Healthcare Analytics Framework

9/14/22, 11:34 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 1/41

Part II Strategies, Frameworks, and Challenges for Health Analytics

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 2/41

5 Grasping the Brass Ring to Improve Healthcare Through Analytics: The Fundamentals

Dwight McNeill

The U. S. healthcare industry faces enormous challenges. Its outcomes are the worst of its peer wealthy countries, its efficiency is the worst of any industry, and its customer engagement ratings are the worst of any industry. Although the industry is

profitable overall, ranking fourteenth among the top 35 industries,1

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end01) it has had difficulty in converting these challenges into business opportunities to do good by improving the health of its customers while doing well for its stockholders.

One of the paradoxes of healthcare is that it uses science (a.k.a. analytics) more than any other industry in the discovery process —that is, to understand causes of diseases and develop new treatments. Yet there is a tremendous voltage drop in deploying this knowledge in the delivery of care and the production of health. McGlynn et al. reported in their classic paper “The Quality of Healthcare delivered in the United States” that patients received recommended care about 55% of the time and that these

deficits “pose serious threats to the health of the American people.”2

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end02) This percentage is close to a coin toss. An example of this shortfall in translating research into practice is the mortality rate amenable to healthcare. Included in this mortality rate are deaths from diseases with a well-known clinical understanding of their prevention and treatment, including ischemic heart disease, diabetes, stroke, and bacterial infections. In other words the science is clear on what needs to be done, and the premature mortality rate from these diseases is an important indicator of the success on executing on the science. It turns out that the mortality rate of the United States is worst among 15 peer wealthy nations. In fact, it is 40% higher than the average mortality rate of the best five countries. This translates into 118,000 lives that would have been saved if one simply lived in

these other countries.3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end03)

The reasons for the voltage drop are many. One is the predilection of highly trained and autonomous practitioners (doctors), who drive 80% of healthcare activity and costs, to rely on intuition rather than data to drive their decision making. This is related to not having information at their fingertips for decision making because it is either not there (research not digested) or is known but not findable (not digitized or available in electronic or paper records). For example, the human error rate in

medical diagnosis is 17%.4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end04) It remains to be seen if the IBM Jeopardy! Watson application—incorporating natural language processing and predictive analytics for differential diagnosis—

will reduce the rate, but it seems likely.5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end05) What is clear so far is that much of the voltage drop has to with the sociology, for example, changing cultures and behaviors, and not the technology. Going forward, analytics needs to make the case that it can produce compelling solutions to vital business challenges.

Delivering on the Promises The promise of analytics in healthcare is huge. The McKinsey Global Institute states that “if U.S. healthcare were to use big data creatively and effectively to drive efficiency and quality, the sector could create more than $300 billion in value every year” through applications such as comparative effectiveness research, clinical decision support systems, advanced algorithms for

fraud detection, public health surveillance and response, and more.6

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end06) IBM estimates that if all the available IT and analytics solutions that it sells to health plans were fully and successfully deployed, a midsized health plan could net a potential $644

million annually in economic benefit.7 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end07) And many other healthcare technology vendors offer the same proposition. For example, SAP, the largest business software company, states that “with the right information at the right time, anything is possible...and with real-time, predictive analysis comes a shift toward

an increasingly proactive model for managing healthcare.”8 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end08)

It is now time to convert the promise of analytics and the potential benefits into practice and demonstrate results with quantifiable savings in terms of dollars and lives. Although analytics has been around a long time in various guises in healthcare, for example, informatics, actuarial science, operations, and decision support, a tipping point may have been achieved with the proliferation of big data, new technologies to harvest, manage, and make sense of it, and the acute need of business to achieve results and indeed transformation in the wake of the Great Recession and Obamacare. It may be a new ball game for analytics.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 3/41

Fundamental Questions and Answers If we boil it down to the fundamentals, businesses in all industries strive to accomplish two goals: Increase revenues and reduce costs. Figure 5.1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05fig01) provides a healthcare value framework for these goals.

Figure 5.1 Healthcare value framework

On each axis, one for revenue increases and the other for cost reduction, more is attained by going up or to the right, respectively. In terms of cost reduction, gains can be made through operational efficiencies and through medical cost reduction. Clearly medical costs represent the majority of costs, and this is where the largest potential savings can accrue. Similarly on the revenue axis, gains can again be made through operational efficiency, but the larger area of opportunity is improving clinical outcomes. The cell with the most potential that combines optimization of both revenues and costs is transformation of the business to radically reposition the company for greater market opportunities. Examples of questions that address each cell of the matrix and require analytics support are provided in Figure 5.1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05fig01) . For the transformation cell, customer analytics using big personal data is the approach for knowing customers and providing them value. Another example is for operational efficiency, which addresses how to reduce data redundancy, enhance data quality, and have one trusted source of truth. The challenge for analytics and IT is to move out and upward from its usual focus on its own operational efficiency to providing value to the big challenges of the business.

So, this part of the book is devoted to strategies, frameworks, and challenges and includes six chapters that provide some fundamental answers to those wanting to step to the plate and get in the new game of health analytics. (Future chapters will provide how-to applications and best practices.) Some questions discussed in Part II (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/part02#part02) are

• What is health analytics and what are the scope and various options?

Jason Burke, in Chapter 6 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch06#ch06) , “A Taxonomy for Healthcare Analytics (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch06#ch06) ,” asserts that the fundamental improvements needed in health and life sciences will only be realized via the deeper insights offered through analytics. He inventories the options in an analytics continuum that ranges from business analytics to clinical analytics. He includes the following five areas in his analytics taxonomy: clinical and health outcomes, research and development, commercialization, finance and fraud, and business operations. He catalogs an analytics “needscape,” which is an inventory of analytics options that encompasses the various needs of healthcare organizations ranging from supply chain optimization to comparative effectiveness.

• What are the various ways to do analytics?

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 4/41

In Chapter 7 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07) , “Analytics Cheat Sheet (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07) ,” Mike Lampa, Sanjeev Kumar, Raghava Rao, et al., provide a “Rosetta Stone” for analytics to allow beginners to learn the language, including terms, acronyms, and technical jargon, quickly. They address the different types of analytics, for example, forecasting and text mining; analytic processes for enterprise scaling, for example, Six Sigma; sampling techniques, for example, random and cluster; data partitioning techniques, for example, test and validation set; a compendium of key statistical concepts, for example, correlation; modeling algorithms and techniques, for example, multiple regression; times series forecasting, for example, moving average; and model fit and comparison statistics, for example, chi square. All in all, it is a great reference for a quick lookup for those involved in analytics work.

• What is its value to the business and how is it determined?

Pat Saporito, in Chapter 8 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch08#ch08) , “Business Value of Health Analytics (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch08#ch08) ,” demonstrates that healthcare faces many challenges including relatively poor outcomes, inefficiency, and low customer satisfaction. Analytics offers a value proposition to use insights derived from data to solve some difficult business issues. But analytics is funded by the business and the Return on Investment (ROI) must be clear for business to invest in any endeavor. Saporito makes the case that analytics must be aligned with the business, first and foremost. She details a number of ways to prove the value of analytics, overcome biases about analytics, and change the culture to be more fact based in its decision making.

• How do I keep out of trouble with my lawyers about privacy concerns?

Thomas Davenport, in Chapter 9 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09) , “Security, Privacy, and Risk Analytics in Healthcare (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09) ,” makes the convincing case that the future of advanced analytics to meet evolving business needs that relies on deep and diverse data sets, often including personal data, can be put on a fast track or compromised, depending on approaches to identity protection and privacy. He suggests that companies that do it well may achieve an advantage in the marketplace. He presents the kinds of adjustments that leading payers, providers, and life science organizations are making to their information security and privacy practices. He also details how healthcare organizations are using risk analytics to bolster their security practices. He comments that analytic leaders are paying closer attention to data ownership and privacy and that lawyers rather than IT teams will determine how to interpret and protect privacy.

• What are some examples of analytics “secret sauce”?

Dwight McNeill in Chapter 10 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10) , “The Birds and the Bees of Analytics: The Benefits of Cross-Pollination Across Industries (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10) ,” addresses how healthcare can learn from other industries, including retail, banking, sports, and politics. McNeill asserts that industries have unique strengths and get proficient in associated “sweet spot” analytics. Other industries are blinded from them and their potential performance is constrained. He focuses on the following areas: Why analytics innovations matter; how to find and harvest analytics sweet spots; what best practices analytics should be adapted in healthcare; and how to put analytics ideas into action by understanding the innovation adoption decision-making process. He proposes seven adaptations that address seemingly intractable healthcare challenges, such as population health, patient engagement, and provider performance.

Notes 1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end01a) . CNNMoney, “Top Industries: Most

Profitable 2009,” accessed February 28, 2013, http://money.cnn.com/magazines/fortune/global500/2009/performers/industries/profits/ (http://money.cnn.com/magazines/fortune/global500/2009/performers/industries/profits/) .

2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end02a) . E. McGlynn et al., “The Quality of Healthcare Delivered to Adults in the United States,” New England Journal of Medicine 348 (2003): 2635-45.

3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end03a) . D. McNeill, A Framework for Applying Analytics in Healthcare: What Can Be Learned from the Best Practices in Retail, Banking, Politics, and Sports (Upper Saddle River, NJ: FT Press, 2013).

4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end04a) . Agency for Healthcare Research and Quality, Diagnostic Errors, http://psnet.ahrq.gov/primer.aspx?primerID=12 (http://psnet.ahrq.gov/primer.aspx?

primerID=12) .

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 5/41

5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end05a) . IBM, “Memorial Sloan-Kettering Cancer Center, IBM to Collaborate in Applying Watson Technology to Help Oncologists,” March 22, 2012, www- 03.ibm.com/press/us/en/pressrelease/37235.wss (http://www-03.ibm.com/press/us/en/pressrelease/37235.wss) .

6 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end06a) . Basel Kayyali, et al., “The Big-Data Revolution in US Healthcare: Accelerating Value and Innovation,” April 2013, www.mckinsey.com/insights/health_systems/The_big-data_revolution_in_US_health_care (http://www.mckinsey.com/insights/health_systems/The_big-data_revolution_in_US_health_care) .

7 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end07a) . P. Okita, R. Hoyt, D. McNeill, et al., “The Value of Building Sustainable Health Systems: Capturing the Value of Health Plan Transformation,” IBM Center for Applied Insights, 2012.

8 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch05#ch05end08a) . SAP, “Global Healthcare and Big Data,” marketing brochure.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 6/41

6 A Taxonomy for Healthcare Analytics

Jason Burke

As described in Chapter 1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch01#ch01) , “An Overview of Provider, Payer, and Life Sciences Analytics (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch01#ch01) ,” by Thomas Davenport and Marcia Testa, the global healthcare ecosystem—healthcare providers, payers, and life sciences firms of all types—is undergoing a transformation. Though priorities vary across organizations and geographies between cost, safety, efficacy, timeliness, innovation, and productivity, one universal truth has emerged: The fundamental improvements needed in health and life sciences will only be realized via the deeper insights offered through analytics.

Most, if not all, of the analytical capabilities needed to drive systemic changes in healthcare have been available in commercial software for decades. Though a multiplicity of reasons exist why analytics have not been deployed more pervasively and comprehensively within healthcare, the reality is that most health-related institutions today have some limited analytical capability and capacity. As executives and leaders develop their respective institutional transformation plans, there is a need to consistently characterize and assess an organization’s analytical capabilities.

