INTRODUCTION

Receiving ambulatory care from multiple providers is common.1 For example, 50% of Medicare beneficiaries nationally see 7 or more different physicians each year.1 This may be clinically appropriate, but it creates challenges, as providers often do not communicate with each other regarding their common patients.2,3 As a result, clinically relevant information, such as the results of tests ordered by another physician, is missing in 1 out of every 7 visits.4 More fragmented ambulatory care (that is, care spread across many providers without a dominant provider) has been associated with more testing,5 more procedures,6 more emergency department visits,7,8,9 and more hospitalizations,7,10 compared to less fragmented care. However, the relationship between fragmented ambulatory care and health outcomes is unclear.11

We previously found that the association between fragmented ambulatory care and subsequent healthcare utilization varies with patients’ health status.5,12 We hypothesize that the association between fragmented ambulatory care and health outcomes might also vary with health status. Healthy people might be able to see multiple providers without experiencing harm from gaps in communication, whereas sicker individuals might be more vulnerable to harm.

Prior to a myocardial infarction (MI), many patients have cardiovascular risk factors,13 such as hypertension and diabetes, which are often managed by multiple ambulatory providers (primary care physicians, cardiologists, endocrinologists, and others).14 We sought to determine whether more fragmented ambulatory care is associated with an increased risk of incident coronary heart disease (CHD) events, overall and by health status.

METHODS

Study Design, Population, and Data Sources

We conducted an ancillary study to the nationwide, prospective Reasons for Geographic and Racial Differences in Stroke (REGARDS) cohort study, using data from 2003 to 2016.15 Institutional review boards of the participating institutions approved the protocol. All participants provided written informed consent.

Between 2003 and 2007, 30,239 community-dwelling African-American and white adults ≥ 45 years old were enrolled in the REGARDS study, with oversampling of African-American adults and individuals living in the Southeastern U.S.15 Baseline data collection involved computer-assisted telephone interviews (CATI) and in-home visits with a physical examination, blood test, urine test, electrocardiogram (ECG), and medication inventory. Participants or their proxies are contacted by telephone every 6 months to detect study endpoints. The report of a potential event triggers adjudication, which involves an expert review of medical records, death certificates, proxy interviews, autopsy reports, the Social Security Death Index, and the National Death Index (NDI).16

For this study, we used REGARDS study baseline data, REGARDS-adjudicated events, and REGARDS-linked Medicare claims data.37 The most frequently missing variable was C-reactive protein (missing for 6%); 85% of the sample was included in the complete case analysis. Second, we restricted the analysis to those ≥ 65 years old at baseline, excluding those who aged into Medicare over the study period. Third, we changed the time period for outcomes from 3 to 6 months immediately following each exposure period. Fourth, we used all unique covariates (not including variables embedded in the definitions of other variables), instead of selecting covariates based on bivariate p values, with a complete case approach. Fifth, we used all unique covariates, with multiple imputation to handle missing covariate values. Sixth, we stratified the sample into 5 self-rated health groups instead of 3.

Analyses were conducted with SAS (version 9.4, Cary, NC) and Stata (version 14, StataCorp, College Station, TX). Multivariate p values < 0.05 were considered statistically significant.

RESULTS

Sample Characteristics

Of the REGARDS study participants, 10,566 met the inclusion criteria (Appendix 3). The mean age of participants was 70 years (Table 1, Appendix 4). More than half of participants (57%) were women, and 34% were African-American. Two-thirds (66%) had very good or good self-rated health. By design, one-fourth (25%) had high fragmentation. Differences in participant characteristics between those with high vs. low fragmentation in the first year of observation are shown in Table 1. Participant characteristics stratified by self-rated health are shown in Appendix 5.

Table 1 Baseline Characteristics, Overall and Stratified by Fragmentation Score in the First Year of Observation
Full size table

Ambulatory Utilization

Those with high fragmentation had an average of 10.7 visits across 6.3 providers, an average of 29% of visits with the most frequently seen provider, and an average fragmentation score of 0.91; by contrast, those with low fragmentation had an average of 8.8 visits across 3.5 providers, an average of 58% of visits with the most frequently seen provider, and an average fragmentation score of 0.63 (Table 2, p < 0.001 for each comparison).

Table 2 Ambulatory Utilization During the First Year of Observation, Overall and Stratified by Fragmentation Score and by Self-Rated Health
Full size table

Within each self-rated health group, those with high fragmentation had more visits, more providers, a lower proportion of visits with the most frequently seen provider, and higher fragmentation scores than those with low fragmentation (p < 0.001 for each comparison).

CHD Events

Participants were followed up for up to 11.8 years (median 5.5 years) (Table 3). Overall, 569 participants (5.4%) experienced an acute incident CHD event, equivalent to an unadjusted rate of 8.8 events per 1000 person-years. Those with worse self-rated health at baseline had more CHD events: 4.0% of those with excellent self-rated health, 5.4% of those with very good/good self-rated health, and 6.9% of those with fair/poor self-rated health had a CHD event (p = 0.001). Adjusted incidence rates are shown in Figure 1.

