overcome them. If

something is not measured, there is no opportunity to study it and further improve

it. There needs to be enhanced understanding of how innovative techniques are

adopted in RO specifically and assess any disparities in technology

implementation in RO in the United States.

To this end, the following chapter describes the methods used to

systematically collect this data. A custom-designed online survey was created to

conduct a quantitative cross-sectional analysis of how MPs practice and utilize

FDA-approved innovative techniques in the academic vs. community setting and

rural vs. urban geographic locations. This provided the framework to develop a

predictive model assessing possible barriers at the individual (MPs) and

organizational (RO clinic) level, based on the previously discussed constructs.

Details of the survey design, sampling procedures, power analysis, data analysis,

threats to validity, and ethical procedures will be presented in the next chapter.

Chapter 3: Research Method

Introduction

As stated in Chapter 1, the purpose of this study was to investigate the

accessibility of innovative services in RO in the United States and assess possible

diffusion patterns using a quantitative cross-sectional survey. Surveying MPs to

measure the diffusion of innovation in RO is in alignment with the ninth essential

function of public health services to “evaluate effectiveness, accessibility, and

quality of personal and population-based health services” (Centers for Disease

Control and Prevention, 2017). As there are currently no indicators or instrument

for measuring the accessibility of innovative services in RO in the United States,

according to the literature review, this study quantified this issue and attempted to

explain the barriers to adoption of innovative technologies in RO. In this chapter,

the focus is on explaining the research design and rationale and the methodology

used. The sampling and recruitment procedures are also discussed, as are the

instrumentation and operationalization of constructs as they relate to the specific

topic. Last, the threats to validity and ethical procedures are addressed.

Research Design and Rationale

This study aimed to measure the innovation score in RO in the United

States. Innovation score is the dependent variable. For the purposes of this study,

innovation score is defined as the sum of innovation indicators on product,

technology, and organizational innovation. Product and technological innovation

were measured as the ratio of the number of innovative techniques available or

purchased by a department divided by the number of the innovative techniques

used clinically, weighed by partial implementation factors. Organizational

innovation was measured as the weighted average of innovations used to

reorganize procedures, work responsibilities, and decision-making.

Independent variables in this study included variables at both the

organizational and individual level. At the organizational level, independent

variables included university affiliation and urbanicity. At the personal level,

independent variables included gender, age, education, attendance of residency,

certification status, interpersonal channels, leadership structure, group size,

opinion leadership, appreciation, and motivation.

The research design selected to investigate the possible connections

between the dependent and independent variables was a quantitative cross-

sectional survey. Survey research is typically used for exploration, explanation, or

description and involves the collection of descriptive, behavioral, and attitudinal

data (Burkholder, 2015). In this study, survey research was used for explanatory

purposes by collecting descriptive and behavioral data. Innovation score is not

directly measured by any organization or vendor; thus, the only way to measure it

was to ask MP responders to describe how they use innovations in their work and

design a custom survey. Even if there were existing data on organizational

characteristics and innovation implementation, individual level characteristics

would not have been included. It was, therefore, necessary to reach out to

individual MPs and collect self-reported data.

The survey was conducted exclusively online. This type of design is

predominant in studies by health professionals, especially doctoral students

(Rudestam & Newton, 2014). Web survey research has been shown to have lower

costs, shorter length of time for field research, more flexible questionnaire design,

higher percentage of questions answered completely and accurately, higher ability

to administer complex instruments, shorter date processing time, lower error due

to manual data entry, faster survey deployment, and a larger amount of available

para-data (Dykema, Jones, Piché, & Stevenson, 2013). Potential disadvantages of

Web survey designs include required access to Internet and email, the requirement

of high computer literacy, increased likelihood of incomplete and erroneous

contact information, lower response rates, slightly lower demographic

representativeness, and decreased ability to administer incentives effectively

(Dykema et al., 2013). For the purposes of this study, a web survey was selected

because email communication is extensively used by and the preferred method of

communication of MPs (American Association of Physicists in Medicine, 2017).

Coincidentally, the use of a web survey aligned with the time and resource

limitations of the study.

Methodology

Population

In 2018, the AAPM directory listed approximately 7,500 members in total,

and approximately 5,500 members are listed as active in the United States (not

student, retired, associate, or corporate members) (American Association of

Physicists in Medicine, 2018). Seventy-seven percent of the membership reported

being active in RO (American Association of Physicists in Medicine, 2018); thus,

there were about 4,200 eligible MPs. Unofficial surveys typically garner an

approximate 5-10% response rate and official surveys an approximate 30-35%

response rate. (American Association of Physicists in Medicine, 2017)

Considering this response rate and the estimated number of AAPM members,

approximately 200 responses were expected for this survey.

Sampling and Sampling Procedures

Primary data was collected for this study using total population

convenience sampling because the population is finite and a list was readily

available. The MedPhysUSA listserv and AAPM Blackboard were used to recruit

participants. The MedPhysUSA listserv is used extensively among MPs for

survey dissemination, exchange of ideas, and general discussion. The listserv is

hosted by Wayne University and had over 4,000 members at the time of study.

Users must subscribe, and there are volunteer moderators. Forum policies were

followed; no official permission was needed to post. The AAPM Blackboard is

hosted by the AAPM and is open to all members for exchange of ideas and

general discussion. No special permission was needed to post the survey. The

survey was also posted on LinkedIn. LinkedIn retains data on user activity, as

described in its privacy policy (“Privacy Policy, LinkedIn,” n.d.). All users agree

to this policy as a condition of using the platform. The survey was designed using

an online survey research platform called SurveyMonkey (SurveyMonkey, 2019).

The invitation post included brief description of the study and a link to the survey.

Based on the expected participant number, a 5% chance of a Type I error

and a 20% chance of a Type II error was reasonable. Statistical power of 80% was

a reasonable initial goal as well. The typical four-to-one weighting was a good

starting point for this study, and it will be further assessed after data collection. If

the preliminary analysis reveals disproportional response rates between the

categories of interest, these numbers may need to be altered. Type II errors, which

encompass not detecting an effect that is actually there, or a false negative

(Salazar, Crosby, & DiClemente, 2015), posed more significant social

implications than Type I errors in this study. Not detecting an underperformance

in community and rural RO centers might have led to a lack of supplemental

efforts from AAPM to boost their utilization of new techniques and continuing

population inequalities. Effect sizes are unknown since innovation in the United

States has not been previously measured. Small to medium effect size of 0.2 were

assumed. Based on these assumptions, G-Power (Faul, Erdfelder, Lang, &

Buchner, 2007) calculated that approximately 200 responses were needed.

Procedures for Recruitment, Participation, and Data Collection

Participants were contacted passively by using the forums mentioned

above. The invitation contained a brief description, incentive information, and a

link to participate. Two reminders were sent approximately 10 days apart. The

survey remained open for a total of thirty days. When participants opened the

link, they were taken to the introductory study page, where they were asked to

consent to participate in the study. The survey contained a total of 70 questions

and took approximately 5-10 minutes to complete (SurveyMonkey, 2018).

Participants were asked multiple demographic and practice information as

described in other sections. Participants were asked to enter an email address if

they wish to receive a $10 Amazon gift card as an incentive to participate. This is

a reasonable amount, as MP’s salaries start at $140,000 (American Association of

Physicists in Medicine, 2018). The incentive was used to encourage participation.

Dykema et al. (2013) analyzed the use of web surveys in clinicians and other

healthcare providers and concluded that to achieve good response rate, incentives

are needed to improve participation rates. Participants will exit the study after

they complete the survey. No follow-up procedures are applicable.

Instrumentation and Operationalization of Constructs

Studies attempting to define and measure innovation in RO rely on custom

survey design, as there is no centralized reporting mechanism. This was described

in Chapter 2 as discussed by M. Jacobs et al. (2015) and Jacobs et al. (2016). This

survey is an extension of the work done by these groups, who validated their

innovation indicators using the Delphi method. Briefly, this Dutch group used

consensus guidelines among RO chairpersons to define innovation in RO and

used semi-structured interviews across 15 RO centers.

Reliability and validity. Reliability was measured using inter-item

reliability and split-half method. Chronbach’s alpha of 0.7 or higher was be used

to assess sufficient evidence of internal reliability (Salazar, Crosby, &

DiClemente, 2015). Face validity and content validity were demonstrated by an

expert panel. The expert panel was used to assess clarity of wording, applicability

of answers, etc. Five MPs who fit the eligibility criteria were contacted in

November 2018 and asked to review the survey and identify any ambiguity in the

wording of the questions. Each expert panel participant was contacted via email

and sent a preliminary version of the survey. Feedback was requested in writing

within a week. Comments were received during a two-week period. Comment

examples included the anonymity of the survey, stratification techniques, and

length of the consent form. Additionally, comments were requested from the

AAPM Technology Assessment office and received within 2 weeks in January

2019. Four blinded field experts reviewed the survey and provided feedback on

the structure and levels of measurement. All comments were used to improve the

survey questions before deploying the study to the target population and enhance

the study’s validity, as discussed in subsequent sections. Construct validity was

demonstrated using principal component analysis (PCA) after data collection.

Operationalization of Constructs. The dependent variable in this study is

innovation score, which is defined as the sum of innovation indicators on product,

technology, and organizational innovation. Product and technological innovation

were measured as the ratio of the number of innovative techniques available or

purchased by a department divided by the number of the innovative techniques

used clinically, weighed by partial implementation factors. Organizational

innovation was measured as the weighted average of innovations used to re-

organize procedures, work responsibilities, and decision making. This is a

continuous variable. For example, assume that there are ten indicators for product,

technology, and organizational innovation and one of the responders has five

available for use in their clinic. Out of those five, they use the first two all the

time (100%), the third one most of the time (75%) and the last two half of the

time (50%). The innovation score for this respondent will be

(5/10)*[(100+100+75+50+50)/5] = 0.5*0.75 = 0.375.

