quantitative research

Journal of Clinical Epidemiology 63 (2010) 1000e1010

A prospective cohort study found that provider and information continuity was low after patient discharge from hospital

Carl van Walraven a,b,*, Monica Taljaard

a , Chaim M. Bell

b,c,d,e , Edward Etchells

c , Ian G. Stiell

f ,

Kelly Zarnkeg, Alan J. Forstera aOttawa Hospital Research Institute, Ottawa, Ontario, Canada

b Institute for Clinical Evaluative Sciences, Toronto, Ontario, Canada

c Department of Medicine, University of Toronto, Toronto, Ontario, Canada

dKeenan Research Centre of the Li Ka Shing Knowledge Institute, St. Michael’s Hospital, Toronto, Ontario, Canada eDepartment of Health Policy Management and Evaluation, University of Toronto, Toronto, Ontario, Canada

f Department of Emergency Medicine, University of Ottawa, Ottawa, Ontario, Canada

g University of Calgary, Alberta, Canada

Accepted 25 January 2010

Abstract

Objective: Continuity of care is composed of provider and information continuity and can change value over time. Most studies that have quantitatively associated continuity of care and outcomes have ignored these characteristics. This study is a detailed examination of continuity of care in patients discharged from hospital that simultaneously measured separate components of continuity over time or determined the factors with which they are associated.

Design Setting: Multicenter, prospective cohort study of patients discharged to the community after elective or emergent hospitaliza- tion. For all physician visits during 6 months after discharge, we identified the physician and the availability of particular information (in- cluding hospital discharge summary and any information from previous physician visits). Four physician continuity scores (preadmission; hospital admitting; hospital consultant; and postdischarge) and two information continuity scores (discharge summary and postdischarge visit information) were calculated for all patients (range: 0e1, where 0 is perfect discontinuity and 1 is perfect continuity).

Results: Four thousand five hundred fifty-three people were followed for a median of 175 days. Both provider (range of median values: 0e0.410) and information (range: 0.220e0.427) continuity scores were low and varied extensively over time. With a few exceptions, con- tinuity measures were independent of each other. The influence of patient factors on continuity varied extensively between the continuity measures with the most influential factors being admission urgency, admitting service, and the number of physicians who regularly treated the patient.

Conclusion: Both provider and information continuity was low in patients discharged from hospital. Continuity measures can change extensively over time, which are usually independent of each other, and are associated with patient and admission characteristics. Future studies should measure multiple components of provider and information continuity over time to completely capture continuity of care. � 2010 Elsevier Inc. All rights reserved.

Keywords: Continuity of care; Time-dependent covariates; Cohort study; Generalized linear mixed model; Continuity of information; Communication

1. Introduction

Continuity of care is considered a cornerstone for opti- mal patient care and is central to primary care medicine [1]. Continuity of care occurs when a patient experiences coherent and linked care over time and is composed primar- ily of provider and information continuity [2]. Provider

* Corresponding author. Ottawa Hospital Research Institute, ASBI-003

Carling Avenue, Ottawa, Ontario, Canada K1Y 4E9.

E-mail address: [email protected] (C. van Walraven).

0895-4356/$ - see front matter � 2010 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2010.01.023

continuity results from an ongoing relationship between a patient and provider over time, whereas information con- tinuity indicates that data from prior events are available for a subsequent patient encounter.

The association between continuity of care and patient outcomes has been frequently studied [3]. However, to completely quantify the association between continuity and patient outcomes, we believe that four issues regarding the measurement and expression of continuitydwhich have received limited attention in the literaturedmust be ad- dressed. First, despite the recognition that continuity of care has multiple components [2], none of the studies in

1001C. van Walraven et al. / Journal of Clinical Epidemiology 63 (2010) 1000e1010

a systematic review of continuity of care and outcome [3] examined both provider and information continuity in a de- fined group of patients. Such analyses are necessary to completely describe continuity in a patient cohort.

Second, provider and information continuity measures both will change value over time at each visit that a patient experiences. Recognizing this by expressing continuity measures as time-dependent variables would let researchers examine the effect of interventions or events on continuity of care. Time-dependent covariates would also improve regression models that determine how continuity is associ- ated with outcomes. They could be used in a proportional hazards model [4,5] or longitudinal analysis. However, in our systematic review [3], only four studies measured and expressed continuity as a time-dependent covariate [6e9].

Third, the direct relationship between distinct continuity measures has not been directly studied. It would not be unexpected that separate continuity measures are related because individual provider visits can have multiple charac- teristics that individually influence those measures. Strong relationships between these continuity measures could introduce multicollinearity into regression models and make their results unreliable.