Toward a Health Analytics Taxonomy For organizations looking to grow their analytical competencies, one initial challenge is simply understanding the inventory of options. What are all of the ways that analytics might help transform the business, and how can priorities be developed against those options? What are the focus areas?

Despite areas of analytical progress in niche market topic areas,1

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch06#ch06end01) a common taxonomy for health analytics has yet to emerge. However, some noticeable trends have surfaced:

• As illustrated in Figure 6.1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch06#ch06fig01) , analytical applications in health and life sciences are increasingly being conceptualized as existing on a continuum between

business analytics (e.g., cost, profitability, efficiency) and clinical analytics (e.g., safety, efficacy, targeted therapeutics).2

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch06#ch06end02)

Figure 6.1 Health analytics continuum

• Whereas organizations have created initiatives targeting the extreme ends of that continuum (e.g., an activity-based costing initiative at a hospital), the largest challenges still reside in moving toward the middle of the continuum: linking clinical and business analytics into a more comprehensive view of health outcomes and costs.

• To successfully link the business and clinical perspectives, data from all three traditionally “siloed” markets—care providers, health plans, and researchers/manufacturers—must be joined to produce a more complete picture of quality, efficacy, safety, and cost.

So in summary, the analytically derived insights needed to drive health industry transformations require industrywide collaboration around shared information and common analytical needs that link clinical and business concerns. A common

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 7/41

analytical landscape, though not supported through industrywide consensus at the moment, can a) form the basis of agreement within an organization for purposes of strategic planning and organizational development, and b) begin to offer a common understanding of the analytical underpinnings of meaningful health transformation.

Drafting a Health Analytics Taxonomy At first glance, it may appear to be an impossible task: How can a single representation of analytical needs capture the breadth, depth, and diversity that currently exist within the global healthcare market. And in truth, it probably cannot. Yet despite differing market structures, business models, and incentives, most healthcare organizations have similar analytical needs: how to identify the best treatments, how to operate more profitably, how to engage customers more effectively, and so on. Though the motivations behind undertaking analytical initiatives may vary, both the analyses and their corresponding data are comparable.

At the highest level, we have observed five areas of analytical competencies that modern health organizations—including providers, payers, and life sciences organizations—are discovering will be needed to successfully compete:

1. Clinical and health outcomes analytics—These analytics are related to maximizing the use of existing treatments and therapies. For example, providers and health plans are both driven to ensure the best treatment is pursued for a particular patient, not just patients in general.

2. Research and development analytics—These analytics are related to discovering, researching, and developing novel treatments and therapies. For example, pharmaceutical researchers and providers need to know where potential clinical trial participants can be located to expedite new drug development.

3. Commercialization analytics—These analytics are related to maximizing sales, marketing, and customer relationship efforts. For example, both providers and health plans are motivated to communicate more frequently and effectively with patients regarding products, services, and treatments.

4. Finance and fraud analytics—These analytics, which could be considered part of business operations (discussed next), relate to ensuring the financial health and stability of the organization. They are called out separately here due to the strategic role that claims, fraud, and risk play within healthcare markets.

5. Business operations analytics—These analytics are related to driving productivity, profitability, and compliance across the various business functions of an institution. For example, health plans and pharmaceutical manufacturers are motivated to ensure optimal operation of call center facilities, staff, and assets.

Across these five domains, a spectrum of analytical needs/scenarios can be described, as depicted in Figure 6.2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch06#ch06fig02) . Each box represents a class of analytics- related needs and capabilities; taken together, an “analytical needscape” emerges that describes the overall landscape of analytical concerns that organizations need to address.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 8/41

Figure 6.2 Health analytics “needscape”

Ultimately, organizations may differ in

• What is included in the analytical needscape—The list of analytical possibilities is large if not endless, and organizations may uncover opportunities for innovation in pursuing analytical scenarios not represented in Figure 6.2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch06#ch06fig02) . The two most important aspects of building the needscape are a) accurately reflecting the needs of the organization, and b) reflecting those needs in a way that is generalizable to more than one market so that discussions around shared data and analytical models can proceed unimpaired.

• Where certain analytical needs sit within the needscape—Should “bending the cost curve” (a colloquialism regarding restructuring clinical and administrative cost models to reflect a post-reform U.S. business climate) be in the “Business Operations” or “Finance & Fraud” domain? In the end, it probably doesn’t matter as long as the needs are represented.

• The relative priorities within the needscape—Organizations may competitively differentiate themselves based on business strategies that imply differing priorities across the various analytical scenarios. Alternatively, their business model may naturally reflect a bias toward certain analytics. These priorities may differ by functional business unit within an organization as well.

The construction of an analytical taxonomy provides a common communication vehicle by which organizational and functional leaders can map those aspects of analytical capabilities critical to the long-term health and success of their business.

Notes 1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch06#ch06end01a) . For example, the detection and

prevention of healthcare financial fraud, or the emergence of adaptive clinical research designs within clinical trials.

2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch06#ch06end02a) . Jason Burke, “The World of Health Analytics,” in Health Informatics: Improving Efficiency and Productivity, ed. Stephen Kudyba (CRC Press, 2010).

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 9/41

7 Analytics Cheat Sheet

Mike Lampa, Dr. Sanjeev Kumar, Dr. Raghavendra, Dr. Raghava Rao, R. Jaiprakash, Dr. Brandon Vaughn, Vijeyta Karthik, Arjundas Kesavdas, Elizabeth Thomas

Analytics offers any business a means of differentiating itself from and outperforming its competition, but for organizations just getting into the analytics arena there can be a steep learning curve. The recruitment of highly trained statisticians, econometricians, and data miners certainly enhances an organization’s capability to build and deploy analytics. But often managers and executives find it difficult to communicate with analysts in a domain replete with acronyms, mathematical concepts, and buzzwords. The statistical concepts and distinctions behind advanced analytics can often be difficult to understand. What is the difference between statistics and data mining? What’s a model? What’s an ROC? What’s AIC? Which is better, a high or low value? Making intuitive and relevant sense of it all becomes a big job. Where does one begin?

In this chapter, we take a step toward demystifying analytics by defining the basic and important concepts and processes. A wide expanse of knowledge is available on the Web in the form of tutorials, primers, and whitepapers on statistics, data mining, forecasting, and so on. So we provide brief descriptions and appropriate references so as not to reinvent the wheel. We present the material in a step-by-step manner to give abstract concepts firm grounding and as a series of tables to facilitate quick reference. We’ve tried to make this more than just an analytics glossary, but an analytics “cheat sheet” that enables beginners to get up to speed and provides a quick reference for the seasoned analyst.

Different Types of Analytics Table 7.1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab01) captures the most popular flavors of analytics in use within the business world today. While these types of analytics serve different purposes, there is considerable overlap in the use of techniques across the types; for example, correlation and covariance are common to all types of analytics, and regression-based techniques are popular in both statistics and data mining.

Table 7.1 Types of Analytics

Type Description Statistics The science of collection, analysis, and presentation of data.1

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end01)

Forecasting The estimation of the value of a variable (or set of variables) at some future point in time as a function of past data.

Data mining The automated discovery of patterns in data through the use of computational algorithmic and statistical techniques.

Text mining The process of using statistical, computational algorithmic and natural language processing techniques to extract information from text in a manner similar to data mining.

Optimization A discipline that uses mathematical techniques to answer questions such as, “what is the best that can happen?” Problems within optimization frequently involve finding optimal solutions while satisfying constraints.

Analytic Processes Performing analytics involves various degrees of science and art. Individual analysts typically carry out analysis in a personal manner. However, scaling this out to a team can become difficult for managers. Consequently, adopting an analytical process framework is usually a good way to help standardize a team on analytics process while helping to spread analytics leading practice across the enterprise. Table 7.2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab02)

describes three common analytic processes that represent excellent, tried and tested methods. Note that when selecting an analytic process, it is important to evaluate the pros and cons of each methodology for your business’s and teams’ needs.

Table 7.2 Analytics Processes

Analytic Process

Description

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 10/41

Analytic Process

Description

SEMMA Attributed to SAS Institute, SEMMA is an acronym that refers to the core process of conducting data mining. SEMMA stands for five steps:

S = Sample: An optional step in which a representative subset of data is extracted from the larger data set. This step can also be used when resource constraints render processing against the larger data set unfeasible.

E = Explore: A step in which the analyst can visually or statistically examine the data to uncover, for example, trends and outliers.

M = Modify: A step in which data modifications such as binning and transformations can be applied.

M = Model: A step in which data mining models are built.

A = Assess: The final step in which the performance of models is assessed to test the validity of the model results.

CRISP- DM

Cross Industry Standard Process for Data Mining. A common and popular2

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end02) data mining process model.

CRISP-DM organizes the data mining process into six phases:

• Business understanding

• Data understanding

• Data preparation

• Modeling

• Evaluation

• Deployment

Six Sigma and DMAIC

Six Sigma3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end03) is a business management strategy that seeks to eliminate defects. Six Sigma employs statistical approaches to quality control to ensure that 99.99966% of subject outputs are statistically expected to be free of defects (no more than 3.4 defects per million produced).

DMAIC4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end04) is an improvement methodology that when combined with Six Sigma provides an analytical, measurements-based framework to increase quality and reduce costs. DMAIC has five phases:

D = Define

M = Measure

A = Analyze

I = Improve

C = Control

Data Scales Understanding the different types of data scales is critical to ensuring you do not lose information or introduce erroneous signals into your data sets during data preparation and analysis. Table 7.3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab03) provides a list of data types and what they mean.

Table 7.3 Scales of Data Measurement

Measurement Scale Description

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-7… 11/41

Measurement Scale Description

Categorical/Nominal Pure categorical measures of a variable; they have no size, order, or degree of differences between groups. They are typically labels or attributes (e.g., occupation, eye color, gender, fraud).

Ordinal Ordinal variables are special categorical variables that enable ordering, but have no relative size or degree of difference (i.e., numerical operations such as addition/subtraction have no meaning). Examples would be “Like” survey items (Strongly Disagree, Neutral, Strongly Agree) and education level (elementary school, high school, some college, college graduate). It is important to note that for this type of variable, if numbers are used to represent the categories, then the numbers are just labels, and the difference between levels does not represent magnitude.

Interval Interval variables are numerical attributes that have order, relative size, and degree of difference, but no absolute zero point. For example, most temperature scales are interval variables. It is possible to order values (e.g., 90°F is greater than 40°F). However, most temperature scales do not have an absolute zero (i.e., 0°F does not indicate the absence of a temperature; rather it is a distinct temperature value of significance). Interval measures are more common for manmade scales, while naturally occurring measures (e.g., weight) tend to be ratio scaled.

Ratio Ratio variables are the same as interval variables except that ratio variables have an absolute zero point. The temperature scale, Kelvin, is a prime example of a ratio scale with absolute zero (equal to −273 degrees Celsius). Ratio scales enable comparisons such as 150 degrees is three times as high as 50 degrees. Most numerical variables (e.g., waiting time, salary (measured as a number), height, weight, etc.) would be on a ratio scale. In most statistical analysis, both the interval and ratio scales are treated as numerical variables, and some programs don’t distinguish between them.

Sampling Techniques Traditionally, statistics has employed sampling as a method of overcoming obstacles associated with not having an entire population available. Sampling helps to assemble a representative subset of individuals from a larger population to enable generalization. Table 7.4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab04) lists common sampling techniques.

