|Home | About | Journals | Submit | Contact Us | Français|
This article presents an interpersonal continuity of care measure.
We operationalized continuity of care as no more than an 8-month interval between any two visits during a 2-year period to either (a) the same primary care physician or (b) the same physician regardless of specialty. Sensitivity analyses evaluated two interval censoring algorithms and two alternative intervals. We linked Medicare Part A and B claims to baseline survey data for 4,596 respondents to the Survey on Asset and Health Dynamics Among the Oldest Old. We addressed the potential for selection bias by using propensity score methods, and we explored construct validity.
Interpersonal continuity with a primary care physician was 17.3%, and interpersonal continuity of care with any physician was 26.1%. Older participants; men; individuals who lived alone; people who had difficulty walking; and respondents with medical histories of arthritis, cancer, diabetes, heart conditions, hypertension, and stroke were most likely to have continuity. Individuals who had never married, were widowed, were working, or had low subjective life expectancy were least likely to have continuity.
Researchers can measure interpersonal continuity of care using Medicare Part B claims. Replication of these findings and further construct validation, however, are needed prior to widespread adoption of this method.
Experts have regarded continuity of care as a major component of good medical care for more than seven decades (Lee & Jones, 1933; Saultz, 2000). Simply put, the underlying assumption is that continuity facilitates interaction and communication between the patient and his or her physician, which helps develop mutual trust, stimulates the flow of information between them, bolsters the accuracy of the physician’s diagnoses, enhances selection of the most effective treatments, and increases patient compliance and adherence (Breslau & Haug, 1977). An extensive, recent literature review (Saultz, 2003) underscored the durability of continuity of care as a key concept. And a recent multinational study reaffirmed its importance to primary care physicians (Stokes et al., 2005).
It is not surprising, then, that in 1996 the Institute of Medicine (IOM) identified continuity of care as a core attribute of primary care because it should result in better quality care and disease management, especially for older adults with one or more chronic conditions. More recently, however, the IOM elevated continuity of care to the status of a primary aim in its comprehensive call for national action to transform health care quality. Specifically, that aim is
To establish and support a continuous healing relationship, enabled by an integrated clinical environment and characterized by the proactive delivery of evidence-based care and follow-up. (IOM, 2003, p. 49)
In that same paragraph, the IOM explicitly recognized Shortell, Gillies, and Anderson’s (2000, p. 129) definition of clinical integration as
… the extent to which patient care services are coordinated across people, functions, activities, and sites over time so as to maximize the value of services delivered to patients.
This is not just an IOM or even an American point of view. Writing in the British Medical Journal, Haggerty and colleagues (2003, p. 1219) noted that “policy reports and charters worldwide urge a concerted effort to enhance continuity.”
But what does continuity of care really mean? The answer to that question is not at all straightforward. To be sure, scholars have discussed and debated the conceptualization of continuity of care for decades (Christakis et al., 2004; Dietrich & Marton, 1982; Ettner, 1996, 1999; Haggerty et al., 2003; Hennen, 1975; Lambrew, DeFriese, Carey, Ricketts, & Biddle, 1996; Mainous, Baker, Love, Pereira Gray, & Gill, 2001; Mainous & Gill, 1998; Rogers & Curtis, 1980; Saultz, 2000, 2003; Saultz & Lochner, 2005; Shortell, 1977; Starfield, 1980; Steinwachs, 1979; Weiss & Blustein, 1996). Consensus has emerged, however, only on the multiple benefits that should accrue from continuity and on the multidimensionality of the concept itself. The expected benefits include improved doctor–patient relationships; enhanced physician knowledge of the patient; greater rapport and disclosure; increased compliance; reduced hospitalization rates; increased patient and physician satisfaction; reduced disability levels, costs, and number of missed appointments; and improved problem recognition and management (Rogers & Curtis, 1980). The major dimensions of continuity of care include informational, longitudinal, interpersonal, geographic, team-based, and familial (Saultz, 2003). This, however, is where the consensus over continuity of care ends (Haggerty et al., 2003).