Finally, the factors that influence continuity have not been extensively studied. Although several studies have used survey methods to examine the association of patient factors with continuity [10e14], the influence of directly measured patient and system factors on continuity of care has not been commonly studied. This information is neces- sary to identify potential confounders in analyses measur- ing the association between continuity and outcomes and infer why continuity might be compromised.

In this study, we addressed these four issues when we studied continuity in a large cohort of patients discharged from hospital to the community.

2. Methods

2.1. Study design

This was a multicenter prospective cohort study of patients discharged to the community from the medical or surgical services of 11 Ontario hospitals (six university- affiliated hospitals and five community hospitals) in five cities after an elective or emergent hospitalization. Included patients had to be cognitively intact, have a telephone, and provide written informed consent. Patients were not included if they were less than 18 years old, discharged from obstetrical or psychiatric services, or discharged to nursing homes. The study was approved by the research ethics board of each participating hospital.

We chose the postdischarge period to study continuity because it is an ideal time period to study continuity. Patients discharged from hospital have a high risk of poor outcomes [15]. Postdischarge patients often have poor provider [9] or information continuity [8,16,17].

2.2. Data collection

Before hospital discharge, patients were interviewed by study personnel to identify their baseline functional status, living conditions, all physicians who regularly treated the patient (including both family physicians and consultants), and chronic medical conditions. The latter were confirmed by a review of the patient’s chart and hospital discharge summary, when available. The chart and discharge sum- mary were also used to identify diagnoses in hospital and medications at discharge.

Patients or their designated contacts were telephoned 1, 3, and 6 months after their hospital discharge to identify the date and physician of all visits that they had. We only counted one visit for the study if patients saw the same phy- sician more than once in a particular day. Emergency room visits and hospitalizations (including same-day surgeries) were not included in this analysis.

For each physician visit, we determined the availabil- ity of both a discharge summary for the index hospitali- zation and information from previous postdischarge visits that the patient had with other physicians. The methods used to collect these data have been previously detailed [18]. Briefly, we used three complimentary methods to elicit this information from each follow-up physician. First, patients gave physicians a survey on which they listed all prior visits with other doctors for which they had information. If this survey was not returned, we faxed the survey to the physician or we phoned the physician or their office staff and adminis- tered the survey by telephone.

2.3. Continuity measures

In this study, we used the framework and terminology of Reid et al. [2], wherein the primary components of overall continuity of care consist of provider and information con- tinuity. For the posthospitalization period, we measured provider continuity for physicians who provided patient care during three distinct phases: the prehospital period, the hospital period, and the postdischarge period. Preho- spital physicians were those classified by the patient as their regular physician(s) (defined as a physician they had seen in the past and were likely to see again in the future). Hospital provider continuity was divided into hospital phy- sician (i.e., the physician to whom the patient was admit- ted) continuity and hospital consultant (i.e., another physician who consulted on the patient during admission) continuity. Information continuity was broken down as discharge summary continuity and postdischarge visit information continuity.

To quantify provider and information continuity, we used Breslau’s Usual Provider of Continuity (UPC ) [19], which measures the proportion of visits with the physician of interest (for provider continuity measures) or the propor- tion of visits having the information of interest (for infor- mation continuity measures). The UPC was calculated as:

Table 1

Details

Provid

A. P

B. H

C. H

D. P

Inform

E. D

F. Po

in

Abb

1002 inical Epidemiology 63 (2010) 1000e1010

UPC 5 ni=N;

C. van Walraven et al. / Journal of Cl

where UPC ranges from 0 to 1 (where 0 is perfect discon- tinuity and 1 is perfect continuity); ni is the number of post- discharge visits to the physician type of interest (e.g., prehospital, hospital, and postdischarge) or the number of visits at which the information of interest (e.g., discharge summary) was available; and N is the total number of post- discharge visits. Details for calculating each provider and information continuity measure are given in Table 1.

Figure 1 illustrates how we calculated each continuity measure over time for a fictitious patient. This figure high- lights that (1) all continuity measures are incalculable before the first postdischarge visit; (2) all continuity measures change value at each visit after during patient observation; and (3) a physician could be more than one physician type (e.g., a physician who treated a patient before the admission in which he/she was the attending physician would be both a prehospital and hospital physician for that patient).