Table 7.4 Sampling Techniques

Technique Description Sampling The process of selecting a subset of individuals from a larger population with a goal of using the sample data to

generalize back to the general population. Random sampling

The same as sampling, but with a probabilistic component to the selection process that allows every individual in the population a chance of being chosen.

Simple Random Sampling (SRS) with and without replacement

Same as sampling, but each individual has an equal probability of being selected. This is the most basic of all sampling designs and serves as the basis of more complicated random sampling designs.

When SRS is performed with replacement, it means that the individual is placed back in the population and can be selected again. When SRS is performed without replacement it means the individual is not placed back into the population and cannot be selected again. This prevents multiple occurrences of the individual.

Undersampling and oversampling

Forms of sampling that adjust the number of individuals up or down in a sample. For instance, undersampling drops the level of a particular type of individual in a sample while oversampling increases the level of a particular type of individual.

Stratified sampling

A technique that breaks a population into groups or subsets, known as strata. It is important that the groups are homogeneous (heterogeneous groupings would be better served by use of cluster sampling designs). For example, a population can be split by gender, with an SRS of males and a separate SRS of females. This sampling design can either select equal numbers of participants from each group (to ensure equal representation) or in proportion to the size of their strata (to mimic the population and ensure that dominant groups remain dominant in the sample).

Cluster sampling

A technique that takes intact groups of a population and randomly selects groups for study. It is important that the groups are heterogeneous. For example, a city can be divided into areas with no inherent characteristics.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 12/41

Data Partitioning Techniques During model construction, the data set will need to be divided into two or three different data partitions each with a different purpose as defined in Table 7.5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab05) . The partitions are constructed using sampling techniques as defined previously in Table 7.4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab04) . Partitioning should not be confused with sampling. The rationale behind partitioning is to build models that can be used for prediction. To do this one needs to train the model using as much historical data as possible; this data set is termed the training set. After training is complete, it is important to assess the model’s capability to generalize to new data by testing the model on data it has not previously seen (i.e., was not included in the training set); this new data set is called the test set. This is the real test of what the model learned from the training set. Validation is the final step, in which a third data set is created to optimize and tweak the model should it be required. Note it is typical to see training and test sets only.

Table 7.5 Data Partitioning Techniques

Technique Description Data partitioning

The process of splitting the data set into two or three partitions for the purposes of building (training), testing, and validating statistical models. By randomly splitting data into these partitions, it allows the researcher to create “two or three samples” from one: one to build the model, one to test/improve the accuracy of the model, and a final one to provide the final validation of the improved model. Since the second and third data are not used in the model building, model accuracy shown when using the second data set establishes validity of the model. By randomly assigning the original sample into three partitions, a researcher has greater assurance that all three partitions are similar.

Training set A partition of the full data set used to build/train models. A rule of thumb is to divide the sample into training and validation in the ratio 60:40, 70:30, or 80:20 (i.e., the training set is typically larger than the test set since it builds the model). If a third data set is needed, these rules of thumb are adjusted (either so that the test and validation set are the same size—both smaller than the training set—or where the validation set is slightly smaller than the test set).

Test set A partition of the full data set (typically smaller than the training set) used to test how well the model can predict/classify new data previously unseen with a goal to possibly change and improve the model.

Validation set

A partition of the full data set used to test the final model. If the initial model is sufficient without the need for improvements, the test and validation sets are done as a singular action (i.e., one test set) instead of two independent processes.

Statistical Overview—Key Concepts To perform effective analytics, a good understanding of key statistical concepts is required. Table 7.6 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab06) provides a description of key statistical concepts to kick start your learning. For example, understanding the difference between correlation and causation (a common mistake) can help you not make spurious claims from your results.

Table 7.6 Key Statistical Concepts

Key Concepts

Description

Causation Relationship between an event (the cause) and a second event (the effect), where the second event is understood

as a consequence of the first.5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end05)

Correlation A statistical measurement of the relationship between two variables/attributes. Correlation is a necessary but insufficient condition for casual conclusions. The range is from -1 to +1 for traditional measures like r (Pearson Product Moment Correlation Coefficient).

Correlation = +1 (Perfect positive correlation, meaning that both variables move in the same direction together)

Correlation = 0 (No relationship between the variables)

Correlation = -1 (Perfect negative correlation, meaning that as one variable goes up, the other trends downward)

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 13/41

Key Concepts

Description

Covariance A type of variability that measures how much two random variables change together. Covariance typically is the foundation of other metrics (e.g., correlation).

Uncorrelated variables have covariance = 0

With correlated variables covariance is non-zero:

Positive values indicate that the two variables show similar behavior.

Negative values indicate that the two variables show opposite behavior.

Dependent variable or Response variable

The variable whose value you would like to predict or explain. For example, if a company wants to predict customer loyalty using various predictors, customer loyalty would be the dependent variable.

Independent variable or Explanatory variable

A variable used to help predict or explain a dependent variable. For example, if a company wants to predict customer loyalty, the various predictors (e.g., cycle time) would serve as independent variables. Analysis is required to assess the significance between independent variables and dependent variables as the independent variable may not be a significant predictor of the dependent variable. In some analyses (e.g., correlational/regression) if no correlation exists, best practice suggests the removal of the noncorrelated independent variable(s).

Null hypothesis

Under Inferential statistics, one wants to test a claim about a population. This claim is split into two competing hypotheses: the null and alternative hypotheses. The null hypothesis (H0) represents a “status quo” perspective that is often what we are attempting to disprove. The alternative hypothesis (Ha or H1) typically represents the change we want to show. For example, if a company wants to predict customer loyalty based on cycle time performance, a null hypothesis could be “Cycle time is not a significant predictor of customer loyalty” versus the alternative hypothesis “Cycle time is a significant predictor of customer loyalty.” Data is collected and tested to see how “unusual” it is under the temporary assumption that H0 is true. Rare or unusual data (often represented by a p-value below a specified threshold) is an indication that H0 is false, which constitutes a statistically significant result and support of the alternative hypothesis.

p-value When performing a hypothesis test, the p-value gives the probability of data occurrence under the assumption that H0 is true. Small p-values are an indication of rare or unusual data from H0 , which in turn provides support that H0 is actually false (and thus support of the alternative hypothesis). In the example presented previously, if data is collected on cycle time and customer loyalty and a correlation test conducted, one might find a very small p-value, which would suggest that H0 is false (and thus that there is indeed a relationship between cycle time and customer loyalty).

Sample size Is the number of observations from a population that have been selected and used for calculating estimates and/or making inferences.

Significance level or Alpha α

Is the amount of evidence required to accept that an event is unlikely to have arisen by chance (and thus contradicts H0). This value is known as the significance level, Alpha, α. The traditional level of significance is 5% (0.05); however, this value can be made more lenient or strict as needed. For example, stricter values can be used for situations that demand stronger evidence before the alternative hypothesis is accepted (e.g., α = 1% (0.01)). In the previous example, if actions from a hypothesis test involving cycle time and customer loyalty could cost a company millions of dollars if wrong, then the test might be conducted at a stricter value to be “more certain” of the results. A value of 5% signifies that we need data that occurs less than 5% of the time from H0 (if H0 were indeed true) for us to doubt H0 and reject it as being true. In practice, this is often assessed by calculating a p- value; p-values less than alpha are indication that H0 is rejected and the alternative supported. When H0 is rejected, the term statistically significant is often used as an abbreviated “catch phrase” of the final result.

Variability Represents the consistency or spread of data referenced against a common point (typically the mean). This is often represented in terms of standard deviation, which acts like an “average distance from the mean.” Other common measures include Range, Interquartile Range, and Variance.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 14/41

Variable/Feature Selection Statistical modeling is an iterative process that can involve hundreds or thousands of independent variables. Clearly, with so many variables that can impact the dependent variable, simple, parsimonious models help reduce complexity and present models with nonsignificant predictors removed. Table 7.7 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab07) describes techniques that help an analyst reduce the number of independent variables to a select critical few.

Table 7.7 Variable/Feature Selection Techniques

Variable/Feature Selection

Description

Forward selection – Stepwise selection

A regression-based model-building method to select variables that are predictive. It starts with an empty model and continues to add variables with the highest correlation to the dependent variable one at a time. Only those variables that are statistically significant are included. This is often used to build models consisting of only statistically significant predictors, and attempts to build the model without the aid of the researcher. In most research, forward methods are preferred over backward elimination, and in general automatic model selection should be used sparingly.

Backward elimination

Similar in concept to forward selection, but the opposite. This technique starts with all the variables included and systematically removes statistically insignificant variables from the multiple regression model. This is often used to build models consisting of only statistically significant predictors, and attempts to build the model without the aid of the researcher.

Principal component analysis

technique used to reduce the number of variables that go into a model, principle component analysis (PCA) identifies those variables (known as the principal components) that account for the majority of the variance observed among the independent variables. This is often used as a variable reduction technique. Although predating factor analysis techniques, this technique is utilized less than the more robust factor analysis approach in modern practice.

Factor analysis A statistical procedure used to uncover relationships among many variables. This allows numerous intercorrelated variables to be condensed into fewer dimensions, called factors. This is often used as a variable reduction technique. For example, if a researcher has more than 100 variables of interest to study, a factor analysis might enable a composite metric to be created that would capture the essence of the 100 variables in a handful of composite measures. The use of factor analysis is often preferred in modern practice over a principal component analysis.

Modeling Algorithms and Techniques Table 7.8 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab08) describes some of the popular algorithms and techniques used in statistical and predictive analysis, though there are many others that are not listed.

Table 7.8 Modeling Algorithms and Techniques

Technique/Algorithm Description Key Statistic(s) Clustering/Segmentation A process that defines the allocation of observations, e.g., records in

a database, into homogeneous groups known as clusters. Objects within clusters are similar in some manner, while objects across clusters are dissimilar to each other.

Model Specifics: Distance

metrics for group structuring; R2

is also available.

Discriminant analysis Useful for (1) detecting variables that allow the greatest discrimination between different naturally occurring groups, and (2) classifying cases into different groups with “better than chance” accuracy (i.e., establishing group membership).

Overall Model:

Chi-square test.

Model Specifics: Correlations to define groups; Classification matrix that shows the accuracy hit rates for classification.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 15/41

Technique/Algorithm Description Key Statistic(s) Linear regression Uses one independent variable to predict a dependent variable. The

type of relationship is considered linear. This model can be adjusted for polynomial patterns. If other nonlinear relationships are of interest, other regression techniques would be used.

Overall Model: ANOVA test (F-

statistic) and R2 value.

Model Specifics: Regression coefficient (slope) and associated p-value.

Multiple linear regression

Uses multiple independent variables to predict a dependent variable. The type of relationship is considered linear. These models can be adjusted for polynomial patterns and interaction effects. If other nonlinear relationships are of interest, other regression techniques would be used.

Overall Model: ANOVA test (F-

statistic) and R2 value.

R2-adjusted sometimes useful.

Model Specifics: Regression coefficients (slopes) and associated p-value

Logistic regression Uses multiple independent variables to predict a binary, categorical dependent variable (e.g., yes/no). The type of relationship is considered nonlinear. The regression is based on likelihood of the binary outcome (presented in terms of log-odds, odds, or even probability of occurrence). Extensions of this technique can be used for categorical data with more than two values.

