Patient social risk factors and continuity of care for Medicare beneficiaries

Abstract

Objective

To identify patient social risk factors associated with Continuity of Care (COC) index.

Data Sources/Study Setting

Medicare Current Beneficiary Survey (MCBS), the Dartmouth Institute, and Area Resource File for 2006‐2013.

Study Design

We use regression methods to assess the effect of patient social risk factors on COC after adjusting for medical complexity. In secondary analyses, we assess the effect of social risk factors on annual utilization of physicians and specialists for evaluation and management (E&M).

Data Collection/Extraction Methods

We retrospectively identified 59 499 patient years for Medicare beneficiaries with one year of enrollment and three or more E&M visits.

Principal Findings

After adjustment for medical complexity, individual‐level social risk factors such as lack of education, low income, and living alone are all associated with better patient COC (P < .05). Similarly, area‐level social risk factors such as living in areas that are nonurban or high poverty, as well as in areas with low specialist or high primary care physician supply, are all associated with better patient COC (P < .05). We found the opposite pattern of associations between these same risk factors and annual patient utilization of physicians and specialists (P < .05).

Conclusions

Medicare patients with multiple social risk factors have consistently better COC; these same social risk factors are associated with reduced patient‐realized access to specialist physician care.

1. INTRODUCTION

Numerous studies in recent years assess the effect of patients′ continuity of care (COC) with their physicians (patient COC) on patient outcomes and medical costs.1, 2, 3, 4, 5, 6, 7, 8, 9, 10 Most of these studies have found that a greater degree of patient COC is associated with better outcomes, such as lower costs, fewer complications of disease, and fewer emergency department (ED) visits or hospitalizations.11 However, there is a lack of research examining the relationship between patient social risk factors and patient COC. Understanding this relationship is important for clinical practice and policy in an era of value‐based payment (VBP), especially in Medicare, where programs such as the Merit‐based Incentive Payment System and the Medicare Shared Savings Program are dependent on the concept of patient COC in the way they hold providers accountable12, 13 and prior research has demonstrated that patients with greater social risk factors have worse outcomes under Medicare VBP.14, 15, 16, 17 In addition, if patient COC is indeed good for patients′ health, then it is important to understand the role social risk factors play in how patients experience COC and access physicians for ongoing evaluation and management (E&M) office visits.

Furthermore, most studies of the effect of COC on patient outcomes have been limited to medical claims data.1, 2, 3, 5, 6, 7 However, it is plausible that there are patient‐level social risk factors not observable in claims data that are associated with COC. It is important to know what these factors are because to the extent they are also correlated with patient outcomes and then prior studies that did not consider these factors are likely to produce biased results.

In this study, we focus on answering two related research questions. Our primary question is as follows: what is the relationship between patient social risk factors and patient COC among Medicare beneficiaries who make annual patient E&M visits? Secondarily, we ask what is the relationship between patient social risk factors and patient‐realized access to physician care for E&M among these same Medicare beneficiaries? We do not have predetermined hypotheses about the direction of these relationships. We use a rich dataset of Medicare claims linked to individual survey responses to assess the association between patient social risk factors and patient COC. We assess patient social risk at both the individual and area level because prior research has shown that both individual and neighborhood social risk factors are associated with patient health outcomes.18, 19, 20, 21, 22, 23, 24, 25 In secondary analyses, we assess whether there are also associations between these same patient social risk factors and realized access to different forms of physician care. We do this by assessing the effect of social risk on a) the number of unique physicians visited for annual patient evaluation and management (E&M) visits and b) whether any specialist physician was involved in annual patient E&M visits.

2. BACKGROUND

Patient COC is a construct referring to the extent to which discrete patient‐physician ambulatory care encounters (ie, face‐to‐face office visits) are connected over time.26, 27, 28 Conceptually, there are three main types of continuity: management, informational, and relational.27, 29 Management continuity refers to organizational and institutional coordination in the delivery of ambulatory care across multiple providers. Informational continuity refers to the availability and communication of patient information across providers. In their focus on managing patients across multiple providers, both management and informational continuity conceptually overlap with care coordination.30 Lastly, relational continuity, which is most valued in primary care, focuses on the ongoing relationship(s) between patients and their providers who have assumed responsibility for patients′ care in the form of regular E&M.27, 29 The overwhelming majority of the research on patient COC focuses exclusively on relational continuity.11

Patient COC is most frequently assessed in health services research with measures that describe the concentration or dispersion of patients′ E&M visits—usually on an annual basis—across ambulatory care physicians.28 The two most commonly used measures are the Usual Provider of Care (UPC) Index and the Bice‐Boxerman Continuity of Care (COC) index.28 The UPC Index characterizes the concentration of a patient′s visits with the single provider who has the plurality of the patient′s visits. The COC index characterizes the dispersion of a patient′s E&M visits across multiple providers and tends to be appropriate for patients with multiple medical visits and complex conditions—such as older adults.28, 31 Both the COC and UPC Indices range from 0 to 1, and higher scores (closer to 1) are interpreted as better COC. Higher scores are achieved when fewer physicians are seen for annual E&M.28, 32 As a result, there is an inverse relationship between patient‐realized access to ambulatory physician care for E&M and patient COC. As more physicians are accessed by patients for their care, continuity with a single physician decreases along with the corresponding scores on the UPC and COC Indices.