2.4. Analysis

For each continuity measure within each patient, we cal- culated the mean daily continuity score as:

PN 1 C

N ;

where C is the continuity score on each day of observation (Table 1). This score was summed over the total number of postdischarge days that the patient had a measurable conti- nuity score (i.e., N ). The mean daily continuity score can be considered an ‘‘incidence density’’ and has two advan- tages. First, it allows the calculation of group-level continu- ity values by using a weighted average of individual-patient

of continuity measures and their calculation

Numerator

er continuity measures

readmission No. of postdischarge visits with MD who

regularly treated patient before admission

ospital No. of postdischarge visits with MD under

whom patient was admitted during index

hospitalization

ospital consultant No. of postdischarge visits with MD who

consulted on patient during index

hospitalization

ostdischarge No. of postdischarge visits with MD who

previously saw patient postdischarge

ation continuity measures

ischarge summary No. of postdischarge visits with MD who

had a copy of the discharge summary

from index hospitalization at the time

of the visit

stdischarge visit

formation

No. of previous postdischarge visits (with

another MD) for which information

was available

reviation: MD, physician.

continuity values. Second, the mean daily continuity ac- counts not only for continuity values but also their duration.

We found that the mean daily continuity score for all provider and information continuities was not normally dis- tributed. For descriptive purposes, we categorized each continuity measure into four groups based on their mean daily continuity score (0; O0 to median continuity; median up to 75th percentile continuity; and 75th percentile to maximum continuity). Because hospital consultant continu- ity was very low in our sample, this was categorized into two groups (0 and O0).

To account for patient clustering within hospitals, we used generalized linear mixed modeling (GLMM) to deter- mine the independent association of patient and admission factors with each continuity measure. PROC GLIMMIX in SAS (Cary, NC, USA) was used to create the models with a beta distribution for the continuity measure, a logit link function, and the KenwardeRoger method was used for computing the denominator degrees of freedom. The GLMM methodology allowed us to express the hospital as a random effects variable, thereby improving the gener- alization of our findings to other hospitals. The importance of all diagnosis nonspecific variables was first determined with univariate GLMM models. Significant variables (i.e., those with a type 3 fixed effects P-value that was !0.05) were offered to the multivariate model in a forward selec- tion manner. Variables were retained if they remained significant in the model. Goodness of fit was evaluated using studentized residual plots.

Calculation of postdischarge visit information continuity scores required physicians to tell us whether they had infor- mation from previous visits that the patient had with other physicians. As described above, this information was provided by paper or phone survey. Of the 23,454

Denominator Notes

No. of postdischarge

MD visits

No. of postdischarge

MD visits

No. of postdischarge

MD visits

Applies only to patients who had

>1 consultation in hospital

No. of postdischarge

MD visits �1 Can be calculated only after first

postdischarge visit

No. of postdischarge

MD visits

No. of previous visits

with another MD

Can be calculated only if patient had

prior visits with another MD. The

mean value at each visit was averaged

to calculate continuity score over time

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

1 C

O N

T I N

U I T

Y S

C O

R E

Pre-hospital Hospital Consultant Post DC

Pre Hospital MD Hospital MD Consultant MD

Post DC MDs

Continuity Scores

WEEKS FROM DISCHARGE

Fig. 1. Illustration of provider continuity measures for a patient following

discharge from hospital. This figure illustrates how we calculated continu-

ity scores at each postdischarge visit for a hypothetical patient. Here, we

focus on provider continuity. The top of the figure identifies the physicians

who treated this patient (prehospital physician: Dr Circle; hospital physi-

cian: Dr Diamond; and hospital consultant: Dr Square). This patient’s first

postdischarge was with Dr Circle. As a result, the prehospital provider con-

tinuity score at the first visit was 1.0 (1 over 1). The rest of the provider

continuity scores were 0. All continuity scores stayed at these values until

the second visit with Dr Triangle, whom the patient had never seen below.

Because Dr Triangle is not a prehospital physician, the prehospital pro-

vider continuity score drops to 0.5 (1 over 2). All of the other provider con-

tinuity scores remain at 0. The third visit was with Dr Square, a hospital

consultant. As a result, the consultant continuity score increases to 0.33

(1 over 3). The prehospital continuity score drops to the same value. This

process was repeated at each visit to calculate all provider continuity

scores at each day of the patient’s observation. Details for calculating con-

tinuity scores are given in Table 1.

Recruited 5035

1+ Follow-up Interview 4761 (94.6%)

No Follow-Up 274 (5.4%)

1+ MD visit recorded 4553 (90.4%)

No MD visits 208 (4.0%)

Complete Follow-Up 100 (2.0%)

Incomplete Follow-Up 108 (2.1%)

Complete Follow-Up 4222 (83.8%)

Incomplete Follow-Up 331 (7.3%)

Fig. 2. Patient follow-up. The study cohort (n 5 4,553) is indicated in

green. Its creation from the originally recruited patients is illustrated.

Red boxes indicate recruited patients with incomplete follow-up. Blue

boxes indicate patients with complete follow-up. Details for loss to

follow-up are given in the study text.