Overall Model: AIC, Chi-square (Hosmer & Lemeshow test).

Model Specifics: Regression coefficients (log-odds, which can be translated back into odds or probability) and associated p- values (based on the Wald Chi- square test).

Nonlinear regression Uses multiple independent variables to predict either a categorical or numerical dependent variable. This type of relationship is nonlinear and cannot be represented by simpler polynomial expressions. Thus, the shapes are other types of curved patterns (e.g., exponential, logarithmic, and so on). Traditional overall regression model statistics do not apply to such models.

Overall Model: SSE, MSE (a

“pseudo” R2 is possible).

Model Specifics: Regression coefficients (meaning varies depending on the algebraic structure of the nonlinear model) and associated p-values.

Decision trees Provide a treelike structure that models the segmentation of data into homogeneous subgroups. The decisions generate rules akin to if...then rules that classify/predict cases. Classification and Regression Trees (CART) are a type of decision tree technique that performs two-way splits on decisions. CHAID is a recursive decision tree technique (CHi-squared Automatic Interaction Detector). CHAID operates by performing Chi-square tests resulting in multiway splits on decisions.

Overall Model: Gini index; entropy measures.

Model Specifics: Tree diagram indicating rules for splits and classification.

Analysis of Variance (ANOVA)

1-way

A statistical method to compare two population means using one factor.

Overall Model: ANOVA test (F- statistic).

Model Specifics: Post-hoc tests or planned contrasts for group differences (along with associated p-values).

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 16/41

Technique/Algorithm Description Key Statistic(s) n-way ANOVA A statistical method to compare two or more population means

using two or more factors. Interaction (moderation) effects are possible to model in this.

Overall Model: ANOVA test (F- statistic).

Model Specifics: Interaction Effects (if applicable), Main Effects, and Post-hoc tests or planned contrasts for group differences (along with associated p-values).

k-Nearest Neighbor k-Nearest Neighbor (k-NN) is a classification algorithm that classifies single instances based on “similar“ records surrounding the item of interest. The number of records used to classify the observation is represented by k.

Overall Model: Euclidean distances used to establish “neighbors“ among the records in question.

Model Specifics: Rules to assign a class to the record to be classified (based on the established distances and classes of its neighbors).

Association analysis/market basket analysis

A method for discovering relationships between variables. Typically used to analyze transactional data in what is known as market basket analysis.

Measures of support and confidence are used to assess models.

Support refers to how often a rule applies: So to use an example of, those who buy apples also buy pears; support represents the number of times apples and pears show up in transactions relative to the total number of transactions.

Confidence refers to the reliability of the rule. So to use the previous example, confidence would measure the number of times apples occur with pears in transactions relative to transactions that contain apples.

Neural networks Nonlinear computational models whose operation and structuring are inspired by the human brain. Neural networks model complex relationships between inputs and outputs to find patterns in data.

Overall Model: Classification Matrix and Error Report.

Model Specifics: Weights that help detail the path model developed.

Time series Regression-like model that uses time as a predictor while controlling for dependencies among the time-based measures.

MAPE, MAE, MSE (see Table 7.9).

Time Series Forecasting Forecasting is an analysis technique that uses historical data to determine the direction of future trends. This type of analysis allows companies to have a long-term perspective on operations over time. Commonly used forecasting methods are presented in Table 7.9 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab09) .

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 17/41

Table 7.9 Time Series Forecasting

Statistic Description Time series Regression-like model that uses time as a predictor while controlling for dependencies among the timebased

measures. An ordered sequence of data recorded over periods of time (weekly, monthly, quarterly, and so on— e.g., the daily closing values of Microsoft stock on the NASDAQ).

Moving average

A technique that averages a number of past observations to forecast the short term. When graphed, it helps in displaying the trends in data that are cyclical and in smoothing out short-term fluctuations. This forecasting technique is considered a basic approach to controlling seasonal fluctuation.

Weighted moving average

simple moving average assigns the same weight to each observation in averaging, while a weighted moving average assigns different weights to each observation. For example, the most recent observation could receive a high weighting while older values receive decreasing weights. The sum of the weights must equal one.

Exponential smoothing

A method similar to moving averages except that more recent observations are given more weight. The weights decrease exponentially as the series goes farther back in time.

ARMA (autoregressive moving average)

A forecasting model of the persistence (autocorrelation) in a stationary time series. The model consists of two parts: an autoregressive (AR) part (which expresses a time series as a linear function of its past values) and a moving average (MA) part (which essentially is modeling/controlling for the noise/residuals in the model). ARMA models can be used to evaluate the possible importance of other variables to the system.

ARIMA (autoregressive integrated moving average)

An extension of the ARMA models that incorporate the element of nonstationary data. This technique helps in making any time series stationary by differencing; this step is especially important in forecasting because stationality is a necessary condition for all data modeling techniques. ARIMA techniques can be used to forecast nonstationary economic variables such as interest rates, GDP, capacity utilization forecasts for call center resource planning, and so on.

ARIMAX models

ARIMAX models are an extension of the traditional ARIMA model, which allows for the incorporation of several explanatory variables into the model design often as covariates. This helps in understanding and quantifying the effect each of the explanatory variables has on the dependent variable. Practically, inclusion of covariates can help in bringing some additional business aspects into the model (e.g., promotional data in a sales forecast).

Model Fit and Comparison Statistics Once a model has been executed, how do you measure overall model performance? Table 7.10 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07tab10) describes the many statistics that enable an analyst to assess model fit and comparative performance; it also indicates whether you should look out for a high or low value. Again, this list is by no means exhaustive.

Table 7.10 Model Fit and Comparison Statistics

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 18/41

Notes 1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end01a) . American Statistical Association,

http://www.amstat.org/careers/whatisstatistics.cfm (http://www.amstat.org/careers/whatisstatistics.cfm) .

2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end02a) . Data Mining Methodology (August 2007), KDnuggets, Polls, http://www.kdnuggets.com/polls/2007/data_mining_methodology.htm (http://www.kdnuggets.com/polls/2007/data_mining_methodology.htm) ; Data Mining Methodology (April 2004), KDnuggets, Polls, http://www.kdnuggets.com/polls/2004/data_mining_methodology.htm (http://www.kdnuggets.com/polls/2004/data_mining_methodology.htm) “What Main Method Are You Using for Data Mining (July 2002), KDnuggets, Polls, http://www.kdnuggets.com/polls/2002/methodology.htm (http://www.kdnuggets.com/polls/2002/methodology.htm) .

3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end03a) . “Six Sigma,” Wikipedia, http://en.wikipedia.org/wiki/Six_Sigma (http://en.wikipedia.org/wiki/Six_Sigma) .

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 19/41

4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end04a) . “What Is Six Sigma?” Six Sigma.com, http://www.isixsigma.com/new-to-six-sigma/getting-started/what-six-sigma/ (http://www.isixsigma.com/new-to-six-

sigma/getting-started/what-six-sigma/) .

5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch07#ch07end05a) . “Causality,” Wikipedia, http://en.wikipedia.org/wiki/Causality (http://en.wikipedia.org/wiki/Causality) .

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 20/41

8 Business Value of Health Analytics

Patricia Saporito

The business value of healthcare analytics, and how to determine it, have been a top priority for those who espouse the importance of analytics. According to a study by Nucleus Research, there is an annual revenue increase possibility if the median Fortune 1000 business increased its usability of data by just 10%, translating to a $2 billion opportunity. The same study states that there is a 1000% return on investment (ROI) for organizations that make analytic investments. The spotlight on analytics continues to shine especially in light of increased expectations driven by the Big Data momentum and its least touted “V”— value.

Analytics in healthcare have been providing value but have largely been limited to siloed application areas, for example, clinical improvement, productivity management, and revenue cycle management. The value of analytics within each of these “silos” is often visible and mature, but their impact and contribution to the business are relatively immature from an enterprise perspective due to lack of an integrated enterprise approach. Therefore, integrating analytics across these domains and understanding the value leverage and correlations between them is still a rich area of opportunity. Organizations can realize this greater value by adopting a value management approach. However, culture and change management will be two significant hurdles to overcome as business users will naturally focus on their own departmental interests. These hurdles can be addressed by incorporating value management best practices and institutionalizing them in Business Intelligence Competency Centers.

Business Challenges Healthcare has huge challenges especially in the United States. Healthcare costs continue to rise, and healthcare organizations continue developing strategies to address them. The Realization of Patient Protection and Affordability Care Act’s (a.k.a. Obamacare) promise of containing costs is far off, and in the short term is creating turmoil in the market as providers, payers, and pharmaceutical companies review and revamp their business models to comply. Regardless of these changes, organizations must take a value-driven approach to analytics, one that states the value of analytics in the terms of the business or face the lack of funding. As healthcare organizations face competing requests for limited resources, investments in analytics are often overridden in favor of operational or transactional systems such as electronic medical records (EMRs), computerized physician order entry (CPOEs), and the like. Even when analytics are addressed, all too often they are operational analytics embedded within these operational systems such as a report of overdue accounts receivables. Analytics that span these transactional systems and can bring greater enterprise value get short shrift. Therefore it is critical that organizations have an analytic strategy that is business driven, is strategically aligned and shows value. This keeps the focus on choosing the right projects, clearly driving ownership and accountability for business results and delivery on commitments for these results. Communicating successful projects with proven value will ensure ongoing funding. Unfortunately, most organizations have not demonstrated compelling value propositions so far.

Value Life Cycle Business owners need to make their needs known to their information technology (IT) partners and tie them to their business imperatives to get their share of funding and so that IT gets funding as well. This starts during the annual planning and budgeting process when the business submits requests for resources. Any significant project will require a business case to obtain resources and for prioritization. Following successful implementation of a project, the business and IT should also engage in a value realization analysis that demonstrates the actual value achieved from the analytic project. Organizations can use the value realization not only to communicate successful results but also for “cascading” funding to support future initiatives.

A value-based management approach to this value life cycle contains three phases: discovery—investing for impact; realization —delivering for business outcome; and optimization—governing for ongoing performance (see Figure 8.1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch08#ch08fig01) ):

• Value discovery develops a value-based business strategy and business case, enabled by technology and aligned with corporate objectives. It answers these questions: What are your business imperatives and how do you align your business and IT strategy? What is the expected impact of addressing these imperatives (business case)? What are the right initiatives to address the value creation opportunities, and how are they prioritized?

• Value realization develops transformational strategies to mobilize, deliver, and measure business results based on insights into leading practices or benchmarks. It answers these questions: How should the business prioritize, mobilize,

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 21/41

and govern programs to deliver value? How can the business case be made actionable at the operating level, and how is value measured? How should the business govern, architect, deploy, and ensure quality of master data? How should the business govern, architect, and set up an information framework for business analytics?

• Value optimization assesses how the implementation and program compares to best or leading practices and recommends areas where the business can drive more value from current investments. Key questions addressed in this phase are: What is the value realized by your program and how can the business derive more value from existing investments? What insights do you have regarding the total cost of ownership? How does the implementation compare to best or leading practices? How can we utilize a periodic business process “health check” to continuously measure our progress versus our peers or analytic leaders? How can the business mobilize and govern a program to optimize success? What services make sense in a shared service center, and what is the right approach to setting up these shared services?