Indeed, in his extensive review Saultz (2003) identified no fewer than 21 instruments designed to measure continuity of care, several of which he noted had never actually been used. Saultz placed these instruments into three categories: (a) those that do not require an assigned (i.e., primary) provider, (b) those that do require an assigned provider, and (c) those that measure continuity of care at the family level. In general, measurement instruments in the first category rely on concentration statistics, including Gini coefficients (Shortell, 1977; Steinwachs, 1979), Herfindahl indices (Eriksson & Mattsson, 1983; Smedby, Eklund, Eriksson, & Smedby, 1986), and related fractionals (Roos, Roos, Gilbert, & Nicol, 1980). These are designed for use with administrative data, although they can also be used with self-report or survey data that provide the number of visits to each of the various physicians seen during the target interval. The major drawback of these instruments is that, by focusing on the largest fractional, they ignore the medical specialty of the physician and assume that the physician seen most often is, in fact, the primary (in the sense of usual) physician.
The next category of continuity of care measures is remarkably similar, except that these measures require the identification of a primary physician. Various themes within this category address fractionals among the identified provider (Breslau & Haug, 1977; Smith, 1995), the duration of the relationship with the usual physician (Weiss & Blustein, 1996), and the rate of provider turnover (Merenstein et al., 2001). Another subset within this category involves patient identification of the usual physician, which requires moving beyond sole reliance on administrative data (Ettner, 1996, 1999). Although these measures do require the identification of a usual provider, they generally do not impose restrictions on who that usual provider can be or how long the patient–practitioner relationship should be in existence, and their algorithms are not theoretically motivated.
The last category of continuity of care measures is specific to family practice. These instruments focus on the extent to which the family unit receives continuity of care from the same family physician. Such measures focus either on a global index (Murata, 1993) or on the average continuity of care in the family with the same provider (Godkin & Rice, 1984). As interesting as these family-based measures may be, they have little applicability outside of family practice, especially for older adults.
This brief review should make two things clear. First, continuity of care is presumably very important, especially for older adults. Second, no theoretically driven measure exists for the standard assessment of continuity of care. This is especially problematic given the increasing availability and use of administrative data for evaluating the quality of health care and disease management (IOM, 2003). Accordingly, we report here on the conceptualization, operationalization, and initial validation of an interpersonal continuity of care measure for use with Medicare Part B claims data.
Our conceptualization is based on, and consistent with, those of Steinwachs (1979), Saultz (2003), and Haggerty and colleagues (2003). Steinwachs identified four factors for consideration: time period (fixed or episodic), organizational context (medical specialty and site), provider (individual or team), and problem type (selected diagnosis or all care). Saultz (2003) identified a hierarchical definition progressing from informational continuity (when medical records are accessible by any health care team member), to longitudinal continuity (when a regular source [“medical home”] exists for each patient), and ultimately to interpersonal continuity (when the patient uses a particular physician for all basic health services). Haggerty and colleagues went to the heart of the issue by noting that continuity of care involves
… the relationship between a single practitioner and a patient that extends beyond specific episodes of illness or disease … and implies a sense of affiliation between patients and their practitioners (my doctor or my patient), often expressed in terms of an implicit contract of loyalty by the patient and clinical responsibility by the provider. (p. 1219)
Simply put, our conceptualization assumed that continuity of care involves an ongoing relationship with a particular physician in the outpatient setting with sufficient frequency for that physician to assume primary responsibility for both the patient’s basic health care needs and his or her overall disease and care management.
That having been said, it is important to note here a salient limitation of our approach. The operational definition of continuity of care that we advance in this article does not require evidence that the physician actually assumed primary responsibility for his or her patient. Because we developed our measure to rely solely on the information contained in Medicare claims, evidence of the assumption of that primary responsibility was simply not available. Thus, although our measure is consistent with the heart of the continuity of care issue as described by Haggerty and colleagues (2003), it only requires that there were a sufficient frequency of visits to the same physician for this to have occurred.