1003C. van Walraven et al. / Journal of Clinical Epidemiology 63 (2010) 1000e1010

postdischarge visits, we had complete information for 18,087 (77.1%).

We therefore imputed missing data with a logistic model that contained all relevant variables, including those from a previous study that examined factors influencing informa- tion continuity [18]. This model had 37 variables, and it estimated the probability that information from each previ- ous visit with another physician was available. Using this point estimate and its standard error, we randomly selected the estimated probability that information was available for a previous visit. This estimated probability was then used in a Bernoulli draw to impute a value of 0 or 1 indicating whether information was available for that particular visit. A total of 10 imputations were used for the analysis. We determined the important variables to be included in the model using a complete case analysis (i.e., visits with miss- ing information were excluded). A GLMM model was then created for each imputed data set. The parameter estimates from each regression model were combined using PROC MIANALYZE (SAS, Cary, NC, USA).

Imputation was not required for discharge summary con- tinuity because the summary identified the date on which it was created and all physicians to whom a discharge sum- mary was sent. Comparing this information with the visit date and physician allowed us to infer whether the physi- cian had a copy of the discharge summary at the time of

the visit. As in a previous study [20], we allowed a time lapse of 3 days for the summary to be sent to the receiving physician.

3. Results

Between October 2002 and July 2006, we enrolled 5,035 patients from 11 hospitals (Fig. 2). Four thousand five hun- dred fifty-three (90.4%) patients made it into our study, of whom 4,222 (83.8% of the original cohort) had complete follow-up for the entire 6-month study. Seven hundred thir- teen (14.2%) patients had incomplete follow-up because 300 were lost to follow-up; 169 refused participation; 128 died; 86 were readmitted to hospital; and 30 were trans- ferred into a nursing home.

Study patients are described in Table 2. Patients were observed in the study for a median of 175 days (interquar- tile range [IQR]: 175e178). During this time, they had a median of four physician visits (IQR: 3e6). The first post- discharge physician visit occurred on a median of 11 days (IQR: 6e20) after discharge from hospital.

3.1. Provider and information continuity

Figure 3 summarizes the mean daily continuity scores for all measures. All continuity distributions, with the exception of consultant continuity scores, had bimodal dis- tributions with modes occurring at the minimum and max- imum values. All continuity measures had median values below 0.5 with hospital physician and hospital consultant continuity having the lowest values (median: 0.078, IQR: 0e0.468 and 0 IQR: 0-0, respectively). The highest provider continuity measure was the postdischarge physi- cian continuity score with a median value of 0.410 (IQR: 0.190e0.792). The median (IQR) discharge summary and postdischarge visit information continuity was 0.427 (0e0.842) and 0.220 (0e0.775), respectively.

Table 2

Description of patient cohort (N 5 4,553)

Factor Value N (%)

Mean patient age (SD) 61.4 (16.8)

Female 2,396 (52.6)

Lives alone 1,053 (23.1)

Charlson score 0 3,508 (77.0)

1 145 (3.2)

2 615 (13.5)

O2 285 (6.2) Chronic disease

Hypertension 1,813 (39.8)

Dyslipidemia 893 (19.6)

Diabetes mellitus 788 (17.3)

Coronary artery disease 650 (14.3)

Cancer 529 (11.6)

Previous surgical procedures

CABG 529 (11.6)

Laparoscopic cholecystectomy 283 (6.2)

Appendectomy 310 (6.8)

Hip arthroplasty 255 (5.6)

Knee arthroplasty 113 (2.5)

No. of admissions in previous 6 months 0 3,091 (67.9)

1 1,089 (23.9)

O1 373 (8.2) No. of activities of daily living requiring

aids

0 4,261 (93.6)

1 165 (3.6)

O1 127 (2.8) No. of MDs who see patient regularly 0 347 (7.6)

1 3,944 (86.6)

2 203 (4.5)

O2 59 (1.3) Index hospitalization description

Median length of stay in days (IQR) 4 (2e8)

Median total number of discharge

medications (IQR)

4 (2e7)

Emergent admission 2,589 (56.9)

Admitted to medical service 1,999 (43.9)

Acute diagnoses

CAD 296 (6.5)

Neoplasm of unspecified nature 246 (5.4)

Heart failure 198 (4.3)

Influenza 141 (3.1)

Cardiac dysrhythmias 123 (2.7)

Acute procedures

CABG 216 (4.7)

Total knee arthroplasty 190 (4.2)

Total hip arthroplasty 124 (2.7)

Appendectomy 113 (2.5)

Colectomy/colostomy 79 (1.7)

No. of complications in hospital 0 3,989 (87.6)

1 396 (8.7)

O1 168 (3.7) No. of consultations in hospital 0 2,829 (62.1)

1 1,402 (30.8)

O1 322 (7.1)

Abbreviations: IQR, interquartile range; CABG, coronary artery bypass

graft; CAD, coronary artery disease.