The independent variables are described in detail in Table 1. University

affiliation was measured as a binary yes or no. Zip code text entry was converted

to categorical using the RUCA continuum (USDA, 2013), explained in Appendix

C. Gender was binary male or female. Age was measured as a continuous variable

and was recoded into categorical. Education was measured as Master’s, Doctoral

or other. Residency status was measured as a categorical variable, (yes, no, no

didn’t need). ABR status was measured as yes, yes/other, or no. Interpersonal

channels were measured as a continuous variable, using the number of meetings

attended. Organizational structure was be measured as a categorical variable as

physicist, physician, administrator. Group characteristics were measured as a

categorical variable based on the size of the group. Opinion leadership was

measured as a binary variable as yes or no. Appreciation was measured as a

continuous variable. Motivation was measured as a continuous variable. Barriers

were measured as continuous variables.

Table 1

Operationalization of Constructs

Question Variable

Level of

measurement RQ

Q3-57 Innovation score

(dependent variable) Continuous 1, 2 3

Q1: university affiliation University affiliation Binary 1a

Q2: location Zip code Categorical 1b

Q3-57 Barriers Continuous 3

Q58: gender Gender Categorical 2

Q59: Age Age Continuous 2

Q60: degree Degree Binary 2

Q61: residency Residency Categorical 2

Q62: ABR status ABR status Categorical 2

Q63: meeting attendance Interpersonal channels Continuous 2

Q64: reporting organizational structure Categorical 2

Q65: # of physicists Group characteristics Categorical 2

Q66: volunteering Opinion leadership Categorical 2

Q67: appreciation Appreciation Continuous 2

Q68: motivation Motivation Continuous 2

Q69: final thoughts Free text - -

Q70: gift card Free text - -

Data Analysis Plan

After completing data collection, SPSS version 25 was used for statistical

analysis. Data were screened for duplicates. Missing entries were reviewed for

randomness, and data were removed pairwise when applicable. The RQs and

hypotheses were as follows:

RQ 1: What are the differences in accessibility to innovation in RO based on

location and type of practice in the United States?

RQ 2: What are the statistically significant factors (demographics, practice details,

context-specific psychological antecedents) that predict the accessibility to

innovation in the RO clinic?

RQ 3: What are the statistically significant barriers that MPs practicing in RO in

the United States face in implementing innovations?

Hypothesis 1a

H1a0: There is no statistically significant difference in accessibility to

innovation in RO between university and non-university hospitals.

H1a1: There is a statistically significant difference in accessibility to

innovation in RO between university and non-university hospitals.

Dependent variable: innovation score (continuous).

Independent variable: university affiliation (categorical, two levels: has

university affiliation, does not have university affiliation).

Hypothesis 1b

H1b0: There is no statistically significant difference in the accessibility to

innovation in RO between metropolitan and non-metropolitan hospitals.

H1b1: There is a statistically significant difference in accessibility to

innovation in RO between metropolitan and non-metropolitan hospitals.

Dependent variable: innovation score (continuous).

Independent variable: urbanicity (categorical, metropolitan and non-

metropolitan).

Hypothesis 2

H20: There are no statistically significant factors predicting innovation

score.

H21: There are statistically significant factors predicting innovation score.

Dependent variable: innovation score (continuous).

Independent variables: reporting structure, size of physics group, age,

gender, DABR status, residency status, meeting attendance, education

level, leadership,(categorical) appreciation, and motivation (continuous).

Hypothesis 3

H30: There are no statistically significant barriers affecting innovation

score.

H31: There are statistically significant barriers affecting innovation score.

Dependent variable: innovation score (continuous).

Independent variables: barriers (continuous).

Analysis plan. For RQ 1 the purpose is to compare groups, thus t-test and

ANOVA were used (Frankfort-Nachmias & Leon-Guerrero, 2015). T-test was

used to compare innovation score for type of hospital variable and ANOVA was

used to compare innovation between the RUCA urbanicity continuum. The t and

F values and p-value were reported, along with means and standard deviations for

all groups. Alternative non-parametric tests, such as the Mann-Whitney U and

Kruskal-Wallis H tests, would have been selected if the dataset was found to have

a significant deviation from normality. If homogeneity of variance was met,

Tukey's honestly significant difference (HSD) post hoc test was used to assess

which specific groups differed. If the data did not meet the homogeneity of

variances assumption, the Games Howell post hoc test would have been

considered.

For research question 2, multiple linear regression was initially planned to

be used if the assumptions of linearity were met. The dependent variable was

innovation score. This is a continuous variable. Independent variables in the

model were respondents’ age, gender, education level, ABR status, residency

status, interpersonal channels, appreciation level, motivation level, opinion

leadership, practice organizational structure, and group characteristics. Alternative

non-parametric tests were selected because the data did not meet normality

criteria. To do so, the dependent variable (innovation score) was recoded to be

binary categorical (below median, above median) and binomial logistic regression

was used instead.

For research question 3, correlation testing was used. The dependent

variable was innovation utilization score. The independent variables are the

individual barriers (lack of evidence, complexity, lack of time or staffing, lack of

training or support, lack of interest, lack of interoperability, and lack of

reimbursement. Pearson’s correlation was selected if the relationship between the

dependent and independent variable is linear. Spearman correlation was selected

if the data did not meet normality criteria.

Threats to Validity

Threats to External Validity

This study will reach the entire MPs population in the United States, thus

threats to external validity due to sampling are limited (Cresswell, 2018). There

is, however, the problem of volunteer bias, as participants who volunteer to be in

a study about innovation may inherently be more innovative. The respondents

may not be representative of the general population in regards to innovation

practice or other characteristics. Additionally, there is inherent generalization

across constructs on how the constructs apply to MPs specifically. For example,

opinion leadership as a construct will be measured using volunteering in the

AAPM. While participation in professional association leadership is a measure of

opinion leadership in other disciplines (B. L. Jacobs et al., 2017), there is an

inherent assumption that this connection transfers to MPs as well. This

generalization across constructs may lead to extraneous or confounding variables

in MPs.

Threats to Internal Validity

Extraneous effects may pose a threat to the internal validity of this study,

as there may be excluded extraneous variables that were elusive during the

literature review. This threat was minimized by the use of an expert panel who

reviewed the survey before implementation. Additionally, there may be personal

biases, which should be minimal due to the quantitative nature of the study.

Participant reactivity may also be present in the study, as participants may

respond as being more innovative than they really are, suspecting that this is the

desired behavior. This may affect the measurement of the dependent variable This

is unlikely to have significant effects on an online professional survey, as this is

most often observed in laboratory settings. These possible effects will be

minimized as much as possible with the use of the appropriate multivariable

analysis (Frankfort-Nachmias & Leon-Guerrero, 2015).

Ethical Procedures

Ethical considerations for the participants. Research participants have a

moral obligation to answer truthfully. There are many surveys performed through

the AAPM, including salary surveys for workforce forecasting, and

confidentiality is not really a concern among members. The participant obligation

will be emphasized at the beginning of the study, where the study's scope and

rationale will be explained, in addition to the time commitment expected by the

participants. Participants will be asked to acknowledge the expectations of their

participation. This acknowledgment will serve as consent and as a “contract”

between the participant and the researcher (Rudestam & Newton, 2014).

Participants who are not MPs practicing in RO in the United States, yet

participated for various reasons, would have violated the ethics of the consent.

Various methods were used to positively reinforce truthful answers. Using

a continuous, dependent-response scale (0-100 values) instead of a discrete,

Likert-type scale for partial implementation is expected to have participants

answer more truthfully, versus a yes or no answer that may not be indicative of

how exactly they are using these techniques.

Ethical considerations for the researcher. The survey poses minimal

risk to the participants, as the questions do not fall into the “traditional” sensitive

information areas. Information can be perceived as sensitive when it involves

cultural taboos, the threat of disclosure, or social desirability. What information is

too private, or taboo, is extremely subjective and highly dependent on the culture

(Epstein, Santo, & Guillemin, 2015). Participants were asked questions about how

they perform their daily work duties, and none of the questions is of sensitive

nature. The survey questions can, however, be seen as a surrogate to professional

performance, and thus affect the respondent’s view of performance and self-

worth. Additionally, some of the questions will be about feelings of appreciation

in the workplace. This may affect participants, as they typically may not have to

think about something like that (Rudestam & Newton, 2014). This effect may be

positive or negative, depending on the person and the situation. Possibility feeling

uncomfortable was thus included in the study consent.

Concerns over privacy will be handled using the anonymity features of the

survey provider. The custom link will be provided along with the introductory

text for the email. Email addresses will not be tracked. Additionally, concerns

over Internet Protocol address (IP) tracking will be addressed by turning the

feature off in the survey provider.

Institutional approval to conduct research. The Walden University

Institutional Review Board (IRB) approved this study on April 24, 2019 (approval

number 04-24-19-0603259, expiring April 23rd, 2020).

Summary

In this chapter, a detailed discussion of the research design and rationale

as a total population convenience sampling cross-sectional survey was presented.

The methodology, population, sampling procedures, instrumentation, and

operationalization of constructs, data analysis plan were discussed. Possible

threats to validity were also explored. The results of the study will be presented in

Chapter 4.

Chapter 4: Results

Introduction

The purpose of this research study was to examine the diffusion of

innovation in RO in the United States. To assess diffusion, a variety of statistical

tests was used to compare innovation score with organizational and personal

characteristics. A summary of the results is presented in this chapter. The

following RQs were answered:

RQ 1: What are the differences in accessibility to innovation in RO based on

location and type of practice in the United States?

RQ 2: What are the statistically significant factors (demographics, practice details,

context-specific psychological antecedents) that predict the accessibility to

innovation in the RO clinic?

RQ 3: What are the statistically significant barriers that MPs practicing in RO in

the United States face in implementing innovations?

Data Collection

The survey was posted on Saturday, April 27, 2019. The invitation is

listed in Appendix B. During the first two weeks of the survey, 136 responses

were collected. The first reminder was sent on Saturday, May 11, 2019, and the

last reminder was sent on Wednesday, May 22, 2019. The survey closed on

Saturday, June 1, 2019, at 11:50 pm. At the survey closure, 265 responses were

collected. Twenty-five responses contained no answers at all. Thus, the final

sample size was 240. Incomplete data entries were excluded pairwise when

applicable. The analysis was conducted using SPSS 25.