We took data from the Survey on Assets and Health Dynamics Among the Oldest Old (AHEAD; Myers, Juster, & Suzman, 1997). AHEAD organizers identified respondents from either household screening conducted during the 1992 multistage cluster sampling process for the companion Health and Retirement Study (Juster & Suzman, 1995) of preretirement-aged adults, or a supplemental sample of persons 80 years old or older identified from the Centers for Medicare and Medicaid Services Medicare Master Enrollment File. Organizers oversampled African Americans, Hispanics, and Floridians. Thus, the data are weighted to adjust for the unequal probabilities of selection due either to the multistage cluster sampling design and/or the oversampling.
Researchers conducted baseline AHEAD in-home interviews in 1993 with 7,447 respondents who were 70 years old or older. The response rate was 80.4%. Complete linkage to Medicare Part A and B claims, as well as to the National Death Index was accomplished for 4,697 participants (63%). Of these, we excluded 101 respondents who had evidence of having been in Medicare managed care during the 2-year pre-baseline period, because managed care plans are not required to report complete data. Thus, our analytic sample involved 4,596 men and women (62% of the original AHEAD cohort) for whom complete data were available. We evaluated the potential for selection bias into the analytic sample by using propensity score methods, specifically incorporating the predicted probability of inclusion as a covariate in our models (D’Agostino, 1998; Rosenbaum & Rubin, 1983; Rubin, 1979).
Translating our conceptualization into a measurement metric for use in Medicare Part B claims involved two phases. First, it required defining a visit. Whereas the analysis of Medicare Part A data is generally straightforward (there is one record per hospital episode), the analysis of Part B (outpatient) data is more complicated. Part B data are structured as “lines” (billable goods, services, or procedures) performed under (within) a specific claim. To restrict our measure to the outpatient setting, we first deleted all inpatient-related line items and claims (i.e., hospital, hospital discharge, hospital consultation, nursing facility, and care plan oversight services). We then deleted all Part B claims for which the “from and through” (service start and stop) dates completely overlapped with Part A hospital stays (admission to discharge dates), with one exception: We included physician claims that had occurred on the day of admission, because these most likely reflected outpatient encounters during which the decision to hospitalize was made. To restrict the measure to physician services provided directly to patients (analogous to what a patient would report as a physician visit), we deleted line items or claims that were not primary care or specialty care (i.e., anesthesiology, drugs, supplies, radiology, labs, pathology). To further ensure that the patients had actually been seen (i.e., that visits had occurred), we then deleted line items and claims without evaluation and management codes, recognizing that this may have resulted in underestimation. After we had deleted duplicates (same day, same provider), all remaining line items with evaluation and management codes qualified as visits.
The second phase in operationalizing our definition of continuity of care involved specifying selections reflecting each of Steinwachs’s (1979) four factors. For time period, there were two choices: the normative expectation threshold and the length of the longitudinal period. Because there are no established guidelines for the former (Wallace, 1998), we consulted several general internists, geriatricians, and gerontological health services researchers to secure face validity (Shadish, Cook, & Campbell, 2002). The general consensus was that for older adults, the majority of whom have several chronic conditions, semiannual visits were the desirable minimum. To ensure longitudinal continuity, we stipulated a 2-year window. Recognizing that vacations and scheduling difficulties require some buffering of what might otherwise be an arbitrary rule, we set the threshold for continuity of care to at least one visit every 8 months over the 2-year window.
For Steinwachs’s organizational context, we chose two approaches. The first (i.e., interpersonal continuity of care with a primary care physician) focused on a standard definition of primary care based on the listed specialty for the UPIN (unique physician identifier number) denoted on each visit. This included general practice, family practice, geriatrician, internist, preventive medicine, multispecialty clinic or group practice (who were not classified to any medical specialty and who satisfied the continuity of care definition for only 26 participants), nurse practitioner, certified clinical nurse specialist, and physician assistant. The second approach (i.e., interpersonal continuity of care with any physician) was more encompassing and defined continuity of care based on visits to the same physician (UPIN) regardless of specialty.