1004 C. van Walraven et al. / Journal of Clinical Epidemiology 63 (2010) 1000e1010

3.2. Time-dependent nature of continuity

Individual patient continuity measures varied exten- sively during their observation period (Table 3). For pread- mission and postdischarge provider continuity, the median

individual-patient range was 0.333 (or one-third of the continuity scale). For all continuity measures except hospi- tal consultant continuity, the 75th percentile of the individual-patient range was at least one-half of the entire continuity scale (i.e., 0.5).

As with the patient level, group-level continuity measures also varied extensively over time (Fig. 4). For each continuity measure, the proportion of people at each continuity score range changed extensively over time. For example, more than 80% of people had no follow-up with a hospital physician in the first week after discharge from hospital. However, this proportion decreased to 50% by 6 months. Such large varia- tions in continuity were seen in all continuity measures except that for hospital consultant, which remained consis- tently low throughout the study with approximately 90% of patients never seeing a hospital consultant in follow-up. In addition, dissemination of information between postdi- scharge physicians improved as time progressed, but infor- mation for any previous visit was always absent in more than one-third of patients throughout in the study.

3.3. Correlation between continuity measures

Figure 5 illustrates that most continuity measures were independent of each other or were only weakly associated with several notable exceptions. Prehospital physician and discharge summary continuity were significantly and posi- tively correlated. Prehospital physician and hospital physi- cian continuity were negatively associated with each other. The strongest correlations existed between postdischarge physician and postdischarge information continuity with values of approximately 0.5.

Figure 5 also illustrates the correlations between conti- nuity measures over time. Most of the correlations changed in the first month after discharge, likely reflecting instabil- ity of the individual continuity measures when the total number of follow-up visits was small. More than a month after discharge from hospital, most correlations remained stable with two exceptions. The correlation between postdi- scharge physician and postdischarge information, as well as that between prehospital and hospital physician, both significantly trended toward unity as time progressed.

3.4. Factors influencing continuity

Table 4 details the independent association of baseline fac- tors with each continuity measure. Some findings are notable. Older patients had significantly better prehospital physician continuity but worse hospital and consultant continuity. As the complexity of patient chronic problems (as reflected by the Charlson score) increased, prehospital provider continuity decreased significantly. An increased number of regular physicians were associated with increased preadmission phy- sician continuity but decreased hospital physician, postdi- scharge physician, and postdischarge information continuity. Patients who stayed in hospital longer had significantly worse

Fig. 3. Provider and information continuity in study patients. This figure summarizes mean daily continuity scores for four provider continuity measures

(AeD) and two information continuity measure (EeF). Each plot presents patient continuity scores for the entire study observation period (see Table 1

for details regarding the calculation of the six continuity measures). Each figure presents these scores (horizontal axis) by groups of 0.1 width. The midpoint

value of each category is presented on the horizontal axis. The vertical axis presents the number of people in each category. Below each plot, we present the

median (‘‘Q2’’) and 75th percentile (‘‘Q3’’) value for each continuity measure.

1005C. van Walraven et al. / Journal of Clinical Epidemiology 63 (2010) 1000e1010

hospital physician continuity but much better discharge sum- mary continuity. This pattern, where worse hospital physician continuity was balanced by improved discharge summary continuity, was also seen when patients were admitted to med- ical instead of surgical services. Emergent admissions had bet- ter continuity with both preadmission physicians and postdischarge physicians but worse continuity with hospital

Table 3

Ranges of continuity measures for individual patients

Minimum 25th Percentile

Provider continuity

A. Preadmission 0 0

B. Attending 0 0

C. In-hospital consultanta 0 0

D. Postdischarge 0 0.087

Information continuity

E. Discharge summary 0 0

F. Postdischarge visit 0 0

a Applies only to those patients who had an in-hospital consultation (n 5 1,

physicians. Finally, having a complication in hospital did not increase hospital physician continuity.

4. Discussion

To our knowledge, this is the most in-depth examination of patient continuity after discharge from hospital. Overall,

Median 75th Percentile Maximum

0.333 0.500 0.923

0 0.500 0.952

0 0 0.917

0.333 0.500 0.894

0.200 0.500 0.923

0.214 0.950 0.950

724).

Fig. 4. Group-level continuity measures over time. These plots present each continuity measure in the study cohort during their observation period after

discharge from hospital. In each plot, the horizontal presents the months from discharge, and the vertical axis presents the percent of people with each con-

tinuity value range. The continuity values that consist each range are presented below each plot. In each plot, gray indicates a continuity measure of 0; light

blue indicates continuity measures between 0 and the median; dark blue indicates continuity measures between median and the 75th percentile; and black

indicates continuity measures between 75th percentile and 1.

