Duration of medical home participation and quality of care for patients with chronic conditions

Abstract

Objective

To examine whether the length of participation in a patient‐centered medical home (PCMH), an evidence‐based practice, leads to higher quality care for Medicaid enrollees with multiple co‐morbid chronic conditions and major depressive disorder (MDD).

Data Sources

This analysis uses a unique data source that links North Carolina Medicaid claims and enrollment data with other administrative data including electronic records of state‐funded mental health services, a state psychiatric hospital utilization database, and electronic records from a five‐county behavioral health carve‐out program.

Study Design

This retrospective cohort study uses generalized estimating equations (GEEs) on person‐year‐level observations to examine the association between the duration of PCMH participation and measures of guideline‐concordant care, including the receipt of minimally adequate care for MDD, defined as 6 months of antidepressant use or eight psychotherapy visits each year.

Data Collection/Extraction Methods

Adults with two or more chronic conditions reflected in administrative data, including MDD.

Principal Findings

We found a 1.7 percentage point increase in the likelihood of receiving guideline‐concordant care at 4 months of PCMH participation, as compared to newly enrolled individuals with a single month of participation (p < 0.05). This effect increased with each additional month of PCMH participation; 12 months of participation was associated with a 19.1 percentage point increase in the likelihood of receiving guideline‐concordant care over a single month of participation (p < 0.01).

Conclusions

The PCMH model is associated with higher quality of care for patients with multiple chronic conditions and MDD over time, and these benefits increase the longer a patient is enrolled. Providers and policy makers should consider the positive effect of increased contact with PCMHs when designing and evaluating initiatives to improve care for this population.

What is known on this topic

• Chronic medical comorbidities negatively affect the quality of treatment for major depressive disorder (MDD).

• The PCMH model of primary care transformation provides a potential mechanism to improve receipt of guideline‐concordant care for patients with multiple chronic conditions, including MDD.

• However, there are few studies that assess the effect of duration of PCMH enrollment on treatment outcomes in order to understand the timing of estimated benefits in quality from PCMH initiation

What this study adds

• This study examined the relationship between duration of participation in a PCMH and receipt of quality MDD care among Medicaid enrollees with MDD and multiple chronic conditions.

• The PCMH model was associated with higher quality of MDD care for patients with MDD and comorbid chronic conditions over time, and these benefits increase the longer a patient is enrolled.

• These findings provide support for the idea that providers and policy makers should prioritize ensuring continuity of PCMH participation to maximize the model's benefits.

1. INTRODUCTION

For individuals with major depressive disorder (MDD), concurrent physical illnesses can complicate the provision of high‐quality, guideline‐concordant mental health care. Chronic medical comorbidity is thought to negatively affect the quality of MDD treatment by causing competing demands and diverting time and resources away from mental health concerns. ¹ Medical comorbidities may also reduce providers' focus on making necessary adjustments to MDD care, such as titrating antidepressant prescriptions and managing psychotherapy referrals. ² Previous research has shown that in patients with untreated MDD, each chronic physical condition reduces the odds that a primary care provider discusses MDD as a possible diagnosis or adjusts MDD treatment. ³ Physical comorbidities may also affect the likelihood of receiving guideline‐concordant psychotherapy for MDD; the presence of three or more medical comorbidities has been shown to decrease the proportion of depressed patients receiving guideline‐concordant psychotherapy visits by 7.39%. ⁴

Given the prevalence and impact of MDD and chronic physical comorbidities, it is important to address low rates of guideline‐concordant care in this population. In 2017, approximately, 17.3 million Americans, or 7.1% of the population, had experienced at least one major depressive episode in the past year, ⁵ and primary care patients with chronic medical illnesses have been found to have two‐ to threefold higher rates of MDD than those without medical comorbidities. ⁶