In terms of Steinwachs’s provider factor, we chose the individual physician to ensure interpersonal continuity of care. That is, to meet the normative expectation threshold of semi-annual visits (buffered to include 8 months) over the 2-year period, the qualifying visits had to have been with the same physician (i.e., UPIN). Finally, we chose not to restrict our definition to a selected diagnosis but to health care in general. Thus, we operationalized the continuity of care measure as no more than an 8-month interval between any two visits to the same primary care physician (or to the same physician regardless of specialty) during the 2-year time frame.
One remaining operational issue involved the potential for censoring. We defined continuity based on the 24-month period immediately preceding each participant’s baseline interview. Strictly speaking, the lengths of the intervals at the beginning and end of that 2-year period would have been censored if we did not have access to claims data for each respondent for up to 8 months before and for up to 8 months after that 24-month period. That is, for the beginning and ending intervals, we would not have been able to determine whether those censored intervals would have been less than 8 months. Because we had data on all participants for at least 8 months before and after the 24-month period, we were able to avoid the censoring issue altogether.
Other researchers, however, might not have access to the 40 months of claims data (24 +8 +8) necessary to fully resolve the potential for censoring. Therefore, we conducted sensitivity analyses to provide insight into how the censoring problem might be addressed if only 24 months of claims data were available. Specifically, we repeated the analyses two ways, both of which used only the 24 months of claims data prior to baseline. The first ignored the censoring issue altogether if the observable lengths of the two censored intervals for that participant were each less than 8 months. That is, we treated these censored intervals as being consistent with continuity of care (i.e., “censored in”). In the second approach, we ignored the censored intervals only if the observable lengths of the two censored intervals for that respondent were each less than 8 months (as above) and neither one was the longest interval for that respondent. If either of the censored intervals were the longest interval for a given participant, then we coded that person’s continuity of care as zero (i.e., as no continuity, or “censored out”).
We further evaluated the validity of our continuity of care measure by using traditional methods (Shadish et al., 2002). Ideally, this would have involved corroborating the validity of our measure by correlating it with a known gold standard for continuity of care. Unfortunately, no such criterion variable exists. In its absence, the next best thing would have been to corroborate the validity of our measure by correlating it with related measures, such as the reports of patients and their providers, especially with respect to whether continuity of care exists, with which physicians and patients it exists, and both the patient’s and physician’s sense of the quality of (or their satisfaction with) their specific patient–practitioner relationship. Lamentably, AHEAD does not contain any such information. Indeed, interviewers did not even ask respondents if they had a regular physician or a usual source of health care, or if they could name the physician seen most often.
We were forced, therefore, to rely on a more indirect approach. In it, we used multiple logistic regression analysis (Hanley & McNeil, 1982; Hosmer & Lemeshow, 1989) to identify associations between participant characteristics obtained at baseline and continuity of care during the 24 months prior to baseline. We based the selection of predictor variables on previous speculations or findings in the literature (Saultz, 2003; Steinwachs, 1979). We organized those variables into sociodemographic, social support, socioeconomic, health lifestyle, morbidity, and health services use categories.
Sociodemographic factors were age, gender, race, marital status, and life expectancy. Social support measures included living arrangements, proxy status, presence of living children, and access to future helpers. Among the socioeconomic factors were employment status, insurance status, income, and veteran status. We used alcohol consumption, smoking, and weight to tap health lifestyle behavior. Morbidity was indicated by self-rated health; difficulty walking; and reporting of arthritis, cancer, diabetes, heart condition, hypertension, stroke, or psychological conditions. The sole measure of prior health services use was whether the respondent reported having been hospitalized in the past year.
Based on the literature, our general expectation was that older adults; men; and respondents who lived alone, had better socioeconomic status and care access, exhibited more healthy lifestyles, or had chronic conditions were most likely to have satisfied the continuity of care threshold. That is, we assumed that the prevalence of continuity of care would be higher in respondents who had these traits than in respondents who did not have them. Because of the large number of variables that were considered, the final multiple regression models retained only those that had achieved statistical significance at one stage or another of several standard model-building approaches (Concato, Feinstein, & Holford, 1993; Concato, Peduzzi, Holford, & Feinstein, 1995; Harrell, Lee, & Mark, 1996; Peduzzi, Concato, Feinstein, & Holford, 1995).