Figure 8.1 Value life cycle (Source: SAP)

More than 10,000 companies, both SAP customers and noncustomers, have participated in more than 30 business process benchmark surveys with almost 600 participating in Business Intelligence specific surveys. Based on this population, SAP has found that high adopters of this value management based approach achieve significantly more value than low adopters. These companies deliver twice as many projects on time and on budget, and show more than 1.5 times greater value than low adopters, regardless of industry (see Figure 8.2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch08#ch08fig02) ).

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 22/41

Figure 8.2 Low versus high value management adopters comparison

Healthcare Analytics Value Framework: Key Drivers There are four key business drivers, or levers, to increase value in healthcare:

• Revenue growth:

• Volume metrics measure new patient acquisition, as well as existing patient retention and growth.

• Price metrics include increased prices per service as well as improved profitability per service and per patient.

• Operating margin:

• General and administrative expenses metrics measure improving patient interaction efficiency and improving administrative service efficiency especially in HR and IT.

• Cost of care metrics include improving care service development and overall delivery efficiency

• Asset efficiency:

• Property plant and equipment (PP&E) metrics focus on improving PP&E efficiency such as bed occupancy and/or room utilization.

• Inventory metrics measure improving inventory efficiency including supplier management and leakage.

• Receivables and payables management metrics improve receivables and payables efficiency, for example, percent of bad debt, percent of late charges, percent of A/R unbilled, and so on.

• Organizational effectiveness:

• Organizational strengths metrics focus on improving management and governance effectiveness (e.g., business planning, business performance management) and improving execution capabilities (e.g., operational excellence, agility and flexibility, strategic assets).

• External factors metrics include key macro- or microeconomic key performance indicators (KPIs) such as unemployment rate.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 23/41

These levers impact three main business goals or areas: clinical performance, operational performance, and financial performance as shown in Figure 8.3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch08#ch08fig03) . Each of these performance areas has several key business objectives:

• Clinical performance:

• Improved clinical quality of care

• Improved patient safety and reduced medical errors

• Improved wellness and disease management

• Improved patient acquisition, satisfaction, and retention

• Improved provider network management

• Operational performance:

• Reduced operational costs

• Increased operational effectiveness and efficiency

• Reduced inventory leakage

• Improved provider pay for performance accountability

• Increased operational speed and agility

• Financial performance:

• Improved revenue

• Improved ROI

• Improved utilization

• Optimized supply chain and HR costs

• Improved risk management and regulatory compliance/reduced fines

• Reduced fraud or abuse leakage

Figure 8.3 Healthcare performance management objectives (Source: SAP)

Validating Actionability and Measuring Performance Improvement with Key Performance Indicators (KPIs) An effective way to develop a business case and also to validate analytic actionability and measurement is by analyzing and “diagramming” goals and objectives by asking the following questions:

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 24/41

• What business questions need to be answered?

• If those questions can be answered, what actions can be taken?

• Which KPIs need to be monitored to “move the needle” for improvement? What leading indicators should also be looked at for intervention before the resultant KPI? Are there any key correlation metrics you should also look at?

• What data are needed to support these objectives? How readily available are they? Where do those data reside; that is, what data sources do you need?

Figure 8.4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch08#ch08fig04) illustrates a framework used for working with a leading academic hospital system to define dashboards and reports to help manage nurse productivity and related analytic objectives.

Figure 8.4 Objective actionability matrix framework applied to nursing productivity (Source: SAP)

Labor is one of the largest costs in healthcare. It is also one of the most critical resources in care delivery. Healthcare organizations face serious shortages of healthcare workers. Vacancy rate for nurses, imaging/radiology technicians, and pharmacists are over 10%; one in seven hospitals reports more than 20% of RN positions are vacant, with some reporting as high as 60%. Nursing productivity is a key analytic area that impacts all three performance areas—clinical, operational, and financial.

The organization already had a staff scheduling system in place that had some analytics, but it did not have an effective capability to integrate and analyze data across admissions, staff scheduling, human resources, and finance for use by the nursing unit managers. They originally identified about 20 KPIs they wanted to monitor; after our analysis we defined 30 metrics and expanded the scope of the data to be integrated.

As part of the business design process, we defined and validated five core objectives for Nursing Productivity:

• Enable compliance with staffing guidelines.

• Ensure patient safety.

• Improve the Human Resource life cycle (e.g., hiring, retention, promotion).

• Meet patient safety compliance guidelines.

• Attain revenue goals.

By using the dashboards and reports developed for this project the nursing unit managers were able to better anticipate and manage their staffing needs and costs. A “what-if” component allowed them to see the impact of one change on other areas—for example, the staff budget impact if they used more or less outside staff or increased permanent staff hours. As a result the healthcare organization elected to adopt a primary strategy of increased use of agency staff to reduce permanent staff overtime and burnout, resulting in reduced vacancy rates and related recruiting costs. They also saw an improvement in patient satisfaction as the permanent staff was not as stressed. All of these savings dropped to the bottom line. Without these expanded analytic capabilities to look across areas, they would have kept to stovepiped staff scheduling and staff budget analyses, a

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 25/41

myopic approach, which was leading to increased permanent staff overtime and budget overruns, staff injuries, staff burnout and turnover, and decreased patient satisfaction.

Business Intelligence (BI) Performance Benchmarks for IT In addition to revenue and expense impacts on the business, healthcare organizations also want to measure and improve their BI IT capabilities and overall performance to support the business. Sometimes IT attempts to justify analytic investments based solely on reducing Total Cost of Ownership (TCO); while TCO is important to reduce IT costs, it’s only part of the value picture. The business case for investment must include business revenue and expense components, as well as TCO.

Once analytic investments have been approved and made, benchmarking is an effective way to measure performance both against internal best performance, and against external best in class in healthcare and across all industries.

Specific BI related objectives and related KPIs as part of an effective BI performance analytics program include

• Effectiveness:

• Usage of BI by employees, executives, and external stakeholders (BI Adoption)

• Level of insights generated through BI usage by business process (BI maturity)

• Efficiency:

• BI project mix (e.g., reports, dashboards, semantic views)

• Cycle times, reliability, and uptime

• BI costs

• BI technology

• BI technology leveraged for analytics, data warehousing, and data management

• Organization:

• BI organization model

• Level of centralization

• Size of support organization

• Best practices adoption

• Importance of best practices

• Current coverage of best practices

• Importance and coverage gap

Figure 8.5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch08#ch08fig05) shows a subset of metrics from the Americas SAP User Group (ASUG)-SAP BI Benchmark Survey, comparing a fictitious company against a benchmark peer group.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 26/41

Figure 8.5 Sample ASUG benchmark report (Source: ASUG, SAP)

Using BI Competency Centers to Institutionalize Value BI Competency Centers (BICCs) can ensure that value management is part of every project and institutionalize it in an organization’s overall BI program. Organizations have been forming BICCs (or Centers of Excellence—CoEs) to leverage best practices and improve operational effectiveness and efficiency to ensure business user satisfaction and demonstrate BI value. BICCs can report to IT or to business areas, often operations or administration; it is too early to see whether they are moving under chief analytic officers.

Regardless of where they report, BICCs play a key role in defining and executing an organization’s BI strategy especially in demonstrating and communicating the value of analytics. These units have responsibility for several key areas of the strategy, including the development, documentation, and communication of the organization’s overall BI strategy; development of the business requirements for a project and prioritizing all projects; defining the business case for each project and identifying supporting capabilities; defining an information taxonomy, architecture, and managing the technology tools; and finally managing the governance, program management, roadmap, measurement, training, and support (see Figure 8.6 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch08#ch08fig06) ).

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 27/41

Figure 8.6 BI strategy framework (Source: SAP)

Two key emerging roles that support value management are KPI analyst and value analyst. The KPI analyst helps define key performance indicators, leading indicators, and correlation metrics in the design phase. The value analyst helps define the anticipated ROI for the business case and validates attained ROI in the post-implementation value assessment. In smaller organizations these functions are often filled by the business analyst role, which can reside in the BICC or in business units. In larger organizations a value analyst may be part of a performance management function in finance or strategy areas. Both play a key role in the business benefit area of the BI strategy framework.

A third role that can contribute to value management is the communications analyst; this developing role can help broadcast analytic successes including quantitative value. Communications analysts are often part of the education and training function with the organization area of the BI strategy framework. They often use virtual communities of interest, or practice, to promote successes and increase employees’ analytic engagement and adoption.

Business Performance Based Approach to Value Using a value-based approach is a shift from a simple “on time and on budget” philosophy to a business outcomes and value- based one. Many companies focus on reducing IT and administrative costs, which are important but have far less impact on the business than revenue opportunities. Analytics-driven companies adopt a broader performance management approach and develop the following strategies and accountabilities:

• They demand that IT investments deliver competitive advantage.

• They focus change on high-impact results.

• They set ambitious business goals enabled by ambitious talent.

• They drive accountability through measurement.

• They embrace a culture of performance obsession.

We have found the following best practices are used to actualize value management:

• Performance improvement—They consistently and proactively measure and compare themselves internally and externally to identify new opportunities to gain more value from business processes.

• Justification—They have a formal process to justify investments that involves stakeholders across the business.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 28/41

• Value realization—During project implementations, they design the solution to realize the value identified in the business case. They develop a key performance improvement framework, complete with clearly identified ownership and accountability.

• Business strategy-IT alignment—Business process leads and IT professionals collaborate on everything from strategic planning to program execution to ensure that their objectives are well aligned.

• Governance and portfolio management—Executive leadership is engaged to help ensure project success.

• Organizational excellence—Programs are staffed and managed by a talented team that is measured based on the quantifiable business value its programs achieve.

Additional Resources ASUG-SAP Benchmarking Forum is a facility to exchange metrics and best practices. It covers 26 business processes with 2,600 participants over 1,400 companies including heath care organizations. The BI survey has been conducted since 2007. Participation is free and open to both SAP and non-SAP customers. See http://www.asug.com/benchmarking (http://www.asug.com/benchmarking) .

Deloitte Enterprise Value Map is a one-page tool that shows the relationship between shareholder value and business operations. A healthcare-specific enterprise value map is also available. Search on “Enterprise Value Maps” at www.deloitte.com (http://www.deloitte.com) .

Global Environmental Management Initiative (GEMI) Metrics Navigator™ is a tool to help organizations develop and implement metrics that provide insight into complex issues, support business strategy, and contribute to business success. The tool presents a six-step process to select, implement, and evaluate a set of “critical few” metrics that focus on an organization’s success. It includes worksheets, series of questions, or checklists for each step. (Note: While GEMI is focused on environmental sustainability, the metrics tools are excellent and apply across industries and processes.) Go to www.gemi.org/metricsnavigator/ (http://www.gemi.org/metricsnavigator/) .

HIMSS Analytics and the International Institute for Analytics (IIA) have partnered to create the DELTA-Powered (Healthcare) Analytics Assessment, a new maturity model that assesses and scores the analytical capabilities of healthcare organizations. This survey and certification program provides a roadmap to gauge how organizations leverage data and analytics. Go to http://www.himssanalytics.org/emram/delta.aspx (http://www.himssanalytics.org/emram/delta.aspx) .

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 29/41

9 Security, Privacy, and Risk Analytics in Healthcare

Thomas H. Davenport

Information security, risk management, and privacy have always been hot topics across healthcare due to the high degree of

government and industry regulation,1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end01) the sensitive medical information that is the industry’s lifeblood, and the complexity introduced by significant outsourcing. While traditional information security software and hardware were designed to protect the huge data volumes and transactions required of deep analytics across payers, providers, and life science organizations, forward-thinking healthcare enterprises are more focused on redefining their approaches to identity protection and privacy to guarantee that when the use of analytics generates new opportunities, they can be put into action.