The patient‐centered medical home (PCMH) model of primary care provides a potential mechanism to improve receipt of guideline‐concordant care for patients with multiple chronic conditions, including MDD, by improving system efficiency and offering enhanced care coordination and condition management. ⁷ , ⁸ , ⁹ Research suggests that the populations that can most benefit from this model are those with chronic mental or physical conditions requiring long‐term management. ¹⁰ , ¹¹ A recent meta‐analysis found that multidisciplinary, team‐based, collaborative care results in greater reductions in illness burden and depressive symptoms than usual care among patients with MDD and comorbid chronic medical conditions. ¹² For Medicaid enrollees with MDD, participation in PCMHs has also been shown to increase the likelihood of medication adherence (4.1 percentage points, p < 0.001) ¹³ and receipt of psychotherapy (4.36 percentage points, p < 0.001). ¹⁴

The evidence regarding the benefit of the PCMH is growing. ⁸ , ¹⁵ , ¹⁶ However, there is little information regarding the effect of duration of PCMH participation on treatment outcomes. ¹⁷ For several reasons, increased duration of PCMH participation may affect receipt of minimally adequate MDD care for people with MDD and comorbid physical problems. Providers may need to prioritize treatments, and a longer duration of participation would allow more time to address numerous issues. Longer participation may, therefore, result in better management of other chronic conditions, giving providers and patients more time to systematically screen for, detect, and treat MDD. Research also shows that continuity of care in a PCMH model is associated with decreased emergency department utilization, and improving continuity may help reduce the utilization of non–primary care services. ¹⁷ Finally, sustained contact with PCMH providers may foster trust between patients and providers, which may lead to greater patient willingness to discuss symptoms and accept recommended therapies.

This paper examines the relationship between duration of participation in a PCMH and receipt of quality MDD care among Medicaid enrollees with MDD and multiple chronic conditions. If the effect of the PCMH occurs only after sustained contact, policy makers may prioritize ensuring continuity of PCMH participation to maximize the model's benefits. Given that duration of care likely serves an important role in condition management for this population, ¹⁸ the primary goal of this analysis is to assess whether the number of months of PCMH participation in a given year affects receipt of minimally adequate care (MAC) for MDD.

2. METHODS

2.1. Sample

The setting for this study was Community Care of North Carolina (CCNC), a regional primary care PCMH program for North Carolina's Medicaid enrollees comprising 14 not‐for‐profit care networks. While some variation exists between networks, all PCMHs share a core set of components, such as population management tools, pharmacy management, case management, evidence‐based programs and protocols, and regular performance metric reporting. ¹⁹ Beginning in 2007, CCNC created specialized chronic care programs to serve high‐cost, high‐need Medicaid enrollees. These services included disease management for those with multiple chronic conditions, improvements in access to urgent care, comingling mental health care in primary care settings, and transition management. ⁹ Patients enroll in a PCMH by selecting a local PCMH from a list of options or by accepting assignment to a PCMH close to their residence. North Carolina Medicaid enrollees not enrolled in a PCMH receive typical Medicaid fee‐for‐service (FFS) primary care. ²⁰ We defined PCMH participation as attribution to a PCMH provider during a given month; enrollees were not required to seek services from their primary care provider during the month to be considered a PCMH participant.

This analysis uses the North Carolina Integrated Data for Researchers, a unique data source that linked North Carolina Medicaid claims and enrollment data with other administrative data including state‐funded mental health services, a state psychiatric hospital utilization database, and encounter data from a five‐county behavioral health carve‐out program that existed during the study period. ⁴ , ²¹ Consistent with other evaluations of CCNC's PCMH program, we limited our analysis to the period when enrollment was ramping up (fiscal years 2008–2010), increasing our opportunity to observe enrollees prior to and during PCMH participation, and improving statistical identification. We do not consider data after 2010 because few changes in enrollment were observed after that time. ²² From the merged data, we selected adults of ages 18–64 years with MDD and at least one of the seven other target chronic conditions (asthma, chronic obstructive pulmonary disease, diabetes, hypertension, hyperlipidemia, seizure disorder, and schizophrenia) selected based on prevalence, costliness, immediacy of treatment effects, and feasibility of constructing claims‐based quality measures. Person‐years before an individual's initial MDD claim were excluded, as were person‐years in which an individual had less than 6 months of Medicaid enrollment after the initial MDD claim.