Although there is much research on the association of patient COC with health outcomes and resource use,11 there is a lack of research on patient social risk factors associated with COC. In fact, we are not aware of any. This is important because the National Academies of Sciences, Engineering, and Medicine have called for improving patient COC as one of the key health systems practices for improving the care of socially at‐risk populations.33 At the same time, however, the National Academies also highlight research findings that socially at‐risk patient populations tend to be clustered in a narrower subset of medical providers than the general population, such as community health centers, and that this may not always be beneficial for their health.33 Indeed, prior research has shown that patients at lower socioeconomic status (SES) are less likely to access medical care in general and are particularly less likely to access specialty care.34, 35, 36, 37, 38, 39, 40 This is of special concern for older adult Medicare beneficiaries, many of whom have complex chronic conditions and multiple care needs. Indeed, prior research in the related field of care coordination shows that care coordination practices are less available among socially at‐risk populations41 but that explicitly including information on social determinants of health in clinical and informational workflows can improve care coordination and management continuity.42 As a result, it is important to understand how patient social risk factors impact Medicare beneficiaries′ COC as well as their realized access to ambulatory care providers, including specialists.

3. METHODS

3.1. Study design and data

This is a retrospective observational study assessing the association between patient social risk factors and patient COC after adjusting for common measures of medical complexity within a multivariable regression framework. In secondary analyses, we assess the effect of social risk factors on different forms of patient‐realized access to physician care. We assess the same‐year effect of social risk factors on patient COC and access to care because we hypothesize a simultaneous effect of patient social risk factors on their interactions with outpatient providers. To do this, we conduct secondary data analysis in the Medicare fee‐for‐service (FFS) population using the Medicare Current Beneficiary Survey (MCBS) for the years 2006‐2013, linking the MCBS survey component to beneficiaries′ FFS Medicare claims. The MCBS is an annual nationally representative survey of the Medicare population conducted by CMS43 and includes important information on many patient nonmedical characteristics. We also link data on hospital service area (HSA)‐level health care supply provided by the Dartmouth Institute44 and county‐level economic conditions from the Area Health Resources File (AHRF) provided by the Health Resources and Services Administration45 to MCBS respondents′ records using their address of residence.

3.2. Study sample

Our study sample includes all community‐dwelling Medicare beneficiaries who are: (a) continuously enrolled in Medicare Part A and Part B for at least one calendar year during 2006‐2013 overlapping their participation in the MCBS and (b) identified as having three or more annual ambulatory evaluation and management (E&M) visits (described below) with national provider identifiers (NPIs) listed on their claims. Restricting the study population to patients with three or more annual E&M visits is necessary to get stable measures of patient COC.46 We exclude patients with missing NPIs on the majority of their E&M visits because we are unable to accurately measure COC for them. In addition, because some of our predictor variables are related to area‐level factors, we exclude individuals that had no listed US Zip code. (This led us to exclude 0.5 percent of the eligible sample, and only 1.7 percent of the included sample had any missing NPIs.)

We use the MCBS cross‐sectional survey weights to compute nationally representative estimates. These weights account for the overall selection probability of each person sampled and include adjustments for the stratified sampling design, survey nonresponse, and coverage error.

3.3. Dependent variables

Our primary dependent variable is patient COC measured using annual patient ambulatory E&M visits for each patient‐year. We identify such E&M visits using the Berenson‐Eggers Type of Service (BETOS) codes in Medicare Part B physician claims as face‐to‐face office visits for new and established patients, home visits, nursing home visits, specialist visits, and consultations. As in prior studies,1, 3 we further limit E&M visits by physician specialty type to PCPs and other medical specialists who provide regular E&M care (eg, cardiology and endocrinology), excluding visits to specialty types deemed inappropriate for regular E&M (eg, critical care and radiology). We also include visits to federally qualified health centers (FQHCs) and rural health centers (RHCs) from Medicare outpatient facility claims. (Details on our E&M visit definition and the codes used are in Appendix S1).