Descriptive Statistics

Independent variables were recoded as needed, taking into account the

number of responses received. ZIP codes entered in the survey were converted to

county, and the Rural-Urban Continuum Codes (Appendix C) were used to

categorize locations as metropolitan and nonmetropolitan (USDA, 2013), as

shown in Table 2. Most responses were received from metropolitan areas.

University affiliation is also listed in Table 2; the sample contained approximately

equal responses between university and non university facilities. Sample

demographics for gender, education and age are shown in Table 3. Descriptive

statistics for respondent appreciation and motivation are shown in Table 4.

Sample professional characteristics for residency, DABR status, number of

meetings in the past 10 years, reporting, number of physicists and volunteering

are shown in Table 5. Principal component analysis was found to not a suitable

test, as the overall Kaiser-Meyer-Olkin measure ranged from 0.588 to 0.634

depending on the factors entered in the model.

Table 2

MPs Characteristics for RO center University Affiliation and Urbanicity

Characteristic N f %

University Affiliation 240

Yes 111 46.3

No 129 53.8

Urbanicity 220

RUCA Cat 1 138 62.7

RUCA Cat 2 50 22.7

RUCA Cat 3 17 7.7

RUCA Cat 4 4 1.8

RUCA Cat 5 4 1.8

RUCA Cat 6 4 1.8

RUCA Cat 7 3 1.4

Urbanicity Binary 1 220

RUCA Cat 1 138 62.7

RUCA Cat 2-9 82 37.3

Urbanicity Binary 2 220

RUCA Categories 1-3 205 93.2

RUCA Categories 4-9 15 6.8

Table 3

MPs Demographics for RO

Characteristic N f %

Gender 188

Male 155 82.4

Female 33 17.6

Education 190

Master’s 118 62.1

Doctoral 72 37.9

Age distribution 183

< = 30 23 12.6

31-40 70 38.3

41-50 36 19.7

51-60 30 16.4

> 60 24 13.1

Table 4

Descriptive Statistics for RO Appreciation and Motivation in MPs

N M SD

Appreciation 190 69.16 26.728

Motivation 190 76.35 23.463

Table 5

MPs Professional Characteristics

Characteristic N f %

Residency 185

Yes

77 41.6

108 58.4

DABR status 188

DABR 150 79.8

not DABR/other 38 20.2

Number of meetings in the past 10 years 179

0-5 104 58.1

6-9 42 23.5

> = 10 33 18.4

Reporting 189

Another physicist 114 60.3

Physician 22 11.6

Administrator 37 19.6

VP/CEO 16 8.5

Number of physicists 191

1-3 98 51.3

≥ 4 93 48.7

Volunteer 185

Yes 72 38.9

No 113 61.1

Dependent variable coding. The dependent variable, innovation score,

was measured using 20 unique indicators in five categories: patient positioning,

patient treatment, treatment planning, quality assurance, and workflow, as shown

in Appendix D. Participants answers were entered using a slider with scale 0-100.

For values entered as 0 or 1 on the slider, it was assumed that the respondents

meant to not move the slider at all and that the type of innovation was not

available. For values entered as 2 to 100 on the sliding scale, it was assumed that

the respondents had the technology available to them. This categorized each

respondent as having or not having the innovation. The innovation indicators were

summed to calculate the innovation score for each category and the total

innovation score for each respondent. This provided a measure for the diffusion of

available innovative techniques and will be subsequently referred to as innovation

score. The distribution is shown in Figure 1, and it appears to be normally

distributed.

Further, the exact number on the scale of individual responses was used to

assess how respondents actually used the innovations available to them. This will

be subsequently referred to as innovation utilization score, and it is distinctly

different from the previously defined innovation score. A similar method

described for innovation score was used for the innovation utilization score. The

mean of the responses in each of the indicators in each category was used to

calculate the innovation utilization score in each category separately. The

innovation utilization score was calculated by adding the innovation utilization

score in the five categories. The distribution is shown in Figure 2, and it appears

to be normally distributed.

The Shapiro-Wilk test for normality was performed for innovation score

and innovation utilization score, using university affiliation and urbanicity as

factors. The results of the Shapiro-Wilk test were not statistically significant for

the innovation score (p = 0.611) and for the innovation utilization score (p =

0.699). Thus, the data were normally distributed. The Q-Q plots were also normal.

The 20 parameters used to measure innovation were assessed for the

reliability of the construct. The scale had a high level of internal consistency, as

determined by a Cronbach's alpha of 0.963.

Figure 1. Distribution of RO innovation score.

Figure 2. Distribution of RO innovation utilization score.

Study Results

Research Question 1 Results

Research Question (RQ) 1: What are the differences in accessibility to innovation

in RO based on location and type of practice in the United States?

H1a0: There is no statistically significant difference in accessibility to

innovation in RO between university and non-university hospitals.

H1a1: There is a statistically significant difference in accessibility to

innovation in RO between university and non-university hospitals.

Dependent variable: innovation score (continuous).

Independent variable: university affiliation (categorical, two levels: has

university affiliation, does not have university affiliation).

Results for Hypothesis 1a. An independent t-test was performed to

determine if there were differences in innovation score between university and

non-university centers for the total innovation score and for the innovation

utilization score. There were no outliers in the data, as assessed by inspection of a

boxplot. There was homogeneity of variances, as assessed by Levene’s test for

equality of variances for both the total innovation score (p = 0.689) and the

innovation utilization score (p = 0.129).

The mean innovation score difference for centers with university

affiliation (M = 13.19, SD = 4.76) is higher than the mean innovation score for

centers without a university affiliation (M = 11.55, SD = 4.63), a statistically

significant difference MD = 1.65, 95% CI[0.38,2.917], t(211) = 2.56, p = 0.011, d

= 0.351. Additionally, the patient treatment innovation score for university centers

(M = 3.04, SD = 1.43) is higher than the patient treatment innovation score for

non-university centers (M = 2.64, SD = 1.34), a statistically significant difference

MD = 0.39, 95% CI[0.021,0.76], t(217) = 2.083, p = 0.038, d = 0.282; the

workflow innovation score for university centers (M = 2.96, SD = 1.51) is higher

than the workflow innovation score for non-university centers (M = 2.50, SD =

1.55), a statistically significant difference MD = 0.46, 95% CI[0.05,0.86], t(217) =

2.217, p = 0.028, d = 0.188. Results are presented in Tables 6 and 7. Cohen’s d

was calculated manually.

Table 6

Group Statistics for RO center Innovation Score and University Affiliation

Category University

Affiliation N M SD SEM

Patient Positioning Yes 106 2.217 1.087 .106

No 113 2.018 1.035 .097

Patient Treatment Yes 106 3.038 1.434 .139

No 113 2.646 1.349 .127

Treatment

Planning

Yes 106 2.292 1.441 .139

No 113 2.115 1.361 .128

Quality Assurance Yes 106 2.311 1.539 .149

No 113 1.956 1.454 .137

Workflow Yes 106 2.962 1.505 .146

No 113 2.504 1.548 .146

Total Yes 103 13.194 4.757 .469

No 110 11.546 4.635 .442

Table 7

Independent Samples T- Test for RO center Innovation Score and University

Affiliation

Category

Levene's Test

for Equality of

Variances

t-test for Equality of Means

F p t df p MD SED

95% CI

LL UL

Patient

Positioning

Equal variances assumed 1.009 .316 1.390 217 .166 .199 .144 -.083 .485

Equal variances not

assumed

1.388 214 .167 .199 .144 -.084 .482

Patient

Treatment

Equal variances assumed .003 .953 2.083 217 .038 .392 .188 .021 .762

Equal variances not

assumed

2.079 211 .039 .392 .188 .0203 .763

Treatment

Planning

Equal variances assumed 1.510 .221 .937 217 .350 .177 .189 -.196 .550

Equal variances not

assumed

.935 214 .351 .177 .189 -.196 .551

Quality

Assurance

Equal variances assumed 1.966 .162 1.758 217 .080 .356 .202 -.043 .754

Equal variances not

assumed

1.755 214 .081 .356 .203 -.044 .755

Workflow

Equal variances assumed 1.152 .284 2.217 217 .028 .458 .207 .051 .865

Equal variances not

assumed

2.219 217 .028 .458 .206 .051 .864

Total

Equal variances assumed .160 .689 2.562 211 .011 1.649 .644 .379 2.917

Equal variances not

assumed

2.559 209 .011 1.649 .644 .379 2.919

The mean innovation utilization score for centers with university

affiliation (M = 59.39, SD = 17.74) is similar to the mean innovation utilization

score for centers without a university affiliation (M = 60.62, SD = 15.97). The

innovation utilization score difference is not statistically significant based on

university affiliation. However, for the five categories measured, the mean

workflow innovation utilization score for centers with university affiliation (M =

54.05, SD = 22.85) is higher than the mean workflow innovation utilization score

for centers without a university affiliation (M = 46.95, SD = 19.92), a statistically

significant difference MD = 7.09, 95% CI[0.78,13.39], t(178) = 2.217, p = 0.028,

d = 0.330. Results are presented in Tables 8 and 9. Cohen’s d was calculated

manually. Thus, the null hypothesis H1a,0 is rejected, and the alternate H1a,1 is

accepted.