Because PCMH participation required Medicaid enrollment during this period and medication adherence can be derived only from Medicaid claims data, we limited the sample to person‐years with at least 6 months of Medicaid eligibility and restricted to months in which individuals were enrolled in Medicaid. Because a large proportion of Medicaid enrollees changed eligibility status at some point during a year, we included partial years of Medicaid eligibility to increase the generalizability of the study results. To focus on the effect of PCMH “dose” on those enrolled for at least part of the year, individuals with no PCMH participation during a given year were excluded. The sample also excludes individuals who received care through assertive community treatment teams during the study period (n = 3611) and, therefore, may have received psychotherapy that is not evident in this dataset.

2.2. Measures

The primary outcome variable, measured yearly, is a binary indicator of receipt of MAC for MDD. Based on clinical guidelines for antidepressant usage ²³ and studies, ²⁴ we defined MAC as receipt of either (1) at least eight psychotherapy visits and/or (2) at least six covered months of antidepressant prescriptions during a 12‐month period. American Psychiatric Association Guidelines recommend a 4–8 weeks of initial treatment period to assess adequacy of antidepressant treatment response followed by a 4–9 months of continuation of treatment once an adequate response is established. ²⁵ The 6 months of coverage used in this study were, therefore, deemed conservative.

Antidepressant usage was measured using prescription claims and based on the proportion of days covered (PDC) or the proportion of days in each month for which dispensed medication was available. ²⁶ An individual was considered to have adequate antidepressant coverage during a 12‐month period if they had at least 6 months with a PDC of at least 80%, a common threshold for measuring medication coverage. ²⁷ Psychotherapy utilization was measured using Current Procedural Terminology codes and included claims for individual, family, or group psychotherapy during a given year.

The primary explanatory variable is the number of months of PCMH participation during each year. Yearly control variables that may affect quality of care included the total number of target conditions, number of additional comorbidities, sex, race, ethnicity, and age. To account for the effect of variation in length of yearly Medicaid enrollment on quality of care, we controlled for number of months enrolled in Medicaid. In addition, we controlled for total previous PCMH participation since the beginning of the study period (excluding the current year) to address the cumulative effect of participation across multiple person‐years. Target diagnoses were based on outpatient, inpatient, and emergency department service use. Total number of comorbidities was measured using 56 diagnostic subcategories derived from the Chronic Illness and Disability Payment System (CDPS). ²⁸ Because the sample is limited to individuals with MDD, the mental health category was excluded from the CDPS conditions. Race was coded as a categorical variable including white, Black, Asian, American Indian, Native Hawaiian/Pacific Islander, and other. Ethnicity was defined as Hispanic or non‐Hispanic. Age was included as a quadratic term based on improved model fit over the linear version as assessed by the Akaike Information Criterion. All variables were defined and measured at the person‐year level.

2.3. Analytic methods

To account for the panel structure and address the clustering of repeated observations of individuals over time, data were analyzed at the person‐year level using generalized estimating equations (GEE) with binomial distribution and logit link. An exchangeable correlation structure was selected due to a lower quasi‐likelihood under the independence model criterion as compared to independent and unstructured correlations. Differential effects were calculated as changes in predicted probabilities using the method of recycled predictions. Standard errors for predictions and marginal effects were calculated using the delta method.

Our primary analysis focused on receipt of MAC, defined as meeting the minimum threshold for either antidepressant use (6 months) or psychotherapy (eight visits) in a 12‐month period. We also examined each of these outcomes independently. We calculated the effects of varying lengths of yearly PCMH participation (e.g., the average effect of an additional month of PCMH participation) to determine the timing of medical homes' effects.

To assess variation across subgroups of enrollees, we stratified the primary model assessing receipt of MAC by enrollee sex, Hispanic ethnicity, and race. In subgroup analyses stratified by race, we assessed outcomes independently for Black and American Indian enrollees. In primary models, covariates include Asian, Pacific Islander, and other race as additional racial categories to improve accuracy and model fit. However, we did not conduct subgroup analyses for these groups due to limited sample size for Asian and Pacific Islander enrollees and difficulty of interpretation for other race.