At the same time, highly public data breaches, stricter reporting requirements, and meaningful use reimbursement are forcing healthcare organizations to manage security and privacy in a dynamic and increasingly predictive manner. Analytics holds real promise in transforming how organizations both establish and report their security posture to key stakeholders, and could be the missing link to allowing organizations to market their risk posture as a competitive differentiator. Similarly, applying analytics should streamline compliance audits saving time and money. The applications with the most promise for IT and information security (“infosec”) teams should focus on predictive analytics, correlation of physical and digital events, risk dashboarding, and assigning monetary value to data breaches.

This chapter first presents the kinds of adjustments leading payers, providers, and life sciences organizations are making to their information security and privacy practices to prepare for the expected explosion in the use of analytics across the enterprise. Second, the chapter details how healthcare organizations are using risk analytics to add a new dimension to their existing security practices, and in the process making their information risk posture more transparent to key stakeholders and regulators.

Securing Analytics Information risk management at the atomic data level has been in place across payers, providers, and life sciences companies to varying degrees for years. Perimeter security (such as firewalls, intrusion detection systems (IDS), and intrusion prevention systems (IPS)), encryption, and data leak protection are common tools used to control the movement and use of sensitive information inside virtually every healthcare enterprise. Even the most basic security infrastructures are designed to scale to handle larger databases and transaction volumes, along with new transaction types, and adding analytical processes should not compromise the technological veracity of the common security infrastructure.

Instead, as the role of analytics is to generate new insights (information sets) from existing public and private data sets, leading organizations are reconsidering the current concepts of permission and privacy, differently within each sector. The age old practice of simply anonymizing input data sets will not satisfy the ever-changing customer privacy and government regulations:

We’ve spent a lot of money on information security over the years, and feel pretty good on that front. It’s really the legal and compliance side that has the hospital leadership most concerned. The idea of permission has to be completely rethought now if we’re going to be able to use any of the results of the analysis we’re now able to run. CISO, NY-based regional clinic.

Providers Primary Data Types: Patient data (patient medical records, PII), clinical research data, financial and operational data

The historical motivators for information security and privacy protections inside provider organizations have been financial liability stemming from exposure of personally identifiable information (PII), and regulatory compliance. Among the three healthcare sectors, providers have the most difficult time with information security due to the different flavors of databases, desktops, mobile devices, diagnostic devices, and personnel (doctors, nurses). In addition, the high number of outsourced service partners the average provider organization shares data with inevitably leads to unintended information exposure.

Inadvertent data leaks have become the norm as permissioned access to data sets by insiders and outsourced service providers have become common. Cleanup costs of data leaks have grown considerably to include lost business to reputational damage, identity monitoring services for the affected patients, and forensic investigation costs to determine the exact source of the leak.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 30/41

While less common, malicious data breaches are certainly of more concern to most providers, and the costs can be considerably higher depending on the intent of the data thief.

While these issues may be exacerbated by heavier use of analytics, patient privacy will be the area where provider organizations will have to be sure that actionable insights can be used without the objection of patients. Traditionally, information security has been the domain of the IT staff. As new analytical techniques use more PII to generate more powerful insights, deciding what levels of permission must be secured to use those insights freely will need to involve the legal department more than IT. Providers who intend to become analytical competitors will need to shift their focus from securing their data to seeking permission to use the insights generated from those data.

Guidance: If analytical results are shared beyond normal boundaries (for example, partners, suppliers, etc.) and identity can be detected, the use of those results needs to be reviewed before taking action.

Trying to lock down all the tricky ways our medical staff use data and analytics to deliver better care is a losing battle for IT, and not their job. Instead, we initiated a dialogue with IT to be sure we don’t get blindsided by a privacy suit. A great example is a recent effort to identify our most expensive patients over a long period of time, and then try to proactively work with those patients. Without proper permission from the patients to use their medical records for that purpose, combined with the financial information we have about them, we might run into a problem.” Legal Counsel, Cambridge, Massachusetts-based hospital

Payers Data: Patient data (patient medical records, PII, patient financial records), actuarial and rate information

Similar to provider organizations, payer organizations are built on collecting and making rate and coverage decisions using PII that allows them to predict insurance risks and set premiums accordingly. While analytics are mostly being applied at an aggregated data level, atomized data will be more and more accessible with analytics, making information security and privacy as critical as ever.

Similar to the provider community, leading payer organizations are involving legal teams more than IT teams as their ability to take action hinges on satisfying privacy rights among the insured community as well as to regulators. The bar for Health Insurance Portability and Accountability Act (HIPAA) compliance in particular will surely go up if consumers complain that new decisioning techniques can better assess an individual’s risk, producing higher premiums for example. Claims of privacy violations could be a potent lever consumer groups could use to slow more efficient decisioning and classification.

Guidance: Negative reactions by consumers to rate increases and uncertainty around the ultimate outcome of the federal healthcare legislation are putting pressure on providers to price both existing and new policies correctly. Scenarios could be imagined where consumer health advocates could claim privacy over patient data used to drive toward rate increases or policy changes.

Life Sciences Data: High-value research and development (R&D—e.g., drug discovery) data, clinical trial results

Unlike providers or payers, pharmaceutical and biotech organizations spend all of their information security resources on protecting highly proprietary research and development information, including clinical trial results, testing results, and drug formation results. Like providers, life sciences organizations are home to a variety of medical and research personalities, each of which has its own data recording and data storage cabinets. So, the focus tends to be on data leakage (securing the perimeter) and database security (access, encryption).

With analytics come more calls on large databases by numerous internal and external parties, which somewhat increases the security risk. But, even more than with providers, the security infrastructure inside most life science companies was built with scale as the number one priority. Since clinical trial data are largely aggregated and anonymized from the beginning, privacy is not an issue faced by life sciences organizations.

Guidance: Privacy is less of an issue here. Analytics will quicken the pace of proprietary information creation, and the information security infrastructure should grow to match that. Most often inside life sciences organizations, success is just having an adequate data inventory, and knowing where new proprietary information is being stored.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 31/41

Risk Analytics Most healthcare information security programs target the basics: internal awareness programs, desktop security, network perimeter protections, data leak prevention, authentication, and database security. Although mundane, leading providers are starting to package an aggregation of the basics into a security posture score and market this as a competitive differentiator. Patients like to know that their healthcare provider or insurer can reliably safeguard their information. And CEOs and board- level advisors like to know how much information risk the enterprise faces. Leading organizations are then calculating the return on investment in risk analytics as both internal (better security intelligence to predict threats and avoid costs) and external (marketing information safety results).

We’re right in the middle of trying to get budget to do some pretty cool things in terms of predicting when bad things will happen, mostly just on the inside from non-malicious users. We have the log files, and can react in near real-time, but we don’t have the analytical tools to go to the next step. A by-product of this should also be a red-yellow-green type of indicator for the senior folks to know what the state of security is whenever they need it. CIO, Massachusetts- based multihospital group

In addition, while regulatory compliance takes constant attention (and always will), new acceptable use standards require security audits and tests to receive acceptable use reimbursement. Like PCI compliance among retail merchants, there now exists a real monetary incentive to continue to improve information security.

Predictive analytics,2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end02) correlation between physical and digital events, risk dashboarding, and valuing data breaches are the key areas where leading organizations are effectively applying analytics to information security today.

Predictive analytics is being applied to information security in isolated instances primarily to anticipate likely times when data breaches or data loss may occur. Generally, historical event patterns can help organizations project when internal processes are most likely to generate inadvertent exposure or loss. Another major area of focus is data exchanges with service provider partners. With enough historical data in hand, risk management teams can predict when a particular incident may be most likely to occur.

Correlating physical and digital events has been a focus of the security information management (SIM) vendors for years, but using advanced analytics to correlate more and more physical logs against digital events will continue. A big area of focus in many healthcare organizations is the intersection between the physical security and digital security groups. SIM software can now integrate and correlate, for example, real-time video surveillance records, network and database activities, and human resource records to make real-time incident response decisions.

Risk dashboards are maturing from the standard green-yellow-red approach to sophisticated pictures of the overall risk posture of an enterprise, complete with financial impact information that allows senior management to prioritize responses to significant information breaches and make proactive investments.

Calculating the monetary value of information security breaches will be a big domain of analytics. Information security has consistently lacked the capability to prioritize investment and responses because the value of the information assets being managed has been so difficult to estimate. With analytics, IT groups should be able to marry financial (value) data with breach information to be able to make more intelligent real-time decisions.

Compliance and Acceptable Use HIPAA compliance garners the majority of the attention when it comes to regulatory compliance among healthcare providers,

followed by Red Flag rules and the Health Information Technology for Economic and Clinical Health Act (HITECH).3

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end03) Actual regulatory penalties suffered due to noncompliance are uncommon, but the perception of regulatory liability drives much of the spending in information security systems today. Will the application of analytics change this? Probably not. But, the leading organizations are certainly seeing an immediate payback to investment in risk analytics in the form of faster and more efficient reporting at lower cost.

Now, a renewed focus on data security has emerged due to its role in meaningful use reimbursement.4

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end04) The rather vague requirements to conduct or review security and implement security updates to qualify for meaningful use reimbursement should generate a short-term bump in

security spending to meet reimbursement requirements.5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end05)

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 32/41

Look to the Future This chapter described how the growth of analytics is forcing all healthcare organizations to reconsider information security and privacy. Generally, existing information security infrastructures were built with scale in mind, and increased analytical activity should not affect an organization’s security posture. Among providers and payers who are seeing the value of analytics, however, much closer attention is being paid to data ownership and privacy. Legal teams, rather than IT teams, will need to spend time understanding how to interpret and protect privacy when analytics create actionable opportunities.

While the biggest returns on analytics will never come in the area of information security, using analytics to improve an organization’s risk reporting and overall understanding of risk posture will become a competitive differentiator in an era of increased regulatory requirements and incentives.

It is clear that the promise of analytics in healthcare could be handicapped without proper attention to security and privacy going forward. On the flip side, customers of all three healthcare groups expect intelligent security safeguarding their personal information that doesn’t hinder their doctors, insurers, and pharmaceutical providers from caring for them quickly, efficiently, and cost-effectively. Analytics can help relieve this tension.

Notes 1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end01a) . Examples are Health Insurance

Portability and Accountability Act (HIPAA), industry-based Electronic Health Records (EHR) standards, Red Flags Rule, and Health Information Technology for Economic and Clinical Health Act (HITECH).

2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end02a) . For a complete review of data types and correlation techniques, see “Security Analytics Project: Alternatives in Analysis,” by Mark Ryan, Secure-DNA.

3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end03a) . The Health Information Technology for Economic and Clinical Health Act (HITECH) is a new federal privacy and security mandate regarding patient information, including mandatory notification of individuals whose information is breached, that was included as part of the American Recovery and Reinvestment Act of 2009 (ARRA), signed into law by President Obama February 17, 2009. A major change is that the new legislation generally requires covered entities and business associates to disclose to their patients any breach involving a patient’s protected health information (PHI).

4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end04a) . See E-health Records: Getting Started with Meaningful Use by Anthony Guerra for a summary of acceptable use reimbursement requirements related to data security systems.