Table 8

Group Statistics for RO center Innovation Utilization Score and University

Affiliation

Category University

Affiliation N M SD SEM

Patient

Positioning

Yes 90 60.985 25.089 2.645

No 98 63.908 23.681 2.392

Patient Treatment Yes 96 76.195 24.681 2.519

No 102 76.769 25.208 2.496

Treatment

Planning

Yes 91 48.092 26.226 2.749

No 95 51.637 24.973 2.562

Quality

Assurance

Yes 85 56.022 28.503 3.092

No 86 61.565 23.893 2.576

Workflow Yes 90 54.045 22.859 2.410

No 90 46.959 19.921 2.099

Total Yes 103 59.393 17.749 1.749

No 110 60.619 15.970 1.523

Table 9

Independent Samples T-Test for RO center Innovation Utilization Score and

University Affiliation

Category

Levene's Test for

Equality of

Variances

t-test for Equality of Means

F p t df p MD SED 95% CI

LL UL

Patient

Positioning

Equal variances

assumed .172 .679 -.822 186 .412 -2.923 3.557 -9.941 4.095

Equal variances

not assumed

-.820 182 .413 -2.923 3.566 -9.956 4.113

Patient

Treatment

Equal variances

assumed .588 .444 -.162 196 .872 -.575 3.546 -7.573 6.423

Equal variances

not assumed

-.162 196 .871 -.575 3.546 -7.568 6.419

Treatment

Planning

Equal variances

assumed .269 .605 -.944 184 .346 -3.545 3.754 -10.952 3.861

Equal variances

not assumed

-.943 182 .347 -3.545 3.758 -10.960 3.869

Quality

Assurance

Equal variances

assumed 3.361 .069 -1.38 169 .170 -5.544 4.020 -13.481 2.392

Equal variances

not assumed

-1.38 163 .170 -5.544 4.024 -13.491 2.402

Workflow

Equal variances

assumed 2.542 .113 2.217 178

.028 7.086 3.196 .771 13.39

Equal variances

not assumed

2.217 175 .028 7.086 3.196 .778 13.40

Total

Equal variances

assumed 2.317 .129 -.531 211 .596 -1.227 2.311 -5.782 3.33

Equal variances

not assumed

-.529 205 .597 -1.227 2.319 -5.798 3.345

Hypothesis 1b

H1b0: There is no statistically significant difference in the accessibility to

innovation in RO between metropolitan and non-metropolitan hospitals.

H1b1: There is a statistically significant difference in accessibility to

innovation in RO between metropolitan and non-metropolitan hospitals.

Dependent variable: innovation score (continuous)

Independent variable: urbanicity (categorical, metropolitan, and non-

metropolitan)

Results for Hypothesis 1b. An independent t-test was performed to

determine if there were differences in innovation score between metropolitan and

non-metropolitan centers for the total innovation score and for the innovation

utilization score. There were no outliers in the data, as assessed by inspection of a

boxplot. There was homogeneity of variances, as assessed by Levene’s test for

equality of variances for both the total innovation score (p = 0.478) and the

innovation utilization score (p = 0.855). It is noted that total innovation score was

assessed using binary RUCA categorization 1 and 2-9, while innovation

utilization score was assessed using binary RUCA categorization 1-3 and 4-9.

Results are presented in Tables 10 to 13.

The mean innovation score for metropolitan centers (M = 12.94, SD =

4.65) is similar to the mean innovation score for non-metropolitan centers (M =

11.69, SD = 4.37). The innovation score difference is not statistically significant

based on metropolitan or non-metropolitan status, even though d = 0.275.

However, for the five categories measured, the mean patient positioning

innovation score for metropolitan centers (M = 2.21, SD = 1.02) is higher than the

mean patient positioning innovation score for non-metropolitan centers (M =

1.89, SD = 1.12), a statistically significant difference MD = 0.31, 95%

CI[0.011,0.612], t(203) = 2.043, p = 0.042, d = 0.293. Additionally, the mean

patient treatment innovation score for metropolitan centers (M = 3.08, SD = 1.36)

is higher than the mean patient treatment innovation score for non-metropolitan

centers (M = 2.47, SD = 1.33), a statistically significant difference MD = 0.62,

95% CI[0.23,1.00], t(203) = 3.145, p = 0.002, d = 0.457. Results are presented in

Tables 10 and 11. Cohen’s d was calculated manually.

Table 10

Group Statistics for RO center Innovation Score and Urbanicity

Category UrbanicityBinary1 N M SD SEM

Patient Positioning RUCA 1 130 2.208 1.025 .090

RUCA 2-9 75 1.893 1.122 .130

Patient Treatment RUCA 1 130 3.085 1.364 .120

RUCA 2-9 75 2.467 1.340 .155

Treatment Planning RUCA 1 130 2.354 1.375 .121

RUCA 2-9 75 2.000 1.356 .157

Quality Assurance RUCA 1 130 2.169 1.536 .135

RUCA 2-9 75 2.160 1.395 .161

Workflow RUCA 1 130 2.823 1.553 .136

RUCA 2-9 75 2.707 1.431 .165

Total RUCA 1 127 12.937 4.653 .413

RUCA 2-9 72 11.694 4.375 .516

Table 11

Independent Samples T-Test for RO center Innovation Score and Urbanicity

Category

Levene's Test for Equality of

Variances t-test for Equality of Means

F p t df p MD SED

95% CI

LL UL

Patient

Positioning

Equal variances

assumed

1.090 .298 2.043 203 .042 .314 .154 .011 .618

Equal variances not

assumed

1.994 143 .048 .314 .158 .003 .626

Patient

Treatment

Equal variances

assumed

.176 .675 3.145 203 .002 .618 .196 .230 1.005

Equal variances not

assumed

3.161 157 .002 .618 .196 .232 1.004

Treatment

Planning

Equal variances

assumed

.536 .465 1.784 203 .076 .354 .198 -.037 .745

Equal variances not

assumed

1.791 156.273 .075 .354 .198 -.036 .744

Quality

Assurance

Equal variances

assumed

1.997 .159 .043 203 .966 .009 .216 -.416 .434

Equal variances not

assumed

.044 167 .965 .009 .210 -.405 .424

Workflow

Equal variances

assumed

3.040 .083 .532 203 .595 .116 .219 -.315 .548

Equal variances not

assumed

.544 165 .587 .116 .214 -.306 .539

Total

Equal variances

assumed

.505 .478 1.849 197 .066 1.243 .672 -.083 2.568

Equal variances not

assumed

1.881 155 .062 1.243 .660 -.062 2.547

The mean innovation utilization score for metropolitan centers (M =

60.73, SD = 16.67) is similar to the mean innovation utilization score for non-

metropolitan centers (M = 55.41, SD = 15.38). The total innovation utilization

score difference is not statistically significant based on metropolitan or non-

metropolitan status, even though d = 0.331. However, for the five categories

measured, the mean patient positioning innovation utilization score for

metropolitan centers (M = 63.96, SD = 23.78) is higher than the mean patient

positioning innovation utilization score for non-metropolitan centers (M = 47.73,

SD = 29.28), a statistically significant difference MD = 16.22, 95%

CI[0.73,31.72], t(173) = 2.067, p = 0.04, d = 0.608. Results are presented in

Tables 12 and 13. Cohen’s d was calculated manually. Thus, the null hypothesis

H1b,0 is rejected, and the alternate H1b,1 is accepted.

Table 12

Group Statistics for RO center Innovation Utilization Score and Urbanicity

Category UrbanicityBinary2

N M SD SEM

Patient Positioning RUCA 1-3 165 63.958 23.784 1.852

RUCA 4-9 10 47.733 29.287 9.262

Patient Treatment RUCA 1-3 175 78.023 23.493 1.776

RUCA 4-9 11 72.897 25.561 7.707

Treatment

Planning

RUCA 1-3 166 50.171 25.228 1.958

RUCA 4-9 11 43.606 28.433 8.573

Quality Assurance RUCA 1-3 152 57.803 25.949 2.105

RUCA 4-9 11 69.955 25.739 7.761

Workflow RUCA 1-3 164 50.896 21.894 1.710

RUCA 4-9 8 40.042 18.073 6.390

Total RUCA 1-3 186 60.726 16.673 1.223

RUCA 4-9 13 55.416 15.385 4.267

Table 13

Independent Samples T- Test for RO center Innovation Utilization Score and

Urbanicity

Category

Levene's

Test for

Equality of

Variances

t-test for Equality of Means

F p t df p MD SED

95% CI

LL UL

Patient

Positioning

Equal variances assumed 1.463 .228 2.067 173 .040 16.22 7.849 .732 31.72

Equal variances not

assumed

1.718 10 .117 16.22 9.445 -4.90 37.35

Patient

Treatment

Equal variances assumed .067 .795 .698 184 .486 5.13 7.339 -9.35 19.61

Equal variances not

assumed

.648 11 .530 5.13 7.909 -12.3 22.52

Treatment

Planning

Equal variances assumed .221 .639 .829 175 .408 6.56 7.915 -9.06 22.19

Equal variances not

assumed

.747 11 .471 6.56 8.794 -12.8 25.91

Quality

Assurance

Equal variances assumed .043 .836 -1.50 161 .135 -12.15 8.098 -28.1 3.84

Equal variances not

assumed

-1.51 12 .158 -12.15 8.041 -29.8 5.45

Workflow

Equal variances assumed .214 .644 1.379 170 .170 10.86 7.875 -4.69 26.40

Equal variances not

assumed

1.641 8 .139 10.86 6.614 -4.38 26.09

Total

Equal variances assumed .034 .855 1.115 197 .266 5.310 4.762 -4.1 14.70

Equal variances not

assumed

1.196 14 .251 5.310 4.439 -4.2 14.83

Research Question 2 Results

Research Question (RQ) 2: What are the statistically significant factors

(demographics, practice details, context-specific psychological

antecedents) that predict the accessibility to innovation in the RO clinic?

H20: There are no statistically significant factors predicting innovation

score.

H21: There are statistically significant factors predicting innovation score.

Dependent variable: innovation score (continuous).

Independent variables: reporting structure, size of physics group, age,

gender, DABR status, residency status, meeting attendance, education

level, leadership, (categorical) appreciation, and motivation (continuous).

Bivariate analysis for appreciation, motivation, and number of

meetings. The Shapiro-Wilk test for normality was performed for innovation

score and innovation utilization score using appreciation, motivation, and number

of meetings as factors (continuous variables). The results of the Shapiro-Wilk test

were statistically significant for all three parameters (p < 0.001). Thus, the data

are not normally distributed. The Q-Q plots were also not normal. Spearman

correlation was used to perform bivariate analysis for appreciation and

motivation. There was a positive correlation between innovation utilization and

participant appreciation (rs = 0.224, p = 0.002) and motivation (rs = 0.215, p =

0.003). Both correlations are of small to medium effect size class, according to

Cohen’s criteria (Ellis, 2010). There was a small negative correlation between

innovation utilization and the number of meetings attended (rs = -0.186, p =

0.013). Since this result is unexpected, the relationship with total innovation score

was also investigated. Results were not statistically significant (rs = 0.067, p =

0.371). Thus, the number of meetings will not be included in the final regression

model.