To test the robustness of our primary analysis, we conducted several sensitivity analyses. To account for unobserved time‐invariant factors, we estimated a linear probability model with person‐level fixed effects; this model was not used in the primary analysis because there are only up to three person‐years available. To test our assumptions about the definition of MAC, we estimated GEE models with varying definitions of minimally acceptable antidepressant use (4, 6, and 9 months). To test whether our results were primarily being driven by individuals who were continuously enrolled in a PCMH, we estimated a GEE model excluding person‐years with the full 12 months of PCMH participation. A sensitivity analysis requiring continuous yearly Medicaid enrollment was also conducted.

To assess whether the effect of increased duration of participation in a PCMH was driven primarily by an increase in primary care visits, we also conducted a stratified analysis, separating enrollees into high and low primary care utilizers. Among enrollees with at least 1 month of PCMH participation, the median number of annual primary care visits was 4; therefore, high primary care utilization was defined as 4 or more primary care visits per year. Finally, to assess whether our results were being driven by a “new user” effect, we conducted separate sensitivity analyses limiting the study sample to enrollees with no PCMH participation in the prior year and excluding PCMH enrollees that had fewer than two primary care visits in the year prior to PCMH participation.

3. RESULTS

The final sample was composed of 118,436 person‐years and 62,538 unique individuals. The sample was 22.2% male, 62.3% white, and 1.8% Hispanic, and the average age was 41 years (Table 1). The mean number of CDPS comorbidities was 12. The average number of months enrolled in Medicaid per year was 11.02. The average number of medical‐home‐enrolled months per year was 8.95 (standard deviation = 3.47) and 39.61% of person‐years had 12 months of PCMH participation. A total of 33,932 person‐year observations (33.72%) met the criteria for MAC; 33,182 observations (28.02%) met the criteria for 6+ months of antidepressants; and 11,196 observations (9.45%) met the criteria for 8+ psychotherapy visits.

TABLE 1.

Patient sample statistics

	Mean (SD) or N (%)
Person‐years	N = 118,436
Medical home months (yearly)	8.95 (3.47)
Prior PCMH months	15.18 (9.52)
Average PCP visits in year prior to PCMH participation	5.30 (5.42)
Medicaid enrollment (yearly)	11.02 (1.76)
Male	26,293 (22.20%)
Age	41.57 (12.018)
Race
White	73,811 (62.30%)
Black	37,904 (32.00%)
Asian	322 (0.30%)
American Indian	2251 (1.90%)
Pacific Islander	49 (<1%)
Other	4099 (3.50%)
Hispanic ethnicity	2174 (1.80%)
# CDPS conditions	12.15 (4.99)

Abbreviations: CDPS, Chronic Illness and Disability Payment System; PCP, primary care physician; PCMH, patient‐centered medical home; SD, standard deviation.

In order to determine whether PCMH initiation was associated with lack of prior access to primary care, we assessed rates of primary care utilization in the year prior to PCMH participation, as well as during PCMH‐enrolled years, as compared to FFS enrollees who never participated in a PCMH during the study period. Figure 1 shows the distribution of primary care visits in these three groups. It is important to note that because this dataset is limited to three fiscal years, we are unable to determine whether enrollees were using primary care prior to the first year in the study period (2008). Overall, differences between the FFS and newly initiated PCMH groups were modest, with PCMH enrollees having slightly higher rates of prior primary care utilization. Among enrollees that became newly engaged in a PCMH in 2009 or 2010, 78.79% had at least one primary care visit in the year prior to initiating PCMH participation, and 88.85% had at least one visit during PCMH‐enrolled years, compared to 72.95% of enrollees with no PCMH participation during the study period. The average number of primary care visits in the year prior to PCMH participation was 5.3, compared to an average of 5.8 visits during PCMH‐enrolled years and 4.9 visits among enrollees that never participated in a PCMH during the study period.

FIGURE 1.