Among the 4,596 AHEAD respondents in the analytic sample, the mean age was 77.5 years, 36.3% were men, 12.4% were African American, 6.2% were Hispanic, 3.5% had never married, 4.7% were separated or divorced, 42.2% were widowed, 37.1% lived alone, and 87.0% had living children. Only one fifth of the participants (21.8%) expressed a 50% or greater expectation for living at least 10 more years. Additionally, 8% worked for pay, 8.1% had Medicaid, 45.6% reported having low incomes, and 18.5% lived in cities of 1 million or more people. Half of these AHEAD respondents (51.2%) had a history of smoking, and 12.7% indicated that they felt the need to cut down on their alcohol consumption. Having fair or poor health was reported by 35.2%, 38.0% had difficulty walking several blocks, 25.2% reported having arthritis, 12.9% reported having cancer, 11.7% reported having diabetes, 28.8% reported having a heart condition, 46.3% reported having hypertension, 9.7% reported having had a stroke, and 7.6% reported having psychological problems. The mean number of self-reported physician visits in the year prior to baseline was 5.0, and 22.9% of respondents reported having been hospitalized during that year. During the 2-year pre-baseline period, the Medicare Part B files contained 123,699 claims (M = 13.5 per person-year) and 236,005 lines (M = 25.7 per person-year). Using our approach (see “Operationalization”), this translated to 54,509 physician visits (M = 5.9 per person-year).
To address the potential for selection bias, we used multiple logistic regression analysis to model inclusion in the analytic sample. In this analysis, we coded the dependent variable 1 if the participant was among the 4,596 persons included in the analysis and 0 if he or she was among the 2,851 not included. We considered all of the previously mentioned covariates and several other baseline characteristics, as the goal of propensity score models is to attain the best statistical fit to the data (D’Agostino, 1998; Rosenbaum & Rubin, 1983; Rubin, 1979). Despite the large number of predictive factors, model fit was modest (C statistic =0.622), although there was no evidence of heteroscedastic error (Hosmer–Lemeshow statistic p = .314).
Because the purpose of the propensity score model was to obtain the predicted probabilities of inclusion in the analytic sample for use as a covariate in the construct validity analyses, we do not present the detailed results here (they are available on request). Briefly, men, individuals who lived in major metropolitan areas, participants for whom the baseline survey data had been obtained by proxy, and persons who had felt unable or been unwilling to estimate their chances of surviving 10 or more years were least likely to be included in the analytic sample (all ps < .01). In contrast, participants who had rated the quality of their neighborhoods as poor or who had reported that religion was very important were most likely to be included in the analytic sample (both ps < .01).
Figures 1 and and22 display the cumulative percentage distributions of the longest interval between visits during the 24-month period prior to baseline that were to the same primary care physician (Figure 1) or to the same physician regardless of specialty (Figure 2). Even using 40 months of data, we could not fully adjudicate intervals beyond 8 months; therefore, Figures 1 and and22 are based on the censored-out algorithm. Given our definition, these intervals may have ranged from 1 day (when the participant saw the same physician each and every day for 24 months) to 732 days (when the participant never saw the same physician twice during the 2-year period). To illustrate what the prevalence of interpersonal continuity of care would be if the 8-month maximum interval specified in our definition were altered, we have added vertical broken lines connecting to the curves at 4, 8, 12, and 16 months and have indicated the corresponding cumulative percentages.
As shown in the figures, using the 8-month maximum interval specified in our definition, the prevalence of interpersonal continuity with a primary care physician was 17.3%, and the prevalence of interpersonal continuity of care with a physician regardless of specialty was 26.1%. When we considered 4- to 16-month interval alternatives, the prevalence of interpersonal continuity of care with a primary care physician ranged from 4.3% to 26.4%, and the prevalence of interpersonal continuity of care with a physician regardless of specialty ranged from 6.0% to 36.4%. Note that for both of the continuity of care measures, there were substantially diminished returns to scale once the maximum monthly interval reached 12 or more months (i.e., both plateaued). And for both of the continuity of care measures, reliance on anything less than a 6-month maximum interval identified extremely (i.e., unrealistically) low (i.e., 10–12%) prevalence rates. Thus, from a data standpoint, the optimal maximum monthly interval appeared to lay between 6 to 12 months.