1006 C. van Walraven et al. / Journal of Clinical Epidemiology 63 (2010) 1000e1010

we found that continuity was low in all spheres of both pro- vider and information continuity; each of these measures can change extensively over time for both individual pa- tients and the entire population; the individual continuity scores were mostly independent of each other; and provider and information continuity was significantly influenced by a few patient and hospitalization factors.

Our results highlight the poor continuity of care that pa- tients experience when they are discharged from the hospi- tal. The median score was less than 50% for all continuity measures. In the 6 months after discharge from hospital, al- most one-third of patients did not see one of their regular treating physicians. In the same period, one-half of patients never saw the hospital physician who treated them during their admission. Consultants who saw patients during their admission rarely saw them after discharge from hospital. We found it encouraging that more than half of patients

had a discharge summary available for more than 50% of their follow-up visits. However, more than half of patients haddon averagedonly a one-in-five chance or less that information from previous visits with other doctors was available at their follow-up visit. These results show the large room for improvement in continuity of patient care when they are discharged from the hospital.

Because increased continuity of care is associated with improved patient outcomes [3], our findings demonstrate a large opportunity to increase continuity of care, which could lead to better patient outcomes after discharge from hospital. This could be accomplished by consciously ensur- ing that patients are seen in follow-up by their preadmission and hospital physicians. We hope that improved information technologies and enhanced provider integration will increase information continuity. These interventions will increase overall patient continuity in the postdischarge period.

Fig. 5. Correlation between continuity measures over time. The correlation between all combinations of continuity measures is presented over time. The

specific continuity measures in each plot are listed along the top row and left column. In each plot, correlation (vertical axis) is presented as the Spearman

correlation coefficient ranging from �1 to þ1. The horizontal axis presents the number of months since hospital discharge (ranging from 0 to 6). MD 5 physician.

1007C. van Walraven et al. / Journal of Clinical Epidemiology 63 (2010) 1000e1010

However, our analysis shows that increasing one conti- nuity measure could decrease another continuity measure. For example, we found that prehospital physician continu- ity was negatively correlated with hospital physician conti- nuity (Fig. 5). We also found that factorsdincluding patient age, the number of regular physicians, emergent ad- missions, and admission to a medical servicedthat were significantly associated with increased prehospital physi- cian continuity were also associated with decreased hospi- tal continuity (Table 4). It is possible that the various continuity components would have different influences on patient outcomes. We therefore believe that it is essential to accurately quantify the influence of the various compo- nents of continuity of care on patient outcomes prior to in- troducing interventions designed to change patient continuity.

We believe that our study makes several notable con- clusions for future studies regarding continuity of care in patients. First, our findings highlight that continuity of

care can change extensively over time for both individual patients (Table 3) and entire patient groups (Fig. 4). This extensive variation in continuity over time highlights the importance of measuring continuity of care in a time- dependent fashion and expressing them as such in analy- ses. Failure to do so could threaten studies that try to determine the influence of continuity on patient outcomes [21]. Second, our findings show the importance of measur- ing multiple components of continuity of care [2]. Our results quantitatively support this proposition given the complete independence between most continuity measures (Fig. 5).

Several factors were often associated with continuity. Patients from medical services had significantly better con- tinuity for prehospital physician, postdischarge physician, discharge summary, and postdischarge information, whereas surgical patients were significantly more likely to have follow-up by the hospital physician (Table 4). Emer- gent admissions also had significantly increased prehospital

Table 4

Independent influence of factors on provider and information continuity rates

Adjusted relative percent change (95% confidence interval)

Provider continuity Information continuity

Baseline factors Comparitor Prehospital MD Hospital Consultanta Postdischarge, MD

Discharge

summary

Postdischarge

information

Patient age increased

by 1 decade

d 14.1 (11.0, 17.2) �10.7 (�13.6, �7.7) �10.3 (�18.1, �1.7) d d d

Female Male 17.3 (7.8, 27.6) d �37.3 (�53.8, �14.9) d d d Charlson score

1 Score of 0 1.4 (�20.0, 28.6) �28.1 (�50.4, 4.4) d d d d 2 �13.9 (�24.1, �2.3) �17.4 (�30.0, �2.5) d d d d O2 �19.3 (�32.5, �3.6) �21.5 (�42.5, 7.1) d d d d