Bivariate analysis for gender, opinion leadership, education, and

residency status. The Shapiro-Wilk test for normality was performed for

innovation score and innovation utilization score using gender, opinion

leadership, education, and residency as factors (binary variables). The results of

the Shapiro-Wilk test were not statistically significant for gender (p = 0.433),

opinion leadership (p = 0.172), education (p = 0.922), and residency (p = 0.981).

Thus, the data are normally distributed. The Q-Q plots were also normal. An

independent sample t-test was performed for gender, opinion leadership,

education, and residency. Results were not statistically significant for gender (p =

0.828), opinion leadership (p = 0.921), and residency (p = 0.402). Results were

statistically significant for education. The mean innovation utilization score for

Master’s degree (M = 61.42, SD = 16.11) is higher than the mean innovation

utilization score for Doctoral degree (M = 56.37, SD = 17.22), a statistically

significant difference MD = 5.06, 95% CI[0.18,9.94], t(188) = 2.046, p = 0.042, d

= 0.278.

Bivariate analysis for age, DABR status, number of meetings,

reporting structure, and number of physicists. The Shapiro-Wilk test for

normality was performed for innovation score and innovation utilization score

using age, DABR status, number of meetings, reporting structure, and number of

physicists as factors (categorical variables). The results of the Shapiro-Wilk test

were not statistically significant for meetings (p = 0.425), age (p = 0.768), DABR

status (p = 0.573), reporting structure (p = 0.739), and the number of physicists (p

= 0.431). There was no statistically significant effect of age (p = 0.38), DABR

status (p = 0.10), reporting status (p = 0.06), and the number of physicists in

practice (p = 0.57).

Binary logistic regression. Three parameters were found to be

statistically significant in affecting innovation implementation: degree,

motivation, and appreciation. Since the two latter are non-linear, a binomial

logistic regression was performed. The dependent variable, innovation utilization,

was recoded to binary (1 = below median, 2 = above median, Mdn = 60.03).

Linearity of the continuous variables with respect to the logit of the dependent

variable was assessed using the Box-Tidwell procedure. The interaction term was

not statistically significant for motivation (p = 0.587), and for appreciation (p =

0.338), thus these two independent variables are linearly related to the logit of the

dependent variable. No corrections were applied. Results of the binary logistic

regression are presented in Tables 14 to 18.

Table 14

Binary Logistic Regression Omnibus Tests of Model Coefficients

χ2df p

Step 1 Step 10.861 3 .013

Block 10.861 3 .013

Model 10.861 3 .013

Table 15

Binary Logistic Regression Model Summary

Step -2 Log likelihood Cox & Snell R Square Nagelkerke R Square

1 250.720a .056 .075

a. Estimation terminated at iteration number 3 because parameter estimates changed by less than

.001.

Table 16

Binary Logistic Regression Hosmer and Lemeshow Test

Step χ2 df p

1 6.187 8 .626

Table 17

Binary Logistic Regression Classification Table

Observed Predicted

Innovation Utilization Score

Below

Median

Above

Median

Percent

Correct

Step 1 Innovation

Utilization Score

Below

Median 60 34 65.7

Above

Median 40 50 55.6

Overall

Percentage 60.8

a. The cut value is .500

Table 18

Binary Logistic Regression Results and Variables in the Equation for Innovation

Utilization (outcome variable)

B S.E. Wald df p OR

95% CI.for

LL UL

Step

Education -.600 .315 3.620 1 .057 .549 .296 1.018

Appreciation .013 .008 2.385 1 .122 1.013 .997 1.030

Motivation .001 .009 .005 1 .944 1.001 .983 1.019

Constant -.821 .560 2.151 1 .142 .440

a. Variable(s) entered on step 1: Education, Appreciation, Motivation.

There were no cases with studentized residuals greater than 2.0. The

Hosmer and Lemeshow Test was not statistically significant (p = 0.626), thereby

suggesting a model fit. The logistic regression model was statistically significant,

χ2(3) = 10.861, p = 0.013. The model explained 7.5% (Nagelkerke R2) of the

variance in innovation utilization and correctly classified 60.8% of cases.

Sensitivity was 55.6%, specificity was 65.7%, positive predictive value was

59.5%, and the negative predictive value was 38.1%. None of the predictor

variables was statistically significant. Thus, the null hypothesis H2,0 is accepted,

and the alternate H2,1 is rejected.

Research Question 3 Results

RQ 3: What are the statistically significant barriers that MPs practicing in RO in

the United States face in implementing innovations?

H30: There are no statistically significant barriers affecting innovation

score.

H31: There are statistically significant barriers affecting innovation score.

Dependent variable: innovation score (continuous).

Independent variables: barriers (continuous).

Seven measured barriers across five categories and in total were assessed

for normality. The results of the Shapiro-Wilk test were mixed, with some values

being statistically significant, and some not being statistically significant (data not

shown). Thus, the data are assumed to not be normally distributed as a group. The

Q-Q plots were also not normal. The seven parameters used to assess barriers

were assessed for the reliability of the construct. The scale had a high level of

internal consistency, as determined by a Cronbach's alpha of 0.882. Spearman

correlation was used to perform bivariate analysis across each category and in

total. Spearman’s rho results are presented in Tables 19 to 25.

There are statistically significant barriers in patient treatment, treatment

planning, quality assurance, workflow, and innovation utilization total. Lack of

interest is a small statistically significant barrier in patient treatment innovation

utilization (rs = 0.199, p<0.05). Lack of inter-operability is a small statistically

significant barrier in workflow innovation utilization (rs = 0.218, p<0.05). Lack of

time and staffing is a small statistically significant barrier in quality assurance

innovation utilization (rs = -0.178, p<0.05). The negative correlation sign is noted

and will be further discussed in Chapter 5. There are multiple small statistically

significant barriers in treatment planning, such as lack of evidence (rs = 0.203,

p<0.05), complexity (rs = 0.175, p<0.05), lack of time and staffing (rs = 0.237,

p<0.01), lack of inter-operability (rs = 0.203, p<0.05), and lack of reimbursement

(rs = 0.269, p<0.01). There are multiple small statistically significant barriers in

total innovation utilization, such as lack of evidence (rs = 0.161, p<0.05), lack of

interest (rs = 0.264, p<0.01), lack of inter-operability (rs = 0.214, p<0.01), and

lack of reimbursement (rs = 0.176, p<0.05). All statistically significant

correlations in the study were of small to medium effect size class, according to

Cohen’s definition (Ellis, 2010). Thus, the null hypothesis H3,0 is rejected, and the

alternate H3,1 is accepted.

Table 19

Correlations for Barriers in RO Patient Positioning Innovation Utilization

Innovation

Utilization

Score Evidence Complexity

Time/

Staffing

Training/

Support Interest

Inter

Operability Reimbursement

Innovation

Utilization

Score

rs 1.000 -.010 -.084 -.014 -.008 .005 .040 .068

P . .915 .331 .870 .929 .949 .644 .459

N 188 118 135 149 131 144 138 121

Evidence rs -.010 1.000 .475** .199* .441** .279** .271** .272**

P .915 . .000 .032 .000 .003 .005 .007

N 118 127 115 116 109 111 107 96

Complexity rs -.084 .475** 1.000 .430** .553** .265** .352** .161

P .331 .000 . .000 .000 .003 .000 .093

N 135 115 146 134 119 125 123 110

Time/

Staffing

rs -.014 .199* .430** 1.000 .535** .387** .208* .218*

P .870 .032 .000 . .000 .000 .018 .018

N 149 116 134 165 136 142 130 117

Training/

Support

rs -.008 .441** .553** .535** 1.000 .527** .404** .299**

P .929 .000 .000 .000 . .000 .000 .002

N 131 109 119 136 142 128 117 105

Interest rs .005 .279** .265** .387** .527** 1.000 .432** .357**

P .949 .003 .003 .000 .000 . .000 .000

N 144 111 125 142 128 162 131 119

Inter

Operability

rs .040 .271** .352** .208* .404** .432** 1.000 .469**

P .644 .005 .000 .018 .000 .000 . .000

N 138 107 123 130 117 131 150 111

Reimbursement rs .068 .272** .161 .218* .299** .357** .469** 1.000

P .459 .007 .093 .018 .002 .000 .000

N 121 96 110 117 105 119 111 134

*. Correlation is significant at the 0.05 level (2-tailed).

**. Correlation is significant at the 0.01 level (2-tailed).

Table 20

Correlations for Barriers in RO Patient Treatment Innovation Utilization

Innovation

Utilization

Score Evidence Complexity

Time/

Staffing

Training/

Support Interest

Inter

Operability Reimbursement

Innovation

Utilization

Score

rs 1.000 .187 -.009 .032 .054 .199*.170 .177

P . .066 .927 .721 .595 .030 .086 .085

N 198 97 118 129 101 119 103 96

Evidence rs .187 1.000 .560** .452** .547** .426** .413** .385**

P .066 ..000 .000 .000 .000 .000 .001

N 97 102 91 91 82 86 82 75

Complexity rs -.009 .560** 1.000 .534** .664** .270** .471** .402**

P .927 .000 ..000 .000 .007 .000 .000

N 118 91 122 112 95 99 93 86

Time/

Staffing

rs .032 .452** .534** 1.000 .558** .367** .280** .364**

P .721 .000 .000 ..000 .000 .005 .000

N 129 91 112 132 100 106 98 90

Training/

Support

rs .054 .547** .664** .558** 1.000 .455** .545** .407**

P .595 .000 .000 .000 ..000 .000 .000

N 101 82 95 100 105 92 86 80

Interest rs .199* .426** .270** .367** .455** 1.000 .474** .519**

P .030 .000 .007 .000 .000 ..000 .000

N 119 86 99 106 92 122 95 85

Inter

Operability

rs .170 .413** .471** .280** .545** .474** 1.000 .513**

P .086 .000 .000 .005 .000 .000 . .000

N 103 82 93 98 86 95 107 83

Reimbursement rs .177 .385** .402** .364** .407** .519** .513** 1.000

P .085 .001 .000 .000 .000 .000 .000 .