Table 1 contains the adjusted odds ratios obtained from the multiple logistic regression analyses of six versions of the continuity of care measures. Columns 1 through 3 reflect defining continuity on the basis of visits to the same primary care physician. Our primary results appear in column 1, where the lengths of the earliest and latest intervals in the 2-year period were fully adjudicated by using the claims for the prior and subsequent 8 months (i.e., 40 months of data). The results from the sensitivity analyses reflecting the two approaches for dealing with censored lengths for the earliest and latest intervals (i.e., 24 months of data) are in columns 2 (censored in) and 3 (censored out). Columns 4 through 6 reflect comparable models when continuity of care was defined on the basis of visits to the same physician (i.e., UPIN), excluding only pathologists and radiologists. We note here that some readers might prefer to see a comparable table containing crude odds ratios instead of adjusted odds ratios. Such a table is available on request from Fredric D. Wolinsky. The main pattern of difference between the two tables is that (a) the risks or protections reflected in the crude odds ratios are generally larger than those associated with the adjusted odds ratios due to confounding, and thus (b) statistical significance is more likely to be achieved among the crude odds ratios. We focus here on the adjusted odds ratios shown in Table 1 because these represent net relationships from which known confounder bias has been eliminated.
The variation between the data shown in columns 1 and 4 results from definitional differences. In column 4 we relaxed the medical specialty criteria for the physician, which increased statistical power (i.e., shrank confidence intervals) and enhanced the effect size (i.e., increased point estimates) for several risk factors. Most notably, when the medical specialty criteria were relaxed, men; participants who reported having cancer, heart conditions, or a stroke; and persons who lived alone were substantially more likely to have continuity of care, with the increased odds ranging from 28.8% to 43.1%. In each case, the effect size was larger and statistically significant compared to the results from the more restrictive definition. In contrast, there were three characteristics (being widowed, working, and reporting low income, which all reduced the likelihood of continuity of care) for which the effect size was equivalent but statistical significance was achieved due to the increased power. In addition, both models indicated that the likelihood of having continuity increased with age; with having difficulty walking; or with having arthritis, diabetes, or hypertension. The likelihood of having continuity decreased for persons who reported greater subjective life expectancy.
As indicated with respect to Figures 1 and and2,2, the optimal maximum monthly interval from an empirical standpoint lies between 6 to 12 months. Although our theoretically driven 8-month interval falls almost at the middle of that range, the interested reader may wonder whether the results shown in Table 1 would differ if based on 6- or 12-month intervals. To address this possibility, we repeated the analyses shown in columns 1 and 4 of Table 1 using 6- and 12-month intervals under the censored-out algorithm (because we did not have sufficient pre-baseline data for all AHEAD respondents to fully adjudicate the 12-month interval, and because the censored-out algorithm most closely approximates the results from full adjudication; see below). Those results (not shown, but available on request) were remarkably consistent. Indeed, both the identification of factors statistically significantly associated with continuity care, as well as the effect sizes estimated for those associations, were very similar to those shown in Table 1. This enhanced our confidence in the 8-month interval that we chose based on theoretical (normative) reasons.
Columns 2, 3, 5, and 6 of Table 1 contain the results obtained when the two approaches for dealing with the censored intervals were used. As shown, the censored-out approach (which ignored the censored intervals only if the observable lengths of the two censored intervals for that participant were each less than 8 months and neither one was the longest interval for that participant; columns 3 and 6) yielded prevalence estimates and adjusted odds ratios remarkably similar to those obtained with full adjudication (based on 40 months of data). In contrast, the censored-in approach (which ignored the censoring issue altogether if the observable lengths of the two censored intervals for that participant were each less than 8 months; columns 2 and 5) substantially overestimated prevalence rates and misidentified variables significantly associated with continuity of care.