Admissions in last 6 months

1 None d �14.2 (�26.1, �0.4) d d d d O1 d �21.7 (�37.5, �1.8) d d d d

No. of MDs who see patient regularly

1 None b �29.2 (�41.6, �14.2) d 10.5 (�6.8, 30.9) d �1.1 (�19.1, 20.8) 2 97.7 (60.5, 143.5) �45.1 (�60.4, �24.1) d �11.8 (�32.1, 14.5) d �32.8 (�51.6, �6.8) O2 141.8 (65.8, 252.7) �55.4 (�73.9, �24.0) d �43.8 (�62.9, �14.8) d �57.8 (�75.6, �27.2)

Hospital length of stay

2e3 days !2 days d �14.8 (�27.1, �0.5) d d 31.7 (12.6, 54.0) d 4e7 days d �21.2 (�32.0, �8.7) d d 62.4 (40.5, 87.7) d O7 days d �40.7 (�50.1, �29.6) d d 54.4 (32.5, 79.8) d

Emergent admission Elective 12.4 (1.2, 24.9) �41.0 (�47.6, �33.5) d 11.8 (0.9, 23.9) d 18.0 (4.2, 33.6) Admitted to medical

service

Surgery 73.2 (55.0, 93.4) �82.2 (�84.8, �79.2) d 29.1 (16.2, 43.5) 103 (83.0, 125) 15.1 (1.0, 31.1)

No. of complications in hospital

1 None 20.8 (3.8, 40.5) d d d d d

O1 �11.8 (�29.7, 10.5) d d d d d Consultation in hospital None d 2.0 (0.4, 3.5) c d d d

Abbreviation: MD, physician.

This table presents the independent association of baseline factors with each of the six continuity measures. These associations are expressed as the adjusted relative percent change in the mean daily

continuity score. Positive values indicate that the baseline factor increased continuity compared with the comparator. a

Applies only to those patients who had an in-hospital consultation (n 5 1,724). b

For the Prehospital Physician Continuity model, the comparator was patients with 0 or 1 physician. c This variable was not included in the model (because its value was ‘‘1’’ for all people in this analysis).

1 0

0 8

C .

v a

n W

a lra

v e n

e t

a l.

/ Jo

u rn

a l

o f

C lin

ic a

l E

p id

e m

io lo

g y

6 3

(2 0

1 0

) 1

0 0

0 e

1 0

1 0

1009C. van Walraven et al. / Journal of Clinical Epidemiology 63 (2010) 1000e1010

physician, postdischarge physician, and postdischarge information continuity but significantly worse hospital physician continuity. These findings indicate that the influ- ence of patient factors on various aspects of continuity can vary extensively.

Our study has several strengths that increase the reliabil- ity of its results. We included a large collection of patients who were discharged to the community from 11 different hospitals across Ontario. Our follow-up and data collection for these patients was very complete. Our data allowed us to calculate multiple provider and information continuities for all patients at all times of their follow-up. This allowed us to examine overall patient continuity from multiple views over a protracted period of time.

Our study also has some noteworthy issues that need to be considered when interpreting its results. First, we are uncertain how representative our results would be in other health care environments. It is possible that continuity of care differs greatly in other countries with different health care systems and practice. However, because our study had very inclusive inclusion criteria and was successful in re- cruiting a large proportion of patients being discharged from the study hospitals, we are confident that our results are representative of patients in Ontario. In addition, our analysisdin which the hospital was expressed as a random effects termdshould improve the validity of generalizing our results to other hospitals in Ontario. Second, our study excluded patients discharged from obstetrical and psychi- atric wards. As such, we are uncertain how our results would apply to these patient populations. Third, we did not measure the patients’ perception of their continuity of care. Eliciting the patient’s view would strengthen the study’s measurement of continuity. Fourth, our measure of information continuity was limited to the presence or absence of information about a previous physician encoun- ter. We did not measure the relevance of that information or whether that information was actually used during the current patient encounter. Finally, our analysis treated all visits the same when calculating the continuity scores and this may be inappropriately simple. For example, hav- ing access to the hospital discharge summary is likely more important to patient care in the early postdischarge period. Future analyses examining how continuity of care influences outcomes may determine the interaction of physician visit types on the association of continuity of care on outcomes.

We assessed continuity of care for more than 4,000 patients discharged from 11 community and university- affiliated hospitals. Our study shows that continuity of care for most patients after they leave the hospital is poor. However, accurate representation of patient conti- nuity requires multiple provider and information mea- sures over time. Future studies need to determine the independent association of these continuity measures with important patient outcomes after discharge from hospital. Before the introduction of interventions to

increase continuity, studies are necessary to determine the independent association of each continuity compo- nent with outcomes.

Acknowledgments

None of the authors have any potential conflicts of inter- est, financial interests, relationships, or affiliations relevant to the subject of their manuscript.