We measure patient COC using the Bice‐Boxerman Continuity of Care (COC) index. As shown in Figure 1, the COC index mathematically characterizes patient COC with a score ranging from 0 to 1 based on the dispersion of unique physicians seen by patients across the total count of their E&M visits.28, 47 Fewer physicians seen by a patient in a year will result in a higher COC index score (closer to 1), interpreted as better patient COC. We then stratify the patients in our sample into quintiles based on their COC index score and identify the patients in the bottom quintile as having low continuity and patients in the top quintile as having high continuity. In Figure 1, we illustrate two examples of low and high continuity using two typical patients. The low continuity patient has six annual E&M visits to four unique physicians—two to the same PCP, two to the same cardiologist, one to an ophthalmologist, and another visit to an endocrinologist—resulting in a COC index score of 0.13. The high continuity patient has five annual E&M visits to two unique physicians—four to the same PCP and one to a cardiologist—resulting in a COC index score of 0.60.

Figure 1.

The Continuity of Care (COC) index—with examples of low versus high continuity

We also measure two secondary dependent variables to represent different kinds of patient‐realized access to annual physician E&M care. First, using our E&M visit definition outlined above, we count the quantity of unique physicians involved in the patient's annual E&M visits. Second, we create a binary indicator of whether or not at least one specialist physician was involved in the patient's annual E&M visits (see the medical specialists listed in Appendix S1 for our identification of specialist physicians).

3.4. Independent variables: patient social risk

Patient social risk is assessed at the individual level with measures of socioeconomic status (SES) and race and at the area level with measures of health care supply and local economic conditions. We use survey responses to assess individual SES as material capital (self‐reported annual income), human capital (highest level of education attained), and social capital (living alone). We also use survey responses to assess race as white, black/African American, and other race. We use the Dartmouth health service area (HSA) data linked to patients′ home addresses to measure local area health care supply as the rate of PCPs per 100 000 and medical specialists per 100 000 population in the patients′ local HSAs.44 We use the AHRF county‐level measures linked to patients′ home addresses to measure local area economic conditions as rural‐urban setting (metropolitan, micropolitan, rural) and poverty rate.45

3.5. Control variables: patient medical complexity

We control for measures of patient medical complexity, disease severity, and supplemental insurance status that we hypothesize impact patient access to ambulatory care and patient COC apart from social risk. We control for age and gender using information from Medicare administrative data. To control for medical comorbidities and disability status, we use the CMS hierarchical condition categories (CMS‐HCC) risk score.48 We also assess patient acute and postacute care intensity with counts of emergency department (ED) visits, inpatient stays, and skilled nursing facility (SNF) stays as these are not only indicators of medical complexity but also reduce the available time patients have for making ambulatory E&M provider visits. Further, we directly control for patient utilization intensity of ambulatory care using the count of annual patient E&M visits and an indicator of whether patients are missing any NPIs on their E&M visit claims. Lastly, we control for patient insurance status with an indicator for any private supplemental insurance and a separate indicator for Medicare and Medicaid dual enrollment.

3.6. Analytic approach

First, we calculate weighted descriptive statistics from the MCBS for all of our dependent, independent, and control variables. We compute these descriptive statistics for the overall study sample, as well as for the study subsamples stratified by low versus high patient COC. We assess statistically significant differences in proportions (or means) in all of our study variables by the stratified study subsamples for low versus high patient COC using the Wald test. We also use the Pearson correlation coefficient and graphical analysis to describe the relationship between patient quintile rank on the COC index and probability of annual specialist involvement and number of unique physicians providing E&M care.

Next, we assess the concurrent‐year association between the patient social risk factor variables identified above and our dependent variable patient COC after adjusting for the control variables listed above. We do this by estimating two multivariable regression models. In the first model, we use logistic regression to assess the association between each of our social risk factor variables and the probability of having high versus low patient COC among all patients at a high or low quintile ranking for patient COC. In the second model, we use ordinary least squares (OLS) regression to assess the average effect of each of our social risk factor variables on patient COC measured as a continuous variable among the entire study sample.

Lastly, we assess the concurrent‐year association between the patient social risk factor variables identified above and our two dependent variables measuring patient‐realized access to annual physician E&M care after adjusting for the control variables listed above. We do this by estimating two multivariable regression models. In the first model, we use Poisson regression to assess the association between each of our social risk factor variables and incidence of unique physicians visited annually for patient E&M. In the second model, we use logistic regression to assess the association between the same social risk factors variables and probability of having at least one specialist physician involved in annual patient E&M.

In addition, to address the possibility that beneficiaries we excluded with <3 annual E&M visits are systematically different from our sample, we conduct a sensitivity analysis including these beneficiaries and re‐estimating the regression models above with patient‐realized access to care as the outcome variables.