5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch09#ch09end05a) . See 2010 HIMSS Security Survey, page 3, for a review and evaluation of risk assessment requirements in connection of with meaningful use reimbursement.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 33/41

10 The Birds and the Bees of Analytics: The Benefits of Cross-Pollination Across Industries

Dwight McNeill

Healthcare can learn a lot from other industries. Industries develop deep and unique strengths in certain areas and get proficient in the associated analytics. Other industries that do not experience the same set of forces do not develop these analytics capabilities. They are blinded from them and their potential performance is constrained. This chapter provides a

guided tour of industries that are somewhat mysterious to healthcare, including retail, banking, politics, and sports.1

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end01) The express purpose is to harvest some ideas and build a bridge to adapt them in healthcare. Indeed, the best analytics from these industries provide insights to address some of the most intractable challenges in healthcare.

This chapter is about discovering ideas in faraway places and building bridges to welcome them home. It addresses four areas:

1. Why analytics innovations matter

2. The discovery process to harvest analytics sweet spots from the four industries

3. The distillation and interpretation of the sweet spots into the most compelling healthcare analytics adaptations

4. A roadmap for putting ideas into action and a model to evaluate the adoptability of the healthcare adaptations

Why Analytics Innovations Matter Analytics in healthcare is a paradox. On the one hand, healthcare is immersed in analytics. It is far ahead of other industries in using science, for example, to understand diseases and develop new cures and treatments. But there is a significant voltage drop between the science and its application in practice. U. S. healthcare faces major challenges, not the least of which are its low standing in the world on key health outcomes, efficiency, cost, disparities, and affordability. And research shows that the odds of

getting the right medical service are just a bit higher than 50/50.2

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end02)

The fact is that we know how to address these challenges, and analytics can be a tremendous support; but there’s a blockage. This is due to a number of factors, including the seemingly overarching prerequisite to digitize the business before anything else can be harvested from analytics. This forestalls other forms of analytics that can lead to real benefits for business today. But it is more complicated than that. There is an amalgam of contributing barriers, including the lack of good coordination among actors in the ecosystem, a perverse payment system that does not reward value, complex products and services, ambiguity about who the customer is, professional autonomy, convoluted market dynamics, multiple vested and powerful interests, and a pervasive, risk-averse culture.

In spite of these challenges, the field must innovate to make change happen. Innovation is critical to the success of any business. A recent MIT survey of 3,000 executives across many industries and countries found that the top business challenge is

“innovating to achieve competitive differentiation.”3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end03) This was far ahead of the usual business challenges to grow revenue and reduce costs. Innovation rises to the top at this time because the Great Recession required businesses to address all possibilities for cost reduction, and many have developed lean organizations. Businesses recognize that they need to grow the top line, and not just by extending the old line. And to grow the top line, they have to transform the business.

This transformation imperative is especially true in healthcare. For example, health insurers face commodity prices for premiums, the demand for transparency, and a flip of the business model from business-to-business to business-to-consumer, among other pressures related to healthcare reform. Similarly, providers are facing market and government pressures to improve outcomes, lower waste, and change the underlying revenue model from fee-for-service to global payments. In response, the leading companies are dramatically changing their identities. For example, some health insurers are becoming health companies, and others are viewing claims processing as one of many product lines as they become information companies. The industry archetypes are eroding, and innovation is charting a new path.

Innovation starts with fresh ideas. Ideas matter. They are the seeds from which innovations grow. Innovating to achieve competitive differentiation is the top business challenge today. And analytics is the high-octane fuel to power innovations to

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 34/41

achieve business breakthroughs.

Ideas can come from comfortable places and unfamiliar places. This chapter concentrates on the latter, not because they are better, but because they are often ignored. They are ignored because they do not necessarily fit our beliefs about how the world works. We tend to seek out information that confirms our positions and ignore the rest, what is referred to as confirmation bias. So outside-in thinking has an inherent hurdle at the outset. But ignoring outside-in thinking can lead to blind spots. For example, legions of dedicated professionals concentrate on the problems within healthcare with the intent to improve them. The blind spot of this process might be characterized this way: “Removing the faults in a stage-coach may produce a perfect stage-

coach, but it is unlikely to produce the first motor car.”4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end04)

Discovery of Healthcare Adaptations The process for discovering ideas from other industries is depicted in Figure 10.1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10fig01) . Note that this approach is based on the firmly held belief that analytics succeeds when it responds to the needs of the business and not when analytics answers—that is, technologies and methodologies—are in search of business questions.

Figure 10.1 Healthcare adaptations discovery process

Industry Challenges

The first stage in the discovery process is industry challenges. This stage seeks to understand the industry, goals, context, challenges, and drivers. These ingredients mix together in a crucible that determines how an industry needs to perform to succeed by developing and honing areas of strength. For example, the banking industry suffered a disastrous plunge in trust and revenues due to its irresponsible subprime mortgage lending practices. This not only contributed to the Great Recession, which cut demand for its products, but also led to onerous regulations and oversight that constrained the business, especially in the area of lending. It subsequently developed greater strengths in risk assessment.

The industry challenges are unique to each industry and are summarized later in this chapter. However, there are challenges common to all the industries that are particularly acute at this time and go beyond generic business challenges, such as increasing revenues, reducing costs, improving the balance sheet, paying attention to customers, and so forth.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 35/41

• Great Recession—This was not a typical, cyclical recession. In fact, it was the worst economic period since the Great Depression. It shattered customer purchasing power and the demand for products and services.

• Hypercompetition—This usual feature of business was put into high gear due to the Great Recession and the need to capture dwindling purchasing dollars. It makes demands on analytics to gather and analyze more diverse and larger volumes of data to know the customer more fully. This is most evident in the retail industry, where personal data can help predict what a customer wants to buy before the purchase.

• Customer power—Customers are taking the upper hand in their relationships with business because they have the data at their fingertips, in their smartphones and tablets, and they use them to make more informed purchasing decisions. The whole process of buying has been accelerated, yet made more deliberate, because of the availability of price information from multiple sellers on the Internet. Similarly, customers want to do business with companies “their way,” on their preferred devices, with access to services 24/7, and expect the business to provide flawless service across channels. Finally, customers want information that makes sense to them and on their terms, which is often obtained from peers and not from marketing departments or government agencies.

• Transformation—The business race continues to have no finish line. But the difference is that the very nature of the business has to evolve and the pace of change is accelerating. The tried-and true ways of succeeding in banking, retail, healthcare, and even sports are up for grabs. The need for transformation creates the appetite for innovation.

• Clicks—Clicks are the sounds of doing business on computing devices. Clicks are challenging the bricks. What could be more indicative of shifting paradigms than the collapse of the structures in which people do business (bricks)? Mobile “rules” for now because it is seemingly a new organ of the body that offers new functionality, is integrated perfectly, and is appreciated. It is almost like a seventh sense for humans.

Industry Strengths

The second stage of the discovery process is the understanding of industry strengths. All industries have unique strengths that, if exploited, can drive business breakthroughs to beat the competition. The unique industry strengths, distilled to their essence, are summarized here:

• For retail, it is clear that the overarching strength of the industry is to acquire, retain, and optimize customers. Marketing’s customer analytics are woven into this strength.

• For banking, the unique industry strength is understanding and minimizing risk. As mentioned earlier, the industry needed to clean up its lending practices for its very survival. Risk assessment has always been a core element of the industry, but it needed to get a lot better to reduce loan defaults without crimping the volume of loans. This became an industry imperative that necessitated the refinement of the industry strength.

• For political campaigns, it is the laser focus on finding, energizing, and persuading voters. This strength was made more powerful through data-driven innovations.

• For sports, it is its ability to engage fans through the full transparency of detailed performance data on its athletes and teams. Athletes are acquired, fired, and improved based on the data. The paradox for healthcare is that sports fans know much more about the athletes who entertain them than patients do about the doctors who make life-and-death decisions about them.

Analytics Sweet Spots

The third stage of the discovery process is identifying the analytics sweet spots that correspond to the industry strengths. The concept of a sweet spot comes from sports. It is the place where a combination of factors results in the maximum impact achieved relative to a given amount of effort. In baseball, for example, it is that place on the bat that produces the most powerful hit. The analogue in analytics is a solution that provides the most power to make the industry strength as successful as possible. In the case of banking, the analytics sweet spot, matched to the industry strength, is a refined FICO score that assesses the capability of a borrower to fulfill promises to repay a loan.

Each industry pushes the envelope in its use of specific analytics, not necessarily because it has more technical sophistication, but because the specific goals, purposes, pressures, and culture of that industry are unique and require better flowering of certain analytics tools. These are the sweet spots that we want to translate and adapt for healthcare.

There are cross-industry analytics themes that shape the manifestation of advanced analytics generally and have a strong influence on the analytics sweet spots for each industry.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 36/41

Modeling Behavior Change

Predictive modeling to change behavior is a powerful, advanced analytics method used across the industries. Most of the breakthrough applications of predictive modeling across industries focus on understanding and changing behaviors of customers. Examples include the following:

• Retail—Determine the probability that a woman is pregnant and her estimated delivery date.

• Banking—Determine the likelihood of divorce as a major predictor of loan default.

• Political campaigns—Determine the messages most likely to convert an undecided voter.

• Sports—Determine what attributes of players contribute to team wins.

It’s the Customer, Stupid

It goes without saying that customers are what make businesses thrive or die. Businesses can be distracted and focus on other priorities but do so at their peril. The pathway to growth is realized by understanding customers and responding in ways to earn their business. The analytics combo of predictive modeling with “boundless personal data” allows unprecedented views into what makes customers tick.

Boundless Personal Data

We live in a surveillance society. There is a huge business and government appetite to know everything about us. For example:

• Retail—“If you use a credit card or a coupon, or fill out a survey, or mail in a refund, or call the customer help line, or open an e-mail we’ve sent you or visit our Web site, we’ll record it and link it to your Guest ID...and we can buy data about

you...(such as) charitable giving and cars you own.5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end05)

• Politics—“The campaigns spent over $13 million on acquisition of data like whether voters may have visited pornography Web sites, have homes in foreclosure, are more prone to drink Michelob Ultra than Corona or have gay

friends or enjoy expensive vacations.”6 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end06)

• Military—The National Counterterrorism Center can use any information the government has collected on you, including “flight records, lists of Americans hosting foreign-exchange students, financial records of people seeking federally backed mortgages, health records of patients at veterans’ hospitals...this obscure agency has permission to

study [any database] for patterns.”7 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end07)

Most of the industries reviewed are using boundless personal data to feed their customer analytics engines. The new data usually come from outside the industry including “open” public databases, data “snatchers,” and Web click trackers. The large increase in the diversity and volume of personal data, in combination with other analytics methods such as predictive modeling and technology game changers, has been a significant factor in solving business problems and demonstrating the value of analytics across industries.

Big Data Promises

The promise of big data is great and alluring. McKinsey & Company proclaims that “it will become the basis of competition,

underpinning new waves of productivity growth, innovation, and consumer surplus.”8

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end08) Boundless personal data is a piece of it. But big data goes beyond that and looks to extract meaning from every digital signal that is emitted. It is likely that harnessing big data will lead to a new world neural system that can measure almost anything. For the moment, it appears that the technology “hows” are ahead of the business “whys” and “whats.” It is unknown how this revolution, like the Internet revolution before it, will play out and when the big promises will be fulfilled.