Table 3 Incident Rates of Acute Coronary Heart Disease Event, Stratified by Self-Rated General Health and by Fragmentation Status
Full size table
Figure 1
figure 1

Absolute adjusted rates of incident coronary heart disease (CHD) events, stratified by healthcare fragmentation status and self-rated general health. Adjusted rates were derived from Poisson models that adjusted for age, gender, race, marital status, educational attainment, income, region, hypertension, hyperlipidemia, diabetes, body mass index, albumin-to-creatinine ratio, C-reactive protein, number of medications, smoking, alcohol use, and lack of social support. Fragmentation status is a time-varying exposure based on the reversed Bice-Boxerman Index (low < 0.85, high ≥ 0.85). See “METHODS” for more details.

Full size image

Overall, the adjusted hazard ratio (HR) for the association between fragmentation and CHD events was 1.14 (95% confidence interval (CI) 0.92, 1.39) (Table 4, Appendix 6). However, there was an interaction between fragmentation and self-rated health (Wald test: p < 0.01). Among those with excellent self-rated health, there was no association between fragmentation and CHD events (adjusted HR 0.91; 95% CI 0.53, 1.57; p = 0.74). Among those with very good or good self-rated health, high fragmentation was associated with an increased hazard of CHD events (adjusted HR 1.35; 95% CI 1.06, 1.73; p = 0.01), compared to low fragmentation. Among those with fair or poor self-rated health, high fragmentation was associated with a trend toward a decreased hazard of CHD events (adjusted HR 0.54; 95% CI 0.29, 1.01; p = 0.052).

Table 4 Association Between Healthcare Fragmentation and Incident Coronary Heart Disease (CHD) Events, Overall and Stratified by Self-Rated General Health (N = 10,556)
Full size table

Results persisted in the sensitivity analyses that used the base case set of covariates with multiple imputation, the 6-month outcome period, all covariates with a complete case approach, and all covariates with a multiple imputation approach (Appendix 7). In the sensitivity analysis that excluded participants < 65 years old at baseline, the results for excellent and very good/good self-rated health persisted, and the results for fair/poor self-rated health reached statistical significance (adjusted HR 0.42; 95% CI 0.20, 0.91). In the sensitivity analysis stratifying self-rated health into 5 groups, confidence intervals were wider and high fragmentation was associated with a decreased hazard of CHD events in the fair self-rated health group (adjusted HR 0.43; 95% CI 0.20, 0.92, Appendix 8).

DISCUSSION

In this 12-year study of 10,556 Medicare beneficiaries from the REGARDS cohort, the association between fragmented ambulatory care and incident CHD events varied with health status. Among those with very good or good self-rated health, having highly fragmented care was associated with an increased adjusted hazard of a CHD event in the 3 months following the exposure. Among those with fair or poor self-rated health, having highly fragmented care was associated with a decreased adjusted hazard a CHD event over the same time period. Among those with excellent self-rated health, there was no association between fragmentation and CHD events. The ambulatory care patterns underlying high vs. low fragmentation were clinically and statistically different from each other: those with high fragmentation had an average of 11 visits to 6 providers per year, with the most frequently seen provider accounting for 29% of the visits, whereas those with low fragmentation had an average of 9 visits to 4 providers per year, with the most frequently seen provider accounting for 58% of visits.

These results are concordant with our previous study of Medicare beneficiaries in New York State, which found that healthcare fragmentation was not associated with hospitalizations among those with 0 chronic conditions, was associated with an increased hazard of hospitalization among those with 1–4 chronic conditions, and was associated with a decreased hazard of hospitalization among those with ≥ 5 chronic conditions.12 Other studies of fragmentation have adjusted for burden of illness (rather than stratify), precluding additional comparisons.6,7,10

The exact mechanisms by which fragmentation could increase or decrease cardiac risk, depending on health status, are not clear. On the one hand, sicker patients may require more providers,1 and providers may make more of an effort to coordinate care for sick patients,39 thereby ameliorating the risk of gaps in communication. Sicker patients may also have access to more ancillary providers like care managers, who may facilitate care coordination.40 On the other hand, for patients with a moderate burden of illness, more fragmented care may lead to worse control of cardiac risk factors, through less aggressive management of those risk factors or discontinuation of beneficial medications.41,42,43,44 Multiple providers may also unknowingly prescribe drugs that interact, exacerbating cardiac risk factors.45,46,47 More work is needed to distinguish between beneficial and deleterious care fragmentation.

Strengths of this study include the national sampling frame, large sample size, long time horizon, homogeneity of insurance benefits across study participants through Medicare, use of a previously validated measure of fragmentation, treatment of fragmentation as a time-varying exposure, adjudicated CHD events, and adjustment for clinically detailed potential confounders.

This study has several limitations. First, it is observational; we cannot rule out residual confounding or infer causation. Second, we did not have data on cardiac risk factor control over time, so we cannot directly test the hypothesized mechanisms above. Third, this study did not measure communication between providers, so the presence of fragmentation cannot be interpreted as the definite absence of communication. Fourth, self-rated health was measured at baseline and may change over time. However, self-rated health at baseline was a significant predictor of CHD events. Fifth, we were not able to account for whether physicians were in the same practice; if physicians practicing together had access to the same medical records, buffering potential harm from fragmentation, this would have biased our study toward the null.

In conclusion, among Medicare beneficiaries with very good or good self-rated health (which was two-thirds of the sample), highly fragmented ambulatory care was associated with an independent increase in the short-term risk of an incident CHD event. More research is needed to determine exactly how fragmentation contributed to this risk and to determine whether interventions are needed to reduce unnecessary fragmentation of care.