N 96 75 86 90 80 85 83 100

*. Correlation is significant at the 0.05 level (2-tailed).

**. Correlation is significant at the 0.01 level (2-tailed).

Table 21

Correlations for Barriers in RO Treatment Planning Innovation Utilization

Innovation

Utilization

Score Evidence Complexity

Time/

Staffing

Training/

Support Interest

Inter

Operability Reimbursement

Innovation

Utilization

Score

rs 1.000 .203*.175*.092 .237** .129 .203* .269**

P . .030 .042 .255 .005 .140 .030 .004

N 186 115 135 155 136 133 114 111

Evidence rs .203* 1.000 .361** .308** .456** .344** .529** .357**

P .030 ..000 .001 .000 .000 .000 .001

N 115 119 107 112 102 105 91 91

Complexity rs .175* .361** 1.000 .470** .495** .392** .517** .532**

P .042 .000 ..000 .000 .000 .000 .000

N 135 107 143 138 126 120 101 98

Time/

Staffing

rs .092 .308** .470** 1.000 .580** .426** .295** .477**

P .255 .001 .000 ..000 .000 .002 .000

N 155 112 138 163 135 131 108 108

Training/

Support

rs .237** .456** .495** .580** 1.000 .399** .574** .376**

P .005 .000 .000 .000 ..000 .000 .000

N 136 102 126 135 144 120 105 100

Interest rs .129 .344** .392** .426** .399** 1.000 .477** .543**

P .140 .000 .000 .000 .000 ..000 .000

N 133 105 120 131 120 145 105 101

Inter

Operability

rs .203* .529** .517** .295** .574** .477** 1.000 .522**

P .030 .000 .000 .002 .000 .000 . .000

N 114 91 101 108 105 105 119 90

Reimbursement rs .269** .357** .532** .477** .376** .543** .522** 1.000

P .004 .001 .000 .000 .000 .000 .000 .

N 111 91 98 108 100 101 90 119

*. Correlation is significant at the 0.05 level (2-tailed).

**. Correlation is significant at the 0.01 level (2-tailed).

Table 22

Correlations for RO Quality Assurance Innovation Utilization

Innovation

Utilization

Score Evidence Complexity

Time/

Staffing

Training/

Support Interest

Inter

Operability Reimbursement

Innovation

Utilization

Score

rs 1.000 .124 .037 -.178*-.083 -.168 -.105 -.027

P . .225 .707 .036 .388 .059 .264 .800

N 171 97 108 140 110 128 116 91

Evidence rs .124 1.000 .503** .383** .470** .377** .466** .338**

P .225 ..000 .000 .000 .000 .000 .002

N 97 108 90 99 87 96 92 81

Complexity rs .037 .503** 1.000 .531** .582** .355** .442** .312**

P .707 .000 ..000 .000 .000 .000 .005

N 108 90 117 113 101 99 97 78

Time/

Staffing

rs -.178* .383** .531** 1.000 .651** .538** .490** .222*

p .036 .000 .000 ..000 .000 .000 .038

N 140 99 113 155 115 130 114 88

Training/

Support

rs -.083 .470** .582** .651** 1.000 .626** .573** .234*

p .388 .000 .000 .000 ..000 .000 .042

N 110 87 101 115 120 105 95 76

Interest rs -.168 .377** .355** .538** .626** 1.000 .507** .482**

p .059 .000 .000 .000 .000 ..000 .000

N 128 96 99 130 105 147 107 89

Inter

Operability

rs -.105 .466** .442** .490** .573** .507** 1.000 .416**

p .264 .000 .000 .000 .000 .000 . .000

N 116 92 97 114 95 107 125 80

Reimbursement rs -.027 .338** .312** .222*.234*.482** .416** 1.000

p .800 .002 .005 .038 .042 .000 .000 .

N 91 81 78 88 76 89 80 102

*. Correlation is significant at the 0.05 level (2-tailed).

**. Correlation is significant at the 0.01 level (2-tailed).

Table 23

Correlations for RO Workflow Innovation Utilization

Innovation

Utilization

Score Evidence Complexity

Time/

Staffing

Training/

Support Interest

Inter

Operability Reimbursement

Innovation

Utilization

Score

rs 1.000 .111 .118 -.051 -.050 -.034 .210* -.034

p . .303 .209 .525 .581 .679 .033 .746

N 180 88 115 155 125 149 103 91

Evidence rs .111 1.000 .420** .295** .411** .546** .543** .387**

p .303 ..000 .006 .000 .000 .000 .001

N 88 91 87 86 85 82 80 75

Complexity rs .118 .420** 1.000 .475** .642** .278** .539** .309**

p .209 .000 ..000 .000 .003 .000 .004

N 115 87 120 115 103 109 92 84

Time/

Staffing

rs -.051 .295** .475** 1.000 .625** .497** .202* .407**

p .525 .006 .000 ..000 .000 .042 .000

N 155 86 115 161 122 146 102 92

Training/

Support

rs -.050 .411** .642** .625** 1.000 .486** .437** .319**

p .581 .000 .000 .000 ..000 .000 .003

N 125 85 103 122 130 119 98 83

Interest rs -.034 .546** .278** .497** .486** 1.000 .278** .428**

p .679 .000 .003 .000 .000 ..005 .000

N 149 82 109 146 119 156 101 89

Inter

Operability

rs .210* .543** .539** .202*.437** .278** 1.000 .352**

p .033 .000 .000 .042 .000 .005 . .002

N 103 80 92 102 98 101 106 74

Reimbursement rs -.034 .387** .309** .407** .319** .428** .352** 1.000

p .746 .001 .004 .000 .003 .000 .002 .

N 91 75 84 92 83 89 74 95

*. Correlation is significant at the 0.05 level (2-tailed).

**. Correlation is significant at the 0.01 level (2-tailed).

Table 24

Correlations for RO Innovation Utilization Score Total

Innovation

Utilization

Score Evidence Complexity

Time/

Staffing

Training/

Support Interest

Inter

Operability Reimbursement

Innovation

Utilization

Score

rs 1.000 .161*.085 .111 .027 .264** .214** .176*

p . .038 .254 .120 .722 .000 .005 .025

N 213 166 183 199 180 195 174 162

Evidence rs .161* 1.000 .458** .353** .517** .391** .495** .445**

p .038 ..000 .000 .000 .000 .000 .000

N 166 166 159 165 155 161 152 145

Complexity rs .085 .458** 1.000 .424** .596** .252** .478** .372**

p .254 .000 ..000 .000 .001 .000 .000

N 183 159 183 183 169 177 164 150

Time/

Staffing

rs .111 .353** .424** 1.000 .609** .500** .393** .394**

p .120 .000 .000 ..000 .000 .000 .000

N 199 165 183 199 178 191 171 159

Training/

Support

rs .027 .517** .596** .609** 1.000 .469** .505** .355**

p .722 .000 .000 .000 ..000 .000 .000

N 180 155 169 178 180 173 160 149

Interest rs .264** .391** .252** .500** .469** 1.000 .384** .435**

p .000 .000 .001 .000 .000 ..000 .000

N 195 161 177 191 173 195 165 157

Inter

Operability

rs .214** .495** .478** .393** .505** .384** 1.000 .569**

p .005 .000 .000 .000 .000 .000 . .000

N 174 152 164 171 160 165 174 147

Reimbursement rs .176* .445** .372** .394** .355** .435** .569** 1.000

p .025 .000 .000 .000 .000 .000 .000 .

N 162 145 150 159 149 157 147 162

*. Correlation is significant at the 0.05 level (2-tailed).

**. Correlation is significant at the 0.01 level (2-tailed).

Table 25

Summary of RO Barrier Coefficients in all Categories for MPs

Patient

positioning

Patient

Treatment

Planning QA Workflow Total

Evidence 0.203* 0.161*

Complexity 0.175*

Time/Staffing 0.237**

0.178*

Training/Support

Interest 0.199* 0.264**

Interoperability 0.203* 0.218* 0.214**

Reimbursement 0.269** 0.176*

*. Correlation is significant at the 0.05 level (2-tailed).

**. Correlation is significant at the 0.01 level (2-tailed).

Summary

The three research questions posed were answered. The null hypotheses

H1a, H1b, and H3were rejected, and null hypothesis H2 was accepted. There are

statistically significant differences in the accessibility of innovative services based

on hospital university affiliation and urbanicity, with more innovative services

being available to patients treated in urban, university hospitals. There are

statistically significant predicting factors that affect the accessibility of innovative

services, with MP appreciation and motivation having a weak predictive

relationship with increased accessibility of innovative services, albeit the model is

not statistically significant. There are statistically significant barriers MPs face in

providing access to innovative services to patients. In Chapter 5 there will be a

detailed discussion on the implications of these results and recommendations for

future research.

Chapter 5: Discussion, Conclusions, and Recommendations

Introduction

The purpose of this study was to investigate the accessibility of innovative

services in RO in the United States and assess possible diffusion patterns using an

online survey of MPs practicing in the United States. The dependent variable was

innovation score. Independent variables were geographic location; practice details

(university affiliation, reporting structure, and size of physics group);

demographics (age, gender, DABR status, residency status, meeting attendance,

and education level); and context-specific psychological antecedents (motivation,

appreciation, and leadership).

There were statistically significant differences in the accessibility of

innovative services based on hospital university affiliation and urbanicity, with

more innovative services being available to patients treated in urban, university

hospitals. Also, there are statistically significant predicting factors that affect the

accessibility of innovative services, with MP appreciation and motivation having

a weak predictive relationship with increased accessibility of innovative services.

Finally, there were statistically significant barriers in patient treatment, treatment

planning, quality assurance, and workflow. Findings indicate that MPs face

challenges in providing access to innovative services to patients. In this chapter,

the results are discussed in context, the study limitations, and recommendations

for future research are made. The chapter concludes with a conclusion to the

study.