Technology Game Changers

Technology advances facilitate the use of data for analytics. Three game changers have been influential:

• Clicks—The Internet has transformed the way businesses communicate, market, do commerce with customers, and collect data about them. One example is the ability to do randomized trials, or A/B testing, of alternative Web site features—for example, how to get the most contributions during a political campaign—on large samples and virtually instantaneously.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 37/41

• Mobile—Earlier we described the challenges and opportunities of mobile. It is seen as the preferred platform for customer communications across industries.

• Hadoop—Hadoop is an open source software framework that allows for the distributed processing of large data sets across clusters of computers using a simple programming model. It is less expensive and easier to get up and running than commercial applications; for example, it is cloud based and works across hardware. It has greatly facilitated the analysis of boundless data.

Looming Privacy Concerns

As boundless personal data increases the utility of analytics to address business needs, it also runs the risk that the “creepy factor” will stop it dead in its tracks. Most of these data are collected and used without the consent of the individual. For example, personal data are collected from children as they traverse the Internet and then used for tailoring marketing messages to them. This concern about privacy is acknowledged but largely ignored, and the response is often to deny its existence. The data are valuable and, while the gate is open, there are few restrictions on their use. But a few breaches of privacy might bring on a spate of consumer complaints and Congressional action.

Sociology, Not Technology

All the analytics sweet spots across the industries are successful because they complement and support important business needs and strengths. Excellent analytics methods can be developed in a bubble outside the realities of the business. But unless they are used to solve business problems, they collect dust and are an expense and not an investment. Making things happen/change/succeed is what business is about. It’s not about the technology; it’s the sociology of getting things done. What is clear from the analytics sweet spots is that the bull’s-eye for the value proposition for analytics is understanding business challenges and strengths and providing tools and expertise in the right way and at the right time to support the business.

Analytics Adaptations for Health and Healthcare The final stage in the discovery process is the translation of industry analytics sweet spots into potential adaptations in healthcare. It is hard to imagine, on the face of it, how banking is anything like healthcare. The task involves connecting dots between the industries to arrive at some revelation. It involves some logic, but is mostly about creativity. Creativity often leads to an epiphany.

For example, in the banking case, and reflecting on the industry strength in risk assessment, one ponders what it is about assessing customer capabilities and risks that might apply in healthcare. What emerged is that one of the most ingrained and intractable issues in healthcare is getting people to be active co-producers of their own health and thereby improve outcomes. People’s behavior is the biggest influencer of health functioning followed by many social determinants. Can the risks, capabilities, and barriers to the fulfillment of doctors’ orders, prescriptions, and plans that rest with the individual be measured and then managed better? This healthcare adaptation, called the Health Improvement Capability Score (HICS), is one of the seven healthcare adaptations that are distilled from the discovery process.

The seven adaptations, listed in Figure 10.2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10fig02) , cluster into three important areas of health and healthcare, including population health, patient engagement, and provider performance. The adaptations for population health include obesity detection, well-being, and my dashboard. For patient engagement, the adaptations include radical personalization and capabilities and support. Provider performance includes the adaptations of peer-to-peer and team centered.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 38/41

Figure 10.2 Seven health and healthcare analytics adaptations

Another example of an adaptation for healthcare comes from retail, banking, and politics...to reduce obesity. The goal is to reduce the incidence of diabetes through early identification of individuals with premorbid obesity, which is followed by targeted intervention programs to reverse weight gain. Many of these individuals will not have received sickness care, or might have received it for other reasons that did not include a risk assessment for diabetes. Therefore, there might not be any relevant data included in a traditional medical record or claims form that speaks to obesity. Retail, banking, and politics have multiple data sources and sophisticated analytics to identify likely buyers, uncommitted voters, or probable defaulters, and they have programs to sell, convince, or reject those so identified.

The analytics adaptation is straightforward. The determination of the degree of obesity is easily determined by a simple calculation of two variables: height and weight. The data are available from firms that aggregate publicly available data. The analytics methods include predictive modeling and the collection and integration of external data. The major challenges are belief systems that question the utility and appropriateness of external data and whether interventions intended to reverse obesity are effective.

Another example comes from sports—to improve provider outcomes. In sports, there is a long tradition of measuring the performance of athletes and making them transparent. One could focus on the paradox that consumers know much more about their sports heroes than about their doctors, though they should know more about a doctor’s “batting average.” But we take a different view. Performance measurement of players has been more about entertainment than about winning games. The pressures on the business of sports have moved the measurement to outcomes (winning) and how the combination of player attributes and individual performance contributes to team wins. Indeed, the sports industry now realizes that after decades of concentrating on individual performance, the business needs to make teams work better.

The analytics adaptation is to develop the ability to measure clinical performance at the care team level and demonstrate its superiority in driving performance improvements and outcomes relative to existing organizational levels of aggregation, such as a hospital or medical group. The measurement should dovetail with the emerging reality that care, especially for the chronically ill, requires a well-functioning team of physicians, other caregivers, communities, and patients. The measurement might well influence the management of care to make the components work together for the good of the whole.

The analytic methods include predictive modeling, external data collection, data integration, and dashboards. There are major challenges, including persistent resistance to performance measurement at a granular level and the belief that individual physicians are the key ingredient in producing patient outcomes.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 39/41

Putting Ideas into Action It is important to generate extra-industry ideas and to show the way across the bridge to healthcare. Ideas are fragile. Eggars and O’Leary think of ideas as seeds. Harkening back to a biblical parable, they note: “Some seeds land on rocky soil. Others are eaten by birds, and some sprout only to be choked by thorns. Only through a fortuitous combination of sun, soil, and water will a

seed grow into a plant and bear fruit.”9 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end09) And progressing ideas to the endpoint of making a real difference is a long journey, and not all ideas deserve the passage.

The endpoint is embedding analytics into the ongoing operations of an organization. The pathway includes six stages. We have discussed the idea phase. The second is the design stage, which involves a plan of action and the justification for implementing it. The third stage is making the decision to adopt the plan, which is discussed in the next paragraph. The next stage is execution. Goethe noted in the eighteenth century, “To put your ideas into action is the most difficult thing in the world.” The next stage is results where the performance of the innovation is evaluated. Unfortunately, most innovations fail from innumerable snags in the delivery of the program. Finally, the last stage is reinvention. A successful innovation changes and adapts through a learning and improvement process.

The adoption decision is complicated. Fortunately, there is body of knowledge that is represented in The Innovation Adoption Factors model (see Figure 10.3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10fig03) ), which describes six areas or domains that need to be considered and mastered to persuade individuals to make the decision to adopt an innovation. The model is composed of two halves. On the left side is the idea stage and on the right is the design stage. The idea stage includes the domains of innovation receptivity, ideation maturation, and urgency/timing. The design stage includes the domains of the innovation’s attributes, the organization’s capabilities, and the readiness to position the innovation for adoption.

Figure 10.3 Innovation Adopters Factors model (Adapted from Rogers on diffusion

theory,10 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end10) Eggers and

O’Leary on getting things done in government,11

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end11) Kingdon on agenda

setting,12 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end12) Hasenfeld

and Brock on policy implementation,13

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end13) Pressman and

Wildavsky on program implementation,14

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end14) Gladwell on the

tipping point model of change,15

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end15) and Plsek16

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end16) and Stacey17

(http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end17) on complexity theory.)

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 40/41

In the book, the six domains are decomposed into 18 factors. The factors become the basis for a guidebook that can be used for evaluation and planning for decisions about innovation adoption. A case study illustrates the scoring, management response, and improvements to the innovation process to make the innovation acceptable.

Notes 1 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end01a) . The chapter is a brief summary of the

book A Framework for Applying Analytics in Healthcare: What Can Be Learned from the Best Practices in Retail, Banking, Politics, and Sports, by Dwight McNeill and published by FT Press.

2 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end02a) . E. McGlynn, S. Asch, J. Adams, et al., “The Quality of Healthcare Delivered to Adults in the United States,” New England Journal of Medicine 348 (2003): 2635- 2645.

3 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end03a) . “Analytics: The New Path to Value,” a joint MIT Sloan Management Review and IBM Institute for Business Value study, Massachusetts Institute of Technology, 2010.

4 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end04a) . Edward de Bono, BrainyQuote.com, Xplore Inc., 2013, www.brainyquote.com/quotes/quotes/e/edwarddebo389925.html (http://www.brainyquote.com/quotes/quotes/e/edwarddebo389925.html) (accessed 3/29/13).

5 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end05a) . Charles Duhigg, “How Companies Learn Your Secrets,” New York Times, February 16, 2012, www.nytimes.com/2012/02/19/magazine/shopping- habits.html?pagewanted=all (http://www.nytimes.com/2012/02/19/magazine/shopping-habits.html?pagewanted=all) .

6 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end06a) . Charles Duhigg, “Campaigns Mine Personal Lives to Get Out Vote,” New York Times, October 13, 2012, www.nytimes.com/2012/10/14/us/politics/campaigns-mine-personal-lives-to-get-out-vote.html? pagewanted=all (http://www.nytimes.com/2012/10/14/us/politics/campaigns-mine-personal-lives-to-get-out-vote.html?

pagewanted=all) .

7 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end07a) . Bill Keller, “Invasion of the Data Snatchers,” New York Times, January 13, 2013, www.nytimes.com/2013/01/14/opinion/keller-invasion-of-the-data- snatchers.html?pagewanted=all (http://www.nytimes.com/2013/01/14/opinion/keller-invasion-of-the-data-snatchers.html?

pagewanted=all) .

8 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end08a) . James Manyika et al., “Big Data: The Next Frontier for Innovation, Competition, and Productivity,” McKinsey Global Institute, May 2011, www.mckinsey.com/insights/mgi/research/technology_and_innovation/big_data_the_next_frontier_for_innovation (http://www.mckinsey.com/insights/mgi/research/technology_and_innovation/big_data_the_next_frontier_for_innovation) .

9 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end09a) . William Eggers and John O’Leary, If We Can Put a Man on the Moon: Getting Big Things Done in Government (Boston: Harvard Business Press, 2009).

10 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end10a) . Everett Rogers, Diffusion of Innovations, 5th Edition (New York: Free Press, 2003).

11 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end11a) . Eggers and O’Leary, If We Can Put a Man on the Moon.

12 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end12a) . John Kingdon, Agendas, Alternatives, and Public Policy (Boston: Little Brown, 1984).

13 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end13a) . Y. Hasenfeld and T. Brock, “Implementation of Social Policy Revisited: A Political Economy Perspective,” Administration & Society 22 (1991): 451-79.

14 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end14a) . J. Pressman and A. Wildavsky, Implementation (Berkeley, CA: University of California Press, 1984).

15 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end15a) . M. Gladwell, The Tipping Point (New York: Random House, 1999).

16 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end16a) . Paul Plsek, “Complexity and the Adoption of Innovation in Healthcare,” presentation to conference on Accelerating Quality Improvement in Healthcare: Strategies to Speed the Diffusion of Evidence-Based Innovations, Washington, DC, January 27, 2003.

9/14/22, 11:35 AM Print

https://content.uagc.edu/print/McNeill.2947.17.1?sections=part02,ch05,ch06,ch07,ch08,ch09,ch10&content=all&clientToken=92efa779-015b-e3a8-… 41/41

17 (http://content.thuzelearning.com/books/McNeill.2947.17.1/sections/ch10#ch10end17a) . R. D. Stacey, Complex Responsive Processes in Organizations: Learning and Knowledge Creation (New York: Routledge, 2001).