The purpose of this article was to report on the conceptualization, operationalization, and initial validation of a measure of interpersonal continuity of care for use with Medicare Part B claims. Conceptually, we assumed that continuity required an ongoing relationship with a particular physician in the outpatient setting of sufficient frequency for that physician to assume primary responsibility for both the patient’s basic health care needs and overall disease and care management. We operationalized the continuity of care measure as no more than an 8-month interval between any two visits to the same physician during the 24 months prior to baseline. We considered two variants: one that required the physician to be in primary care and one that did not.
Based on this approach, we estimated prevalence rates of 17.3% for interpersonal continuity of care with a primary care physician and 26.1% for interpersonal continuity of care with any physician. Because many older adults routinely see non-primary care physicians (especially cardiologists, rheumatologists, and pulmonologists) for the management of their chronic conditions as their regular source of health care, the second of our definitions (which relaxes the constraints on physician specialty) is preferred, and we recommend using it. Placing these prevalence estimates in context, however, is not entirely straightforward. On the one hand, the IOM (1996) has declared continuity of care to be a key component of primary care and has made continuity of care a primary aim of its comprehensive call for national action to transform health care quality (IOM, 2003). Given that similar calls for the centrality of continuity of care in health care reform have been advocated worldwide (Haggerty et al., 2003), one might have expected, or at least desired, substantially larger prevalence estimates.
On the other hand, the evidence on the extent to which fragmentation has permeated health care delivery in America is extremely convincing (Shortell et al., 2000; Starr, 1982, 2004). Moreover, the latent effects of current reimbursement policy are so problematic and far-reaching that the American College of Physicians (ACP; 2006) recently called for revolutionary reforms to restructure health care delivery in America to restore fiduciary agency and professionalism (Parsons, 1951; Wolinsky, 1993). Thus, the prevalence rates estimated in this study may well be accurate reflections of how much American medicine has changed in the last half century (Hafferty & McKinlay, 1993; Starr, 1982, 2004).
Furthermore, we are unaware of any directly comparable normative data. Self-reports from the 2003 National Health Interview Survey (National Center for Health Statistics, 2005) indicated that 83.5% of adults had a usual source of care, and self-reports from the 1995 Behavioral Risk Factor Surveillance System (Centers for Disease Control and Prevention, 1996) indicated 85.7% of adults had a regular source of care. Although these two data sources are remarkably consistent with each other, researchers cannot consider them to be benchmarks for direct comparison. The reason is that the National Health Interview Survey and Behavioral Risk Factor Surveillance System define usual or regular source based solely on the place where the care is received and do not consider whether the same provider was seen, what the provider’s medical specialty was, or whether the provider was seen with sufficient frequency to assume primary responsibility for the patient’s basic health care and disease management.
In the absence of established guidelines (Wallace, 1998), we secured the face validity (Shadish et al., 2002) of the maximum interval between physician visits on which our continuity of care measure is based by consensus building among several general internists, geriatricians, and gerontological health services researchers. Face validity, however, is relatively weak by itself. Therefore, we examined the construct validity (Shadish et al., 2002) of our measure using multiple regression analyses. As expected, older participants; men; persons who lived alone; respondents who had difficulty walking; and persons who had medical histories of arthritis, cancer, diabetes, heart conditions, hypertension, and stroke were more likely to have continuity. Individuals who had never married, were widowed, were working, and had low subjective life expectancy were less likely to have continuity. These results enhance our confidence in the measure’s validity.
Some additional evidence can be brought to bear on the construct validity of our preferred continuity of care measure (which allows the regular doctor to be either a primary care physician or a medical specialist) by reaching forward to the first follow-up interviews. Interviewers asked AHEAD respondents during those 1995/1996 re-interviews if they had had cholesterol screening or flu shots during the past year. Among respondents who had continuity of care, 67.9% reported having had cholesterol screening and 64.9% reported having had flu shots versus 60.2% and 57.0%, respectively, among those who did not (ps < .0001). These results further increase our confidence in the measure’s validity.