Distribution of primary care visits by medical home participation status

Our primary analysis showed a positive and significant association between additional months of PCMH participation and receipt of MAC (Table 2). The average predicted probability of receiving MAC ranged from 18.6% (p < 0.01) at 1 month of participation to 37.7% (p < 0.01) at 12 months (not shown). Compared to those with a single month of PCMH participation, we found no statistically significant association with receipt of MAC for individuals enrolled for 2–3 months. However, individuals that participated in a PCMH for 4 or more months during a given year were more likely to receive MAC. This effect increased with additional exposure time (Supplementary Figure S1); individuals with 4 months of PCMH participation experienced a 1.8 percentage point increase in the likelihood of MAC (p = 0.018) and those with 12 months of PCMH participation exhibited an increase of 19 percentage points (p < 0.01). This increase appears to be driven primarily by antidepressant usage; in separate models for the two components of the composite outcome, individuals with 12 months of participation had a 17.7 percentage point increase in guideline‐concordant antidepressant use (p < 0.01) and a 2.8 percentage point increase in the likelihood of having at least eight psychotherapy visits (p < 0.01). Prior‐year PCMH participation conferred a small but statistically significant negative association with MAC; a 1‐month increase in prior participation was associated with a −0.25 percentage point decrease in the likelihood of MAC.

TABLE 2.

Marginal effects of duration of PCMH participation on receipt of guideline‐concordant care

	Minimally adequate care	Psychotherapy only	Antidepressants only
# PCMH months
2	0.0033	0.0097	−0.0059
	(0.0075)	(0.0055)	(0.0063)
3	0.010	0.0018	0.0082
	(0.0075)	(0.0054)	(0.0064)
4	0.018*	0.0032	0.012
	(0.0074)	(0.0052)	(0.0063)
5	0.041**	0.0031	0.035**
	(0.0078)	(0.0053)	(0.0067)
6	0.099**	0.012*	0.093**
	(0.0077)	(0.0052)	(0.0068)
7	0.13**	0.013*	0.12**
	(0.0078)	(0.0053)	(0.0070)
8	0.15**	0.014**	0.15**
	(0.0074)	(0.0050)	(0.0066)
9	0.15**	0.0096*	0.16**
	(0.0072)	(0.0049)	(0.0063)
10	0.15**	0.017**	0.16**
	(0.0069)	(0.0048)	(0.0060)
11	0.17**	0.023**	0.16**
	(0.0066)	(0.0046)	(0.0057)
12	0.19**	0.028**	0.18**
	(0.0062)	(0.0044)	(0.0052)
Prior PCMH participation	−0.0025**	−0.00059**	−0.0022**
	(0.00014)	(9.72e‐05)	(0.00012)
# CDPS conditions	0.0018**	−0.00052*	0.0024**
	(0.00033)	(0.00021)	(0.00031)
Male	−0.037**	−0.0070**	−0.035**
	(0.0038)	(0.0024)	(0.0036)
Race
Black	−0.14**	−0.0081**	−0.15**
	(0.0034)	(0.0022)	(0.0031)
Asian	−0.056	−0.014	−0.047
	(0.031)	(0.019)	(0.030)
American Indian	−0.079**	−0.044**	−0.059**
	(0.012)	(0.0060)	(0.011)
Pacific Islander	−0.062	−0.049	−0.033
	(0.082)	(0.0370)	(0.082)
Other	−0.048**	−0.030**	−0.038**
	(0.0094)	(0.0053)	(0.0091)
Hispanic ethnicity	−0.041**	0.0095	−0.050**
	(0.012)	(0.0088)	(0.011)
Age	0.0034**	−0.00085**	0.0040**
	(0.00014)	(8.00e‐05)	(0.00013)
Medicaid months	0.050**	0.014**	0.050**
	(0.0010)	(0.00068)	(0.0010)
N (person‐years)	118,436	118,436	118,436

Note: Delta method standard errors in parentheses (**p < 0.01, *p < 0.05).

Abbreviations: CDPS, Chronic Illness and Disability Payment System; PCMH, patient‐centered medical home.

In the primary analyses, we observed racial differences in the likelihood of receiving guideline‐concordant care. Racially marginalized groups and enrollees of Hispanic ethnicity were less likely to receive MAC, with Black enrollees observed to have the largest difference: a 14.3 percentage point reduction in the likelihood of MAC (p < 0.01) as compared to non‐Hispanic white enrollees. These differences were estimated using predictive margins and should be interpreted as the residual direct effect (RDE) of race and ethnicity after adjusting for all measured covariates. The RDE does not account for the effect of factors that may mediate disparities in receipt of MAC, such geographic or socioeconomic characteristics. ²⁹ , ³⁰ In exploratory subgroup analyses of receipt of MAC stratified by race and ethnicity (Table 3), we found that Black, American Indian, and Hispanic enrollees did see a benefit from increased contact with the PCMH model and were more likely to receive MAC with longer durations of PCMH participation.