In all of our models, we include year fixed effects to account for secular trends in technology and health care delivery system change. Standard errors are clustered at the individual beneficiary level and adjusted for the complex survey design of the MCBS and intra‐person correlation over time. We report all our regression results with 95% confidence intervals.

We performed analyses using SAS version 9.4 and Stata version 14.

4. RESULTS

4.1. Characteristics of the study population

Of 170 125 109 (weighted) patient years eligible for inclusion in the study, we excluded 794 757 (0.5 percent of eligible observations) that did not have a valid US zip code, were missing NPIs on a plurality of their E&M visits, or were missing data on an important covariate of interest (Figure S1). Our final study population consisted of 169 330 352 patient years (59 499 unweighted patient years and 21 316 unique patients). This was the sample we used for both our descriptive and regression analyses.

Descriptive statistics for our study population are shown in Table 1. The results stratified by patients with low versus high continuity on the COC index show an inverse relationship between patient COC and realized access to physician care. Specifically, patients with higher (ie, better) COC see fewer unique physicians annually for E&M and are less likely to visit a specialist for E&M (P < .001). This relationship is highlighted in Figure 2, showing a positive gradient between COC quintile and the probability of any specialist involvement annually (r = −.58, P < .001) and the number of unique physicians providing E&M care annually (r = −.65, P < .001).

Table 1.

Descriptive statistics for the patient study population

	Overall	Continuity of Care (COC) index quintilea
		Low continuity	High continuity	P‐value*
Total number of patient years, unweighted, N	59 499	11 604	12 186	‐
Total number of patient years, weighted, N	169 330 352	33 887 026	33 864 105	‐
Values displayed as percentages (Unless otherwise noted)	100.0	100.0	100.0	‐
Dependent variables
Continuity of Care (COC) index (0‐1), Mean (SD)	0.37 (0.27)	0.10 (0.05)	0.83 (0.18)	<.001
Total physicians visited for E&M, Mean (SD)	3.7 (2.2)	6.0 (2.3)	1.7 (0.9)	<.001
Visited a specialist physician at least once	82.8	96.5	44.7	<.001
Individual social risk factors
Education
No highschool or college education	22.2	13.9	31.8	<.001
Highschool/Some college education	57.2	56.9	56.0	.095
College/Graduate school education	20.7	29.2	12.2	<.001
Annual income
<$25 000	42.8	32.1	55.9	<.001
≥$25 000	33.3	36.0	28.2	<.001
≥$50 000/Unknown (Reference)	23.9	31.9	15.9	<.001
Lives alone	31.5	29.0	33.9	<.001
Race
White	86.7	89.7	83.0	<.001
Black	8.6	6.6	10.9	<.001
Other	4.7	3.7	6.1	<.001
Residential social risk factors
Rural‐Urban area
Metropolitan	73.8	82.3	65.9	<.001
Micropolitan	16.8	11.7	20.7	<.001
Rural	9.4	6.0	13.3	<.001
PCP supply in HSA, Per 100 000, Mean (SD)	72.7 (18.4)	73.2 (18.4)	72.8 (19.2)	.112
Specialist supply in HSA, Per 100 000, Mean (SD)	45.7 (14.8)	48.1 (14.6)	43.8 (15.5)	<.001
Adults below poverty, Per 100 in County, Mean (SD)	14.9 (5.7)	14.1 (5.3)	15.7 (6.0)	<.001
Control variables—medical claims
Age, Mean (SD)	72.2 (11.7)	72.6 (10.8)	70.5 (13.1)	<.001
Gender
Female	58.0	57.5	59.2	.112
Male	42.0	42.5	40.8
E&M visits, Mean (SD)	9.5 (6.8)	10.8 (7.1)	7.9 (6.9)	<.001
Missing ≥1 NPIs on E&M Visits	1.7	1.9	1.2	<.001
CMS‐HCC risk score, Mean (SD)	1.1 (0.9)	1.2 (1.0)	1.0 (0.8)	<.001
Inpatient days, Mean (SD)	1.7 (6.1)	1.9 (6.9)	1.5 (5.4)	.002
Emergency department visits, Mean (SD)	0.7 (1.6)	0.7 (1.7)	0.6 (1.6)	.032
Skilled nursing facility days, Mean (SD)	1.1 (7.4)	1.1 (7.1)	1.1 (7.5)	.913
With private supplemental insurance	67.7	74.6	55.5	<.001
With Medicaid dual enrollment	16.8	12.6	24.4	<.001

Weighted estimates from the 2006‐2013 Medicare Current Beneficiary Survey (MCBS) for all community‐dwelling fee‐for‐service Medicare beneficiaries with at least one year of enrollment in Medicare Part A and Part B and having completed the fall survey round of the MCBS (after applying exclusion criteria), as well as having at least three evaluation and management patient visits. Using the MCBS cross‐sectional weights accounting for the overall annual selection probability of each person sampled and including adjustments for the stratified sampling design, survey nonresponse, and coverage error.