100

Sample Characteristics

Known proportions of organizational and personal characteristics were

retrieved from the most recent salary survey for RO MPs (American Association

of Physicists in Medicine, 2018). This report lists primary employment settings as

private or community hospital, government hospital, medical school or university

hospital, college or university, government hospital, medical physics service

group, physician group, industrial or commercial firm, and cancer center. It is

unclear if some categories are correlated--for example, there can be a cancer

center with a university affiliation. This self-identification can be subjective; this

is the reason this classification was not selected for this study. The report is

useful, however, in providing some baseline descriptive statistics. Based on this

report, 17% of master’s-level MPs work in centers with university affiliation, and

83% of PhD physicists work in centers with university affiliation. On average,

50% of MPs are employed by centers with a university affiliation, 26% of MPs

are women, 55% hold a master’s degree, 89% are certified by the ABR, and 28%

graduated from a residency.

In analyzing the sample descriptive statistics and comparison to known

proportions, it is concluded that there are similarities in percentages of university

affiliation, type of degree, and DABR status, but not for gender and residency.

The respondent gender ratio was higher for male and higher for residency

graduates. There are no publicly available proportions for age distribution, but the

respondent age distribution appears to be skewed towards younger respondents,

101

which is in alignment with the higher response rate for residency graduates. No

exact known proportions are known for urbanicity, meetings attended, reporting

structure, number of physicists, volunteer status, appreciation or motivation. It is

thus concluded that the sample was overall representative of the population.

Open-Ended Feedback

The study included a free-text comment section at the end of the survey.

Many of the comments revolved around the cost of innovation, which was

expected. An unexpected common thread, however, was concerns about

maintaining safety. MP participants were highly concerned that using innovations

and changing the status quo might be inconsistent with maintaining patient safety.

This finding was unexpected because all innovations in question are FDA-

approved and commercially available. The pathway to deteriorating safety would

be possible only through poor implementation. This circles back to workflow

innovation. A conclusion is that MPs do not feel innovation improves patient

safety and/or they do not know how to safely implement innovations in the clinic.

Interpretation of the Findings

University Affiliation and Innovation

In Research Question 1a it was demonstrated that centers with a university

affiliation have a higher mean innovation score than centers without a university

affiliation (MD = 1.65, 95% CI [0.38,2.917], t(211) = 2.56, p = 0.011, d = 0.351).

As discussed in the literature review in Chapter 2, there are many outcome

differences between academic and nonacademic centers. While the difference in

102

innovation score is likely not the only factor contributing to outcome differences,

it is a factor that needs to be incorporated in future models. The results of this

study are in congruence with similar studies performed in the United States (see

Pfister et al., 2015). It is interesting to note that the two categories with

statistically significant results are patient treatment and workflow. The parameters

affecting the patient treatment innovation score (stereotactic body radiosurgery,

stereotactic cranial radiosurgery, robotic radiosurgery, intraoperative radiation

therapy, and flattening free beams) are techniques that are only available in newer

accelerators, which are in their majority multimillion-dollar investments.

In Research Question 1a it was also demonstrated that centers with

university affiliation have a higher mean workflow utilization score than centers

without a university affiliation (MD = 7.09, 95% CI [0.78,13.39], t(178) = 2.217,

p = 0.028, d = 0.330). This finding is interesting because organizational

innovation has not previously been studied in RO in the United States, as typically

publications focus on technological differences (see Chen, 2014)). The results of

this study are in congruence with the published results from the Netherlands (M.

Jacobs, 2017). Improving workflows can be a low-risk, high-yield opportunity for

many centers lacking the funds for large investments. A curriculum with core and

adjunct tools for MPs is currently under development through the Medical

Physics Leadership Academy Working Group (J. Johnson, personal

communication, MPLAW Retreat, May 2019). The lack of statistically significant

differences in other categories is a positive finding for the industry, as it implies

103

that once centers break through the barrier of purchasing innovative technologies,

there are no major differences in utilizing them.

Urbanicity and Innovation

In Research Question 1b it was demonstrated that urban centers have a

higher innovation scores in patient positioning (MD = 0.31, 95% CI

[0.011,0.612], t(203) = 2.043, p = 0.042, d = 0.293) and patient treatment (MD =

0.62, 95% CI [0.23,1.00], t(203) = 3.145, p = 0.002, d = 0.457). Thus, the results

of this study are in congruence with similar studies performed in the United States

(Charlton et al., 2015). The quantitative assessment did not reveal large

differences for total innovation, which was an unexpected result. This is possibly

due either to the low power of the study or the RUCA classification that was used

in this study; greater granularity may be necessary. The results were plotted on a

map of the United States for qualitative analysis, as shown in Figure 3. The

population was superimposed with innovation score (darker green, higher

population). Additionally, all RO the centers that are currently operational in the

United States are superimposed as black squares (IAEA DIRAC database, 2018).

The heatmap represents centers that are more innovative (red) versus less

innovative (blue). The most innovative centers are in close proximity and in areas

with high population density. Conversely, areas with low population density have

the lowest innovation score. This qualitative assessment does support the claim

that urban centers provide more innovative treatments, despite the absence of

large effect sizes and statistical significance.

104

Figure 3. Map of the United States with innovation utilization score. Hawaii is

not to scale. No data were received for Alaska. Map layers include population

density as shades of green and RO centers as squares.

105

While differences in patient treatment are relatively easy to explain due to

purchasing decision and competition in urban centers, the differences in patient

positioning may not be so obvious. Patient positioning is typically decided at the

time of simulation and is the primary responsibility of the radiation therapists.

Historically there is great variation in MPs involvement in patient positioning and

setup reproducibility, with some MPs being very involved, and some MPs being

absent in the simulation process (Clements et al., 2018). The introduction of

mandatory MP residencies is closing this gap. The increase in hypo-fractionated

treatments has also changed this dynamic, as discussed in the 2014 AAPM

summer school on “safely and accurately delivering high precision, hypo-

fractionated treatments” and AAPM reports (Halvorsen et al., 2017). However,

there may be discordance of information flowing to the American Society of

Radiologic Technologists. Another possible explanation is that in urban centers,

patients “shop around” for their treatment, with higher socioeconomic status

patients often requesting or demanding certain types of treatment (Martin,

Thomas, Harden, & Burnet, 2015; Roubou & Alexopoulou, 2015). Frequent

examples from personal experience include prone breast treatments or large full-

body immobilization.

Education and Innovation

In Research Question 2, it was demonstrated that the mean innovation

utilization score for MPs with a Master’s degree is higher than the mean

innovation utilization score for Doctoral degree (MD = 5.06, 95% CI[0.18,9.94],

106

t(188) = 2.046, p = 0.042, d = 0.278). Out of the 118 respondents with a Master’s

degree, 41.5% were employed by a center with a university affiliation, and 58.5%

of respondents were employed by a center without a university affiliation. Out of

the 72 respondents with a Doctoral degree, 58.3% were employed by a center

with a university affiliation, and 41.7% of respondents were employed by a center

without a university affiliation. The survey did not ask participants the topic of

their doctoral degree. It is possible that MPs with doctoral degrees in medical

physics are employed in university hospitals with higher innovation scores, while

MPs with doctoral degrees in other fields are employed in non-university

hospitals with lower innovation scores. This discrepancy does not exist for

Master’s degree, as the Master’s degree must be in Medical Physics to qualify for

certification. Even though the results are statistically significant, education is

likely to have a more complicated relationship with innovation. It is possible that

Masters level physicists are more motivated and competitive, but this requires

further study. This may partially explain the poor model fit in Research Question

Appreciation and Motivation in Innovation

In Research Question 2, it was demonstrated that there is a small positive

correlation between innovation utilization and participant appreciation (rs = 0.224,

p = 0.002) and motivation (rs = 0.215, p = 0.003). Even though the correlations

were small, the results are in congruence with prior published studies in general

and in the health care setting (Wisdom et al., 2014). It is important to note that

107

this is the first time these parameters have been measured for MPs, and results are

in agreement with studies done on other health care professionals (Strömgren,

Eriksson, Bergman, & Dellve, 2016). Appreciation and motivation are often

considered “soft skills” that may be shunned by MPs in leadership positions

(Gutierrez, Halvorsen, & Rong, 2017). This common misconception is declining

since the introduction of the Medical Physics Leadership Academy and the 2016

summer school (J. Johnson, personal communication, July 14, 2019). The results

of this study will serve to strengthen the base of evidence supporting intra-

personal skills and clinical performance.

Barriers to Innovation Utilization

Four categories were found to have statistically significant barriers: patient

treatment, treatment planning, quality assurance, and workflow. Statistically

significant barriers were also detected for innovation utilization total. There were

no statistically significant barriers detected for patient positioning, possibly due to

lack of statistical power. Training and support was not a statistically significant

barrier for any of the categories investigated. This result is in agreement with the

AAPM TG 100 report listing training as the least effective tool of a quality

management program (Huq et al., 2016).

In the patient treatment category, lack of interest is a statistically

significant but with small effect size barrier in patient treatment innovation

utilization (rs = 0.199, p<0.05). It is noted that the patient treatment category

involved very advanced techniques, see Appendix D for details. This lack of

108

interest can be further explained in the context of university affiliation and

urbanicity. Interestingly, university centers in metropolitan areas had even higher

correlation coefficients for lack of interest in patient treatment innovation

utilization (results not shown). This finding may be an indicator of complacency

due to market domination (Martin et al., 2015; Roubou & Alexopoulou, 2015).

In the treatment planning category, five out of the seven barriers were

found to be statistically significant: lack of evidence (rs = 0.203, p<0.05),

complexity (rs = 0.175, p<0.05), lack of time and staffing (rs = 0.237, p<0.01),

lack of inter-operability (rs = 0.203, p<0.05), and lack of reimbursement (rs =

0.269, p<0.01). This is an alarming result; even though the coefficients are small,

there are multiple barriers. The highest coefficient, reimbursement, is the obvious

first choice for discussion. The treatment planning Current Procedural

Terminology (CPT) codes have been steadily decreasing, stagnant or bundled in

(“Reimbursement-American Society for Radiation Oncology (ASTRO)”, n.d.). In

health care in general, it has been well demonstrated that financial incentives

increase innovation utilization (Darzi & Parston, 2013). This conlusion is

congruent with the second-highest coefficient, lack of time and staffing resources.

In the absence of reimbursement, faced with low staffing support, the barrier to

innovation is significant. Facilities with mixed vendor products may face high

inter-operability issues, which are unfortunately unsolvable at the clinic level.