TABLE 3.

Marginal effects stratified by race and ethnicity

	Black	American Indian	Hispanic
# PCMH months
2	0.0303	0.0920	0.189
	(0.0191)	(0.0889)	(0.0815)
3	0.00348	0.0559	0.172
	(0.0189)	(0.0847)	(0.0797)
4	0.0149	−0.00303	0.189
	(0.0184)	(0.0896)	(0.0821)
5	0.0514**	0.235*	0.175
	(0.0182)	(0.0798)	(0.0832)
6	0.109**	0.158	0.135
	(0.0168)	(0.0814)	(0.0806)
7	0.104**	0.219*	0.179
	(0.0169)	(0.0750)	(0.0816)
8	0.132*	0.183	0.259*
	(0.0161)	(0.0792)	(0.0751)
9	0.129*	0.232*	0.228*
	(0.0157)	(0.0738)	(0.0733)
10	0.129*	0.144	0.256*
	(0.0153)	(0.0734)	(0.0724)
11	0.152*	0.214*	0.262*
	(0.0150)	(0.0702)	(0.0711)
12	0.167*	0.238*	0.274*
	(0.0146)	(0.0687)	(0.0699)
Prior PCMH participation	−0.00214*	−0.00254	−0.00162
	(0.000229)	(0.00101)	(0.00102)
# CDPS conditions	0.00356*	0.00642**	0.00124
	(0.000539)	(0.00239)	(0.00249)
Male	−0.0235*	−0.0718*	0.0105
	(0.00670)	(0.0288)	(0.0274)
Black			−0.137*
			(0.0500)
Asian			−0.0588
			(0.143)
American Indian			−0.0919
			(0.134)
Pacific Islander			−0.0401
			(0.0996)
Other race			−0.113*
			(0.0243)
Hispanic ethnicity	0.00325	−0.0757
	(0.0417)	(0.137)
Age	0.00287**	0.00465**	0.00218
	(0.000241)	(0.00103)	(0.00102)
Medicaid months	0.0372**	0.0681**	0.0454*
	(0.00176)	(0.00880)	(0.00732)
Observations	37,904	2251	2174

Delta method standard errors in parentheses (**p < 0.01, *p < 0.05).

Abbreviations: CDPS, Chronic Illness and Disability Payment System; PCMH, patient‐centered medical home.

We also observed differences in receipt of MAC by sex; in the primary models, men were 3.7 percentage points less likely to receive MAC compared to women (p < 0.01). In subgroup analyses stratified by sex (not shown), men saw a statistically significant association between additional months of PCMH participation and increase likelihood of receipt of MAC beginning at 5 months.

3.1. Sensitivity analyses

In sensitivity analyses, the fixed‐effects model did not show a significant association between PCMHs and MAC until 7 months of participation. Otherwise, the fixed‐effects model produced slightly smaller effect sizes than the GEE model, but with similar magnitude and direction of effects, suggesting that primary analyses were not likely confounded by unobserved time‐invariant factors. Although a large portion of the sample participated in a PCMH for 12 months in each year (“continuously enrolled”), this group did not appear to drive the results; models excluding 12‐month PCMH enrollees produced similar effect sizes. The same was true for models limiting the analysis to enrollees who were continuously enrolled in Medicaid for 12 months (Supplementary Table 1). Models using varying definitions of minimally adequate antidepressant usage (4 and 9 months) produced results similar to the primary results.

Limiting the sample to new PCMH enrollees without prior PCMH participation in the study period demonstrated that new enrollees exhibited somewhat larger improvements than the total sample. At 4 months, the overall population of PCMH enrollees had an increase in MAC of 1.8 percentage points (compared to those enrolled for only 1 month), while new users had a 4.4 percentage point increase. At 10 months, new users exhibited the greatest differential improvement from the total sample; new users were 24 percentage points more likely to receive MAC, while all PCMH enrollees exhibited an increase of 24 percentage points.