Abbreviations: E&M, evaluation and management; NPI, national provider identifier; N, number; SD, standard deviation.

^aBice‐Boxerman Continuity of Care (COC) index quintile rank (bottom 0‐0.16; top 0.55‐1.00).

^* P‐value for difference in proportions (or means for costs) by Continuity of Care (COC) index patient quintile categories (Wald tests). Survey estimation commands were used to adjust P‐values for the complex survey design of the MCBS and robust clustered on individuals to account for within person correlation due to the same persons appearing in the data more than once over multiple years.

Figure 2.

Specialist and physician involvement by Continuity of Care (COC) index quintile

As expected, there is a strong negative relationship (P < .05) between level of patient COC and 8 of 10 measures of patient medical complexity and insurance status (Table 1). The general pattern of findings shows sicker patients utilize more physicians and have lower (ie, worse) COC. Finally, we find a strong relationship between level of patient COC and 4 of 4 measures of individual social risk and 3 of 4 measures of residential social risk (P < .001). Patients with more individual social risk factors (less education, lower income, living alone, and black or other race) and more residential social risk factors (living in low specialist supply, nonurban, or high‐poverty areas) also have higher (ie, better) COC.

4.2. Patient social risk factors associated with continuity of care

The results of our logistic regression model estimating the relationship between patient social risk factors and odds of high versus low patient COC are shown in Table 2. The results of our OLS regression model estimating the same relationship with COC measured as a continuous variable among all patients are also shown. After adjustment for medical complexity, individual‐level patient social risk factors such as no high school or college education (2.87 OR, 95% CI, 2.50‐3.30), income <$25 000 (2.02 OR, 95% CI, 1.80‐2.27), living alone (1.11 OR, 95% CI, 1.01‐1.21), black race (1.35 OR, 95% CI, 1.16‐1.58), and other race (1.28 OR, 95% CI, 1.05‐1.57) are all associated with increased odds of high versus low patient COC. Similarly, area‐level social risk factors such as living in micropolitan (1.45 OR, 95% CI, 1.06‐1.98), rural (1.48 OR, 95% CI, 1.11‐1.97), high‐poverty (1.11 OR, 95% CI, 1.03‐1.19), or higher PCP supply (1.22 OR, 95% CI, 1.08‐1.37) areas are associated with increased odds of high versus low patient COC. However, living in an area with a higher specialist supply (0.75 OR, 95% CI, 0.65‐0.85) is associated with decreased odds of high versus low patient COC. These same risk factors are all significant (P < .05) in the same direction for our OLS regression results.

Table 2.

Patient social risk factors associated with continuity of care

	Odds of high vs low continuity of carea		Overall effect on continuity of careb
	OR	(95% CI)	OLS β	(95% CI)
Total number of patient years, unweighted N	23 790		59 499
Total number of patient years, weighted N	67 751 131		169 330 352
Individual social risk factors
Education
No highschool or college education	2.87	(2.50, 3.30)	0.09	(0.07, 0.10)
Highschool/Some college education	1.69	(1.50, 1.89)	0.04	(0.03, 0.05)
College/Graduate school education (Reference)	‐	‐	‐	‐
Annual income
<$25 000	2.02	(1.80, 2.27)	0.06	(0.05, 0.07)
≥$25 000	1.32	(1.19, 1.46)	0.02	(0.01, 0.03)
≥$50 000/Unknown (Reference)	‐	‐	‐	‐
Lives alone	1.11	(1.01, 1.21)	0.01	(0.00, 0.01)
Race
White	‐	‐	‐	‐
Black	1.35	(1.16, 1.58)	0.03	(0.01, 0.04)
Other	1.28	(1.05, 1.57)	0.02	(0.00, 0.04)
Residential social risk factors
Rural‐urban area
Metropolitan (Reference)	‐	‐	‐	‐
Micropolitan	1.45	(1.06, 1.98)	0.03	(0.00, 0.05)
Rural	1.48	(1.11, 1.97)	0.03	(0.01, 0.06)
PCP supply in HSA (1 SD Increase)	1.22	(1.08, 1.37)	0.02	(0.01, 0.03)
Specialist supply in HSA (1 SD Increase)	0.75	(0.65, 0.85)	‒0.02	(−0.03, −0.01)
Adults below poverty in county (1 SD Increase)	1.11	(1.03, 1.19)	0.01	(0.00, 0.01)

Weighted estimates from the 2006‐2013 Medicare Current Beneficiary Survey (MCBS) for all community‐dwelling fee‐for‐service Medicare beneficiaries with at least one year of enrollment in Medicare Part A and Part B and having completed the fall survey round of the MCBS (after applying exclusion criteria), as well as having at least three evaluation and management patient visits. Using the MCBS cross‐sectional weights accounting for the overall annual selection probability of each person sampled and including adjustments for the stratified sampling design, survey nonresponse, and coverage error.