Dr van Walraven had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

This study was conducted using funding from Canadian Institutes for Health Research and the Physicians’ Services Incorporated Foundation. Neither funding agency had any role in the conduct of the study.

Dr Forster is a Career Scientist with the Ontario Ministry of Health and Long-Term Care.

References

[1] Harris MF, Frith JF. Continuity of care: in search of the Holy Grail of

general practice. Med J Aust 1996;164:456e7.

[2] Reid R, Haggerty J, McKendry R. Defusing the confusion: concepts

and measures of continuity of healthcare. Ottawa, Canada: Canadian

Health Services Research Foundation; 2002. pp. 1e50.

[3] van Walraven C, Oake N, Jennings A, Forster AJ. The association be-

tween continuity of care and outcomes: a systematic and critical re-

view. J Eval Clin Pract 2010 Jun 11. [Epub ahead of print].

[4] Allison PD, editor. Estimating Cox-regression models with PROC

PHREG. In: Survival analysis using the SAS system. Cary, NC:

SAS Institute Inc.; 2000. p. 111e84.

[5] Fisher LD, Lin DY. Time-dependent covariates in the Cox

proportional-hazards regression model. Annu Rev Public Health

1999;20:145e57. [6] Christakis DA, Wright JA, Koepsell TD, Emerson S, Connell FA. Is

greater continuity of care associated with less emergency department

utilization? Pediatrics 1999;103(4 Pt 1):738e42.

[7] Christakis DA, Mell L, Koepsell TD, Zimmerman FJ, Connell FA.

Association of lower continuity of care with greater risk of emer-

gency department use and hospitalization in children. Pediatrics

2001;107:524e9. [8] van Walraven C, Seth R, Austin PC, Laupacis A. Effect of discharge

summary availability during post-discharge visits on hospital read-

mission. J Gen Intern Med 2002;17(3):186e92.

[9] van Walraven C, Mamdani MM, Fang J, Austin PC. Continuity of

care and patient outcomes after hospital discharge. J Gen Intern

Med 2004;19:624e45.

[10] Turner D, Tarrant C, Windridge K, Bryan S, Boulton M, Freeman G,

et al. Do patients value continuity of care in general practice? An

investigation using stated preference discrete choice experiments.

J Health Serv Res Policy 2007;12:132e7.

[11] Baker R, Boulton M, Windridge K, Freeman GK. Interpersonal con-

tinuity of care: a cross-sectional survey of primary care patients’ pref-

erences and their experiences. Br J Gen Pract 2007;57(537):283e9.

[12] Mainous AG III, Goodwin MA, Stange KC. Patient-physician shared

experiences and value patients place on continuity of care. Ann Fam

Med 2004;2:452e4.

[13] Love MM, Mainous AG III, Talbert JC, Hager GL. Continuity of care

and the physician-patient relationship: the importance of continuity

for adult patients with asthma. J Fam Pract 2000;49:998e1004.

1010 C. van Walraven et al. / Journal of Clinical Epidemiology 63 (2010) 1000e1010

[14] Christakis DA, Kazak AE, Wright JA, Zimmerman FJ, Bassett AL,

Connell FA. What factors are associated with achieving high continu-

ity of care? Fam Med 2004;36(1):55e60.

[15] Forster AJ, Murff HJ, Peterson JF, Gandhi TK, Bates DW. The

incidence and severity of adverse events affecting patients

after discharge from the hospital. Ann Intern Med 2003;138(3):

161e7.

[16] Bell CM, Schnipper JL, Auerbach AD, Kaboli PJ, Wetterneck TB,

Gonzales DV, et al. Association of communication between

hospital-based physicians and primary care providers with patient

outcomes. J Gen Intern Med 2009;24:381e6.

[17] Kripalani S, LeFevre F, Phillips CO, Williams MY, Basaviah P,

Baker DW. Deficits in communication and information transfer

between hospital-based and primary care physicians: implications

for patient safety and continuity of care. JAMA 2007;297:831e41.

[18] van Walraven C, Taljaard M, Bell C, Williams MV, Basaviah P,

Baker DW. Information exchange among physicians caring for the

same patient in the community. Can Med Assoc J 2008;179:1013e8. [19] Breslau N, Reeb KG. Continuity of care in a university-based

practice. J Med Educ 1975;965e9.

[20] van Walraven C, Seth R, Laupacis A. Dissemination of discharge

summaries. Not reaching follow-up physicians. Can Fam Physician

2002;48:737e42.

[21] van Walraven C, Davis D, Forster AJ, Wells GA. Time-dependent

bias due to improper analytical methodology is common in promi-

nent medical journals. J Clin Epidemiol 2004;57:672e82.