We also conducted sensitivity analyses to assess whether these results are being driven by new access to primary care, rather than the PCMH model itself (Table 4). In analyses stratified by total number of annual primary care visits, individuals with at least four primary care visits in a given year exhibited a statistically significant association between months of enrollment and receipt of MAC at the p < 0.05 level at 5 months of PCMH participation, while individuals with three or fewer primary care visits saw a statistically significant increase in receipt of MAC after 4 months. Among individuals in the high‐visit group, 12 consecutive months of PCMH participation were associated with a 20.0 percentage point increase in MAC (p < 0.01), compared to 13.7 percentage points in the low‐visit group (p < 0.01). In models assessing MAC for enrollees that had at least two PCP visits in the year prior to PCMH engagement, we also observed effects that are similar to the primary models in significance, direction and magnitude. The marginal effects in this subgroup were, in general, slightly larger than those of the full analytic sample, suggesting that the effect is not driven by solely new access to primary care.

TABLE 4.

Marginal effects of duration of PCMH participation stratified by primary care visits

	Minimally adequate care
	High (4+) visits	Low (<4) visits	2+ Visits in year prior to PCMH participation
# PCMH months
2	−0.00371	0.00858	−0.0249
	(0.0108)	(0.0108)	(0.0169)
3	0.00773	0.00927	−0.0171
	(0.0110)	(0.0107)	(0.0180)
4	0.0179	0.0141	0.0393*
	(0.0107)	(0.0106)	(0.0181)
5	0.0398**	0.0377**	0.0358
	(0.0110)	(0.0114)	(0.0209)
6	0.0990**	0.0892**	0.135**
	(0.0108)	(0.0114)	(0.0227)
7	0.128**	0.111**	0.122**
	(0.0109)	(0.0118)	(0.0260)
8	0.154**	0.113**	0.186**
	(0.0103)	(0.0114)	(0.0231)
9	0.153**	0.129**	0.162**
	(0.00995)	(0.0110)	(0.0230)
10	0.162**	0.131**	0.201**
	(0.00958)	(0.0107)	(0.0227)
11	0.180**	0.137**	0.171**
	(0.00925)	(0.0102)	(0.0258)
12	0.200**	0.168**	0.159**
	(0.00872)	(0.00928)	(0.0282)
Prior PCMH participation	−0.00250**	−0.00193**	0.00148
	(0.000187)	(0.000242)	(0.00136)
# CDPS conditions	0.00112**	0.000972*	6.74e‐05
	(0.000420)	(0.000481)	(0.00103)
Male	−0.0432**	−0.0196**	−0.0638**
	(0.00489)	(0.00513)	(0.0114)
Race	−0.147**	−0.131**	−0.119**
Black	(0.00426)	(0.00471)	(0.0109)
	−0.124**	−0.00129	−0.0627
Asian	(0.0384)	(0.0432)	(0.100)
	−0.0843**	−0.0837**	−0.00722
American Indian	(0.0137)	(0.0187)	(0.0405)
	−0.0985	0.0233	Omitted
Pacific Islander	(0.0994)	(0.133)
	−0.0597**	−0.0250*	−0.0165
Other	(0.0113)	(0.0143)	(0.0277)
	−0.0289	−0.0596**	0.0298
Hispanic ethnicity	(0.0152)	(0.0161)	(0.0406)
	0.00337**	0.00332**	0.00137**
Age	(0.000172)	(0.000194)	(0.000461)
	0.0548**	0.0460**	0.0628**
Medicaid months	(0.00142)	(0.00156)	(0.00627)
N (person‐years)	75,244	43,192	9405

Delta method standard errors in parentheses (**p < 0.01, *p < 0.05).

Abbreviations: CDPS, Chronic Illness and Disability Payment System; PCMH, patient‐centered medical home.

4. DISCUSSION

Given the prevalence of multiple chronic conditions among Medicaid enrollees, it is important to investigate models of care such as the PCMH that espouse potential to improve quality of care for people with multiple chronic conditions. We found that the duration of PCMH participation increases the likelihood of receiving minimally adequate MDD care. This association appears to be particularly strong for antidepressant use. This may be because primary care providers are able to prescribe antidepressants themselves, while psychotherapy is dependent on referral to behavioral health specialists, who may have capacity constraints that may dampen potential utilization.