Abbreviations: β, regression coefficient; CI, confidence interval; N, number; OR, odds ratio; OLS, ordinary least squares; SD, standard deviation.

^aLogistic regression results. Odds ratio for being in the highest vs lowest quintile rank of the Bice‐Boxerman Continuity of Care (COC) index. All estimates and standard errors have been adjusted to account for the MCBS complex sampling methodology, as well as within person correlation due to the same beneficiaries appearing in the data more than once over multiple years. Model adjusts for the control variables listed in Table 1 and year fixed effects to control for secular trend.

^bOLS regression results. Modeling the marginal effect of the independent variables on the dependent variable, the Bice‐Boxerman Continuity of Care (COC) index, modeled as a continuous linear variable. All estimates and standard errors have been adjusted to account for the MCBS complex sampling methodology, as well as within person correlation due to the same beneficiaries appearing in the data more than once over multiple years. Model adjusts for the control variables listed in Table 1 and year fixed effects to control for secular trend.

4.3. Patient social risk factors associated with number of physicians involved annually in patient E&M

Table 3 shows the results of our model estimating the association between patient social risk factors and incidence of unique physicians visited annually for patient E&M. After adjustment for medical complexity, no high school or college education (0.86 IRR, 95% CI, 0.84‐0.88), income <$25 000 (0.93 IRR, 95% CI, 0.91‐0.94), and black race (0.96 IRR, 95% CI, 0.93‐0.98) are all associated with decreased incidence of unique physicians visited annually for E&M. Similarly, area‐level social risk factors such as living in micropolitan (0.94 IRR, 95% CI, 0.90‐0.98), rural (0.92 IRR, 95% CI, 0.89‐0.96), or higher PCP supply (0.98 IRR, 95% CI, 0.96‐0.99) areas are associated with decreased incidence of unique physicians visited annually for E&M. However, living in an area with a higher specialist supply (1.04 IRR, 95% CI, 1.02‐1.06) is associated with increased incidence of unique physicians visited annually for E&M.

Table 3.

Patient social risk factors associated with number of physicians involved annually for evaluation and management and odds of any specialist involvement

	Incidence of unique physicians visited for annual E&Ma		Odds of any specialist involvement in annual E&Mb
	IRR	(95% CI)	OR	(95% CI)
Total number of patient years, unweighted N	59 499		59 499
Total number of patient years, weighted N	169 330 352		169 330 352
Individual social risk factors
Education
No highschool or college education	0.86	(0.84, 0.88)	0.47	(0.42, 0.53)
Highschool/ Some college education	0.94	(0.92, 0.95)	0.66	(0.60, 0.74)
College/ Graduate school education (Reference)	‐	‐	‐	‐
Annual income
<$25 000	0.93	(0.91, 0.94)	0.61	(0.55, 0.67)
≥$25 000	0.97	(0.96, 0.99)	0.80	(0.73, 0.88)
≥$50 000/Unknown (Reference)	‐	‐	‐	‐
Lives alone	0.99	(0.98, 1.00)	0.94	(0.87, 1.01)
Race
White	‐	‐	‐	‐
Black	0.96	(0.93, 0.98)	0.94	(0.84, 1.05)
Other	0.97	(0.94, 1.00)	0.86	(0.73, 1.02)
Residential social risk factors
Rural‐urban area
Metropolitan (Reference)	‐	‐	‐	‐
Micropolitan	0.94	(0.90, 0.98)	0.85	(0.73, 0.99)
Rural	0.92	(0.89, 0.96)	0.68	(0.58, 0.81)
PCP supply in HSA (1 SD Increase)	0.98	(0.96, 0.99)	0.84	(0.79, 0.90)
Specialist supply in HSA (1 SD Increase)	1.04	(1.02, 1.06)	1.27	(1.18, 1.37)
Adults below poverty in county (1 SD Increase)	0.99	(0.98, 1.00)	0.92	(0.87, 0.97)

Weighted estimates from the 2006‐2013 Medicare Current Beneficiary Survey (MCBS) for all community‐dwelling fee‐for‐service Medicare beneficiaries with at least one year of enrollment in Medicare Part A and Part B and having completed the fall survey round of the MCBS (after applying exclusion criteria), as well as having at least three evaluation and management patient visits. Using the MCBS cross‐sectional weights accounting for the overall annual selection probability of each person sampled and including adjustments for the stratified sampling design, survey nonresponse, and coverage error.