There is a growing number of publications on evidence for treatment planning

innovations and their effect on productivity and outcomes (Gintz et al., 2016;

109

Wang, Dong, Liu, & Xing, 2017), however, the results of this study show that

critical mass has not yet been reached. The complexity of treatment planning

innovations continues to be discouraging to users. These results may be of

particular interest to treatment planning vendors and to AAPM workgroups

writing recommendations for treatment planning.

In the quality assurance category, lack of time and staffing is a small

negative statistically significant barrier in quality assurance innovation utilization

(rs = -0.178, p<0.05). The negative relationship means that lower staffing levels

are correlated to higher innovation in quality assurance. This result can be

explained in the context of daily clinical work. It is reasonable to assume that

MPs have been forced into using innovations in quality assurance to compensate

for reduced staffing levels. For example, the first category indicator, portal

dosimetry, may take some initial time to set up, but the time gains for performing

device-less quality assurance for every patient multiple times per week is a

significant efficiency gain in the clinic. A similar concept applies to other

indicators in the quality assurance category (Eckhause et al., 2015; Thompson et

al., 2018). While it is positive that MPs are using quality assurance innovations,

the instigating factor may be convenience. This is in contrast with treatment

planning findings: when there is reduced staffing or time for treatment planning,

MPs do not use the innovations to gain time (for example with automatic planning

or automatic contouring) but instead do not use these innovations. This may be

due to the significant resource investment needed to create some of these

110

downstream benefits, and the involvement of other groups, such as dosimetrists

and radiation oncologists, who may not be positive towards these innovations

either (Georg & Thwaites, 2017). In contrast, physicists operate independently in

tasks in the quality assurance category.

In the workflow category, lack of inter-operability is a small statistically

significant barrier in workflow innovation utilization (rs = 0.218, p < 0.05). In the

context of workflow, inter-operability is not to be interpreted in a technical

context, but more as the compatibility of heterogeneous business processes. In

this frame of reference, MPs are encountering barriers in workflow innovations

when the innovations are perceived to be incompatible with existing policies and

procedures or standards of practice. This result is in agreement with prior results

in the Netherlands (M. Jacobs et al., 2017). It is possible that MPs have so far

been reluctant to get involved in process improvement and business development

planning. As previously discussed, these avenues are now being explored by the

AAPM and the Leadership Academy.

In total innovation utilization, there are multiple small statistically

significant barriers, such as lack of evidence (rs = 0.161, p < 0.05), lack of interest

(rs = 0.264, p<0.01), lack of inter-operability (rs = 0.214, p < 0.01), and lack of

reimbursement (rs = 0.176, p<0.05). The coefficients are small, yet statistically

significant. MPs face non-supportive organizational cultures in their clinics. This

is a key finding of this study. Clinics as organizational entities have selected to

reject innovations and maintain status quo, possibly due to the incorrect

111

association of innovations as a compromise to safety, the cornerstone of modern

RO.

Limitations of the Study

The presented study has limitations. Due to the study design, there was a

possible selection bias. The study may not have reached some MPs, especially

those who practice in rural areas. Since there are no known proportions of MPs

per ZIP code, the effect of this limitation is impossible to calculate. Comparison

with known proportions of university versus non-university centers showed a

reasonable degree of agreement, which implies that selection bias was not a

significant source of bias in this study. Another possible source of bias mentioned

in Chapter 1, information bias, could also have influenced the results.

Unfortunately, there is no way to assess the magnitude of this effect either. Both

of selection bias and information bias are inherent to the study design.

Additionally, as mentioned in Chapter 4, there was a high level of internal

consistency, as determined by Cronbach's alpha of 0.963. This statistic in

combination with the face and content validity of the expert panel review leads to

the conclusion that the constructs have high reliability. However, there were many

assumptions made in the operationalization of constructs. It is possible that not all

predictive parameters were included in the model, or operationalized

appropriately. This would partially explain the poor fit of the predictive model.

Additionally, the operationalization of constructs may not be transferable outside

112

the United States, thus results should be applied with caution outside of the

United States.

Furthermore, there are statistical limitations. The effect sizes used to

calculate a priori power were hypothetical and chosen conservatively. Post hoc

analysis for university status reveals that based on the detected sample effect size,

the power of the study was 0.72. This may be slightly smaller than the intended

0.8, yet still within reason. Conversely, the power for the RUCA continuum

ranged from 0.52 to 0.88, depending on the model selected. This is because of the

selected RUCA continuum and the low response rates from areas closer to the

rural end of the spectrum. It is uncertain if the effect sizes measured in this study

are true population effect sizes or sample effect sizes, thus results should be

interpreted with caution until effect sizes are confirmed by future studies.

Lastly, this was a cross-sectional study, thus the study design does not

allow the investigation of temporal relationships and possible causality between

the dependent and independent variables. Causal effects are typically

demonstrated by experimental study designs (Salazar et al., 2015). However, the

results are congruent with theoretical causal structures used in population health

(Darzi & Parston, 2013).

Recommendations

It is recommended that this study can be repeated in the future under the

aegis of the AAPM, possibly by a task group or workgroup. If so, the

questionnaire or survey should be designed to limit selection and information

113

bias, the two biggest limitations of this study. This could be achieved by

addressing the study to the department chair, manager, or chief physicist.

Changing the unit of measurement to the RO department, instead of the individual

MPs could also serve to minimize the aforementioned sources of bias. Even

without AAPM support, the results of this study should be validated by other

researchers. It would be particularly interesting to perform similar studies outside

the United States, assess generalizability and compare effect sizes.

Additionally, it is recommended that there is further theoretical

exploration on appreciation and motivation in health care leadership theory. There

is rich literature on emotional intelligence, organizational culture, and employee

motivation, but the connection to appreciation is not obvious. Very little has been

written about appreciation in the work environment. The Society for Human

Resource Management, (2012) claims that even though 51% of supervisors claim

they recognize good performance, only 17% of the same organization’s

employees feel recognized. The term “appreciative leadership” is often used in

popular management literature. Scholarly work does address appreciative

behaviors relating to praise and recognition, using verbiages such as support,

respect, constructive feedback, social reinforcement, and appreciation (Stocker,

Jacobshagen, Krings, Pfister, & Semmer, 2014). A possible expansion to the path-

goal theory of leadership (House, 1996), as it applies to healthcare, may be

necessary to fill this gap.

114

Another important aspect for further research is the transferability to

Radiation Therapy Technologists and Radiation Oncologists. These professionals,

along with MPs are in the front lines of daily clinical practice. The American

Society for Radiologic Technologists (ASRT) would be an obvious choice for

collaboration and dissemination of these findings. Similarly, collaborating with

Radiation Oncologists and ASTRO would be the logical route to further the

presented research and delve into qualitative and quantitative aspects of

organizational culture. MPs do not practice in isolation. It is possible that a better

predictive model could be derived using inter-team and intra-team dynamics

(Reiter-Palmon, 2017). Removing the barriers to innovation utilization in the

clinic must be a collaborative team effort.

Implications

Instrument Development and Benchmarking

During the literature review stage of this study, there was an evident

relationship between the ninth essential public health service to “evaluate

effectiveness, accessibility, and quality of personal and population-based health

services” and innovation accessibility in RO. There was no available data or an

instrument to effectively measure innovation in RO. This gap in itself is a barrier

to improving the equity of delivery of appropriate treatment to all cancer patients.

The need for a metric of innovation utilization was previously an abstract concept

discussed only in theory. This study has now delivered an instrument to quantify

innovation in RO. This quantitative instrument, along with qualitative work done

115

by others, will be used to improve innovation utilization in RO. The data collected

in this study can serve as a benchmark of the state of innovation today, with plans

on how to improve it in the future.

Social Change Implications

Inequalities in health are parallel to inequalities in health care (Frenk,

2013). To improve public health further in the 21st century, there needs to be an

inclusion of factors outside of traditionally-defined health care (DeSalvo, 2017).

Disparities in access to advanced care have an impact on cancer survival. This

statement may be considered contradictory by some, but it is well supported by

recent literature (Nardi et al., 2016; Pfister et al., 2015; Wolfson, Sun, Wyatt,

Hurria, & Bhatia, 2015). There are is an abundance of differences between centers

that may have a causal effect on improved cancer survival. Innovation is only one

of these parameters. This study did not attempt to show causal effects, as this can

only be demonstrated by clinical trials (B. L. Jacobs et al., 2017). What this study

did demonstrate, however, is that there are indeed differences in innovation

accessibility in RO in the United States. The connection between innovation and

improved cancer survival has been made by many authors; innovation-based care

models are under discussion in reimbursement health care reform (Alvarnas,

Majkowski, & Levine, 2015; Nardi et al., 2016). Thus, using the results of this

study to further how innovation is measured in RO in the United States, and

assess how the measured barriers can be minimized is a positive social change.

Public health is expanding beyond government agency programs to a broader

116

cross-sectoral practice (DeSalvo, 2017). RO as a community is in a position to

further engage public health aspects that have a collective impact on population

health.

Conclusion

In this study, innovation in RO in the United States was measured for the

first time, through the development of a new survey instrument. Rural centers and

centers with no academic affiliation are trailing behind in innovation

implementation. Motivation and appreciation were shown to be statistically

significant personal factors influencing innovation utilization, but no predictive

model was possible. Barriers in RO innovation implementation were also

assessed, with treatment planning showing the most statistically significant

barriers.

RO practitioners follow an ethos of “as low as reasonably achievable”

every day, making every attempt possible to reduce dose to patients. We do this

almost subconsciously, as it has been engrained in our training as common sense.

Getting MPs to view embracing innovation as part of their culture will only be

possible if innovation is not considered to be competing with safety. The

complementary relationship between safety and innovation is being discussed in

many other health care fields. If every MPs practiced using “as innovative as

reasonably achievable” as their mantra, similar to “as low as reasonably

achievable”, patients would benefit immensely. This is a worthy end goal to be

striving towards. This study provides a small but promising step in this direction.

117

Although the exact number of lives saved or extended because of innovations in

daily practices in RO may never be known, it is certainly worth it to try to make

every treatment as innovative as reasonably achievable.

Students also viewed