There are several reasons that increased duration of PCMH participation may increase the likelihood of patients with multiple chronic conditions receiving adequate care for MDD. Increased duration of care and greater number of follow‐up visits may improve the likelihood of receipt of MAC in the long term because providers see patients more frequently and have more opportunities to address patient needs. In analyses stratified by high and low primary care utilization, we found that patients in the high utilization group saw a larger benefit from increased duration of PCMH enrollment. However, we also found that the association between increased duration of PCMH participation and receipt of MAC persisted in the low utilization group, suggesting that PCMHs offer additional benefits beyond increased contact with primary care providers. For example, improved care coordination may increase the likelihood that patients receive referrals to psychotherapists and/or adjustments to antidepressant prescriptions.

It is also important to note that we observed a small but statistically significant negative association between PCMH participation in a prior year and the outcomes of interest. While somewhat counterintuitive, this result may imply that while all enrollees exhibit an increase in receipt of MAC as the duration of their participation increases, new users may experience an initial “boost” in effect of PCMH participation compared to patients with previous PCMH participation. This theory is supported by the results of a sensitivity analysis, limiting the sample to enrollees with no prior PCMH participation during the study period; new PCMH enrollees demonstrated larger effects than the total sample. It is also possible that this result is not due to a depreciation in the effect of PCMHs but rather to the episodic or recurrent nature of MDD. Patients with extended periods of PCMH participation in prior years may be less likely to receive MAC because their symptoms resolved and treatment has been stopped.

Stratified analyses show that the effect of additional months of PCMH participation varies by race, ethnicity, and sex. This is consistent with previous research, which shows that while the PCMH model may improve overall quality of care, the effect of the model can be inconsistent across subgroups. ³¹ , ³² , ³³ Among Black, American Indian, and Hispanic enrollees, we found that additional months of PCMH participation were associated with increases in the likelihood of receiving MAC, although these groups were less likely to benefit than their non‐Hispanic white counterparts. Future research should further examine the potential for increased exposure to the PCMH model to reduce disparities in quality, as well as the role of potential mediating factors such as socioeconomic status.

There are several limitations to this analysis. We derived measures of antidepressant use from claims data rather than observing actual adherence. However, previous studies have shown that claims‐based measures of adherence produce similar results to measures based on self‐report. ²⁷ Unmeasured omitted variables may also bias estimates and measures of precision. However, our fixed‐effects analysis, which controlled for time‐invariant differences across individuals, yielded very similar results. Additionally, it is possible that that the observed association between PCMH enrollment duration and MAC is in part a function of increased access to primary care. However, sensitivity analyses excluding individuals with fewer than two primary care visits in the year prior to PCMH participation show results that are similar to the primary models, suggesting that effects are not driven solely by new primary care access. Moreover, although we required at least 6 months of Medicaid enrollment and controlled for variation in the number of additional Medicaid months observed in each year, it is not possible to determine whether the marginal effect of increased yearly Medicaid enrollment is due to increased observability of MAC in a claims‐based dataset or to an independent effect of Medicaid enrollment on receipt of care. That is, our analysis was unable to examine potential effects on treatments received outside the Medicaid program. Finally, because of the way this sample was constructed, these results are limited to individuals on Medicaid with some PCMH enrollment in each year and cannot be generalized to all patients with MDD and multiple chronic conditions.

Despite these limitations, our findings provide support for the idea that the PCMH model can improve quality of care for patients with multiple chronic conditions and MDD with increasing exposure time. Providers and policy makers should consider the positive effect of increased duration within PCMHs when designing initiatives. Future research in this area should examine factors associated with better PCMH engagement and strategies to facilitate patient engagement to achieve higher care quality.

Supporting information

Supplementary Table 1 Marginal effects among cohort with continuous Medicaid enrollment

Figure S1 Supplementary figure

Swietek KE, Domino ME, Grove LR, et al. Duration of medical home participation and quality of care for patients with chronic conditions. Health Serv Res. 2021;56(S1):1069‐1079. 10.1111/1475-6773.13710