Abbreviations: CI, confidence interval; IRR, incidence rate ratio; N, number; OR, odds ratio; SD, standard deviation.

^aPoisson regression results for incidence of total unique physicians involved in annual patient E&M. All estimates and standard errors have been adjusted to account for the MCBS complex sampling methodology, as well as within person correlation due to the same beneficiaries appearing in the data more than once over multiple years. Model adjusts for the control variables listed in Table 1 and year fixed effects to control for secular trend.

^bLogistic regression results for odds of specialist physician involvement in annual patient E&M. All estimates and standard errors have been adjusted to account for the MCBS complex sampling methodology, as well as within person correlation due to the same beneficiaries appearing in the data more than once over multiple years. Model adjusts for the control variables listed in Table 1 and year fixed effects to control for secular trend.

4.4. Patient social risk factors associated with annual specialist involvement in patient E&M

Table 3 also shows the results of our model estimating the association between patient social risk factors and odds of annual specialist involvement in patient E&M. After adjustment for medical complexity, no high school or college education (0.47 OR, 95% CI, 0.42‐0.53) and income <$25 000 (0.61 OR, 95% CI, 0.55‐0.67) are associated with decreased odds of visiting a specialist at least once annually for E&M. Similarly, area‐level social risk factors such as living in micropolitan (0.85 OR, 95% CI, 0.73‐0.99), rural (0.68 OR, 95% CI, 0.58‐0.81), high‐poverty (0.92 OR, 95% CI, 0.87‐0.97), or higher PCP supply (0.84 OR, 95% CI, 0.79‐0.90) areas are associated with decreased odds of visiting a specialist at least once annually for E&M. However, living in an area with a higher specialist supply (1.27 OR, 95% CI, 1.18‐1.37) is associated with increased odds of visiting a specialist at least once annually for E&M.

4.5. Sensitivity analysis

Table S1 shows the result of our sensitivity analysis estimating the association between patient social risk factors and physician involvement (total unique and any specialist) in annual patient E&M, including beneficiaries with <3 annual E&M visits. We find results very similar to our main analysis in Table 3, although somewhat larger in magnitude for incidence of unique physicians visited annually. In addition, we now find that black race is significantly associated (0.83 OR, 95% CI, 0.74‐0.93) with decreased odds of visiting a specialist at least once annually for E&M. This sensitivity analysis suggests that the association of social risk factors with decreased patient‐realized access to physician E&M is even stronger among patients with few annual E&M visits and that racial disparities may exist in access to specialists. This finding draws attention to the limitation of the standard claims‐based measures of patient COC that require excluding patients with <3 annual E&M visits from the analysis.

5. DISCUSSION

In this nationally representative study of Medicare FFS beneficiaries who make annual E&M visits with physicians, we find that patients with greater social risk factors have consistently better COC. These social risk factors include low income, lack of educational attainment, lack of social support (living alone), black or other nonwhite race, higher area‐level poverty, and rural location. This finding is robust to adjustment for patient medical complexity and enrollment in supplemental insurance. It is also robust to whether we compare patients at high versus low COC or assess the average effect on COC across all patients.

In addition, we find that among this same population, those with more social risk factors access fewer physicians overall for their E&M visits and are much less likely to access a specialist. This finding, too, is robust to adjustment for medical complexity and supplemental insurance status. As a result, it appears that patient social risk factors simultaneously contribute to better patient COC and to reduced access to specialty care for ambulatory E&M care among Medicare FFS patients. Furthermore, we also find that greater area‐level PCP supply is associated with better patient COC but less patient‐realized access to physicians overall, and especially, specialists. Conversely, greater area‐level specialist supply is associated with lower levels of patient COC, but more patient‐realized access to specialists.

These findings may at first appear contradictory. Having more social risk factors is generally associated with worse health outcomes.18, 19, 20, 21, 22, 23, 24, 25 But, greater COC is generally associated with better health outcomes.1, 2, 3, 4, 5, 6, 7, 8, 9 So what is going on here? First, it is important to be clear that our data do not imply that lower subspecialty (ie, primary) care or better COC is a bad thing for patients, and, in particular, for socially at‐risk patients. Indeed, prior research finds that higher local area PCP supply is associated with lower mortality and preventable hospitalization rates among Medicare beneficiaries and the overall US population.49, 50, 51 These studies controlled for area‐level social risk factors and specialist physician supply. In addition, numerous prior studies find that patient COC is associated with better outcomes, including lower costs, lower mortality rates, fewer complications of disease, and fewer emergency department (ED) visits or hospitalizations.1, 2, 3, 4, 5, 6, 7, 8, 9, 52 Similarly, our own descriptive results show that patients in the top COC index quintile have fewer inpatient days and emergency department visits than patients in the bottom quintile. Nonetheless, we caution that our descriptive results should not be taken as causal because the variables were measured simultaneously, and thus, we are unable to establish temporal precedence between the cause (COC) and the effect (outcomes). Others have also cited lack of temporal precedence and the potential for reverse causality in many prior studies of the effect of COC on outcomes as a reason for caution.10, 11, 30