Nonetheless, additional construct validation is needed before our measure of continuity of care should be recommended for widespread adoption. Because no gold standard for continuity of care exists, our measure needs to be examined using Medicare Part B claims that can be correlated with the reports of patients and their providers about whether continuity of care exists, with which physician each patient has continuity of care, and both the patient’s and the physician’s senses of the quality of (and satisfaction with) the target patient–practitioner relationship. We have been able to identify only one data set to date that holds considerable promise (Wei, Morgan, & Virnig, 2006). It already contains extensive survey data about the patient–practitioner relationship (e.g., access, quality, having a regular care place and duration of its use, having a regular provider and length of that relationship, and satisfaction with overall health care) from a large, representative sample of Medicare beneficiaries that could be (but has not yet been) linked to their Part B claims. We are entering into a collaborative relationship with these investigators and their sponsors to both replicate the continuity of care prevalence rates estimated here and conduct more rigorous construct validation. We cannot, however, initiate those analyses until the appropriate modifications to their original Centers for Medicare and Medicaid Services Data Use Agreement have been approved and other logistical arrangements have been made.
Even after that, and assuming that we find support for the construct validation of our measure, our measure will not be without limitations. The most important of these is that our measure relies solely on Medicare Part B claims. As such, it is, at best, an indirect measure of continuity of care. Specifically, it indicates only that there have been a sufficient number of visits within the specified time period to the same physician for that physician to have been able to assume primary responsibility for that patient. Our measure cannot, however, say that such fiduciary agency actually occurred. Moreover, the additional data for construct validation that the collaborative study (Wei et al., 2006) will provide do not contain any evaluation of the patient–practitioner relationship from the physician’s perspective.
These caveats notwithstanding, our continuity of care measure has considerable promise for gerontology, geriatrics, and health care policy. As stated at the outset of this article, it is assumed that continuity of care results in better quality health care and disease management, especially for older adults with one or more chronic conditions (IOM, 1996). Indeed, this is why the IOM (2003) has made continuity of care a primary aim of its comprehensive call for national action to transform health care quality, and why the ACP (2006) has made continuity of care a centerpiece of its call for revolutionary reforms to restructure American health care delivery. At the core of both the IOM and ACP proposals is the “medical home” to be established for each patient by a primary care physician who provides the coordinating function for the health maintenance and health care of patients with multiple chronic conditions, as well as the appropriate reimbursement to the primary care physician for that coordinating function. Although the reasoning behind the IOM and ACP proposals is compelling, the evidentiary base is underwhelming for such sweeping initiatives.
To a considerable degree, the paucity of the evidentiary base is the legacy of not having a standard measure for the assessment of continuity of care available for use with administrative data. We designed our measure for use with claims data, and we encourage future researchers to use it to evaluate the associations between continuity of care and (a) health status trajectories (e.g., self-rated health, activities of daily living, instrumental activities of daily living, and lower extremity limitations), (b) health-related quality of life, (c) patient and physician satisfaction, (d) the use of preventive health services, (e) hospitalization for ambulatory care sensitive conditions (Agency for Healthcare Research and Quality, 2002), and (f) mortality. If that research demonstrates that continuity of care improves health status trajectories, improves health-related quality of life, increases patient and physician satisfaction, increases the use of preventive health services, minimizes hospitalizations for ambulatory care sensitive conditions, and delays mortality, then implementation of revolutionary reforms to restructure American health care delivery like those proposed by the ACP should be encouraged and supported.
This work was supported by National Institutes of Health Grants R01 AG-022913 and R03 AG027741 to Dr. Wolinsky. Dr. Wolinsky is associate director and Dr. Rosenthal is director of the Center for Research in the Implementation of Innovative Strategies in Practice at the Iowa City Veterans Affairs Medical Center, which is funded through the U.S. Department of Veterans Affairs, Veterans Health Administration, Health Services Research and Development Service (HFP 04-149). The opinions expressed here are our own and do not necessarily reflect those of any of the funding, academic, or governmental institutions involved.