We believe that our study results highlight three concerns related to physician access and patient COC that warrant future investigation. First, patients with more social risk factors appear to have less access to higher subspecialty care. There also appear to be racial disparities in access to specialists. Multiple studies find that regular medical care by specialists in the ambulatory setting is associated with better quality of care, fewer hospitalizations, and lower mortality for individuals with chronic conditions.10, 53, 54, 55, 56 This does not detract from the value of primary care, but it does indicate reason for concern about a lack of specialist access among socially at‐risk Medicare beneficiaries, many of whom have chronic conditions. Second, patients with more social risk factors may have better COC (as measured on the standard claims‐based indices) due to structural factors that limit their access to and choice of providers. In rural or high‐poverty areas, such patients may simply have access to a limited supply of providers.57 Again, this does not imply that patient COC, in itself, is a bad thing, but it does suggest reasons for caution in how COC is measured and in assumptions about how it is experienced among socially at‐risk patients. Those with more social risk factors have bigger obstacles to accessing physician care that meets their needs (clinical and interpersonal) due to fewer resources at the individual level (time, knowledge, flexibility, and control of resources) and at the area level (distance and transportation to providers, availability of providers). There is also an equity issue here since such patients tend to be clustered in a narrower subset of providers than the general population that may not offer the same level or quality of care.33

Lastly, our results draw attention to the inadequacy of the standard claims‐based indices of COC for measuring the complex nature of continuity among patients. Patient COC is a multidimensional construct comprising management, informational, and relational continuity.27, 29 In particular, continuity includes organizational and informational coordination of the same patients across multiple providers, in addition to a regular relationship with the same provider(s) over time. However, most prior studies of patient COC rely solely on the claims‐based indices of COC that only capture the relational dimension of continuity.11 As others have pointed out, care coordination is an overlapping, but not identical, construct with COC that may be especially important to measure and implement among complex patients, such as Medicare beneficiaries.30

Our findings have implications for researchers, policy makers, and providers. First, there is a need to focus on improving ambulatory care access and quality, in addition to patient COC, among socially at‐risk Medicare beneficiaries. Second, it is important for providers to screen for and address social risk factors and needs among the patients for whom they provide regular E&M. Both the National Academy of Medicine′s and CMS′ Accountable Health Communities program encourage providers to screen for and address their patients′ social needs.58, 59 Third, it is time for researchers and policy makers to expand the way they measure and promote patient COC. In particular, the standard claims‐based measures should be supplemented with measures that reflect the complexity of patient care needs, such as access to specialists and coordination of care.

Finally, our findings also raise concerns about unmeasured confounding in previous studies of patient COC. Social risk factors are here associated with better patient COC, and prior research also shows that social risk factors tend to be associated with poorer patient outcomes. We know from prior research, for instance, that low SES is negatively associated with patient outcomes.20, 21, 22, 23, 24, 25, 60 Thus, prior studies assessing the effect of COC on patient outcomes that did not control for patient social risk factors may be biased toward the null hypothesis of no (or even a negative) effect. As a result, it is possible that prior studies finding a positive effect of COC on outcomes may have understated that effect.

We note several limitations. First, our sample is limited to the Medicare FFS population. This excludes 25 percent of the Medicare population enrolled in managed care during this period.61 Second, despite our inclusion of control variables not included in many previous studies of COC, we acknowledge the presence of unobserved patient factors. To the extent that we have not fully captured disease severity or opportunity to access physicians, among other unobserved factors, our estimates could be biased.

6. CONCLUSION

In conclusion, FFS Medicare patients with multiple social risk factors have consistently better COC. In addition, these same social risk factors are associated with reduced patient‐realized access to specialist physician care. Further research is needed on access to care and different forms of patient COC among Medicare patients at greater social risk.

CONFLICT OF INTEREST

KJJ is an Assistant Professor at Saint Louis University. JM and JMH have no conflict of interest.

Supporting information

 

 

Johnston KJ, Mittler J, Hockenberry JM. Patient social risk factors and continuity of care for Medicare beneficiaries. Health Serv Res. 2020;55:445–456. 10.1111/1475-6773.13272