The Effects of the Maryland Medicaid Health Home Waiver on Emergency Department and Inpatient Utilization Among Individuals with Serious Mental Illness

Abstract

Objective

The Maryland Medicaid health home program, established through the Affordable Care Act’s Medicaid health home waiver, integrates primary care services into specialty mental health programs for adults with serious mental illness (SMI). We evaluated the effect of this program on all-cause, physical, and behavioral health emergency department (ED) and inpatient utilization.

Method

Using marginal structural modeling to control for time-invariant and time-varying confounding, we analyzed Medicaid administrative claims data for 12,232 enrollees with SMI from October 1, 2012 to December 31, 2016; 3,319 individuals were enrolled in a BHH and 8,913 were never enrolled.

Results

Health home enrollment was associated with reduced probability of all-cause (PP: 0.23 BHH enrollment vs. 0.26 non-enrollment, p<0.01) and physical health ED visits (PP: 0.21 BHH enrollment vs. 0.24 non-enrollment, p<0.01) and no effect on inpatient admissions per person-three-month period.

Conclusion

These results suggest the Maryland Medicaid health home waiver’s focus on supporting physical health care coordination by specialty mental health programs may be preventing ED visits among adults with SMI, although effect sizes are small.

1. INTRODUCTION

Individuals with serious mental illnesses (SMIs) like schizophrenia and bipolar disorder experience elevated prevalence of physical health conditions, including diabetes mellitus, hypertension, and dyslipidemia, and die 10–20 years earlier than the overall population, primarily due to cardiovascular disease.(1–5) The obesogenic side effects of antipsychotic medications and higher levels of health risk behaviors, including tobacco smoking, poor diet, and physical inactivity, are key drivers of this enhanced risk.(6–9)

In addition, those with SMI experience barriers to accessing high-quality physical health care. A recent review found that sub-optimal quality of physical healthcare was particularly problematic for people with SMI covered by Medicaid, the insurer for nearly 70% of U.S adults with SMI.(10,11) Historically, mental and physical health care services, as well as substance use disorder (SUD) treatment and social services, have been financed and delivered in separate systems, creating barriers to access, inhibiting care coordination and impeding the sharing of health information.(12–15) As many people with SMI receive the majority of their services in the specialty mental health care system, the “behavioral health home” (BHH) model, in which primary care services are integrated into specialty mental health settings,(15) are increasingly being implemented in the U.S.

To date, however, there are limited evaluations of BHH models. Results of two randomized controlled trials (RCTs) suggest that the BHH model may hold promise for improving physical health care quality and health outcomes among people with SMI.(16,17) However, the few evaluations of real-world BHH programs outside of the RCT context have shown mixed results, with some studies suggesting that BHHs may decrease psychiatric hospitalizations and ED usage and improve rates of preventive screening and physical health outcomes(18–23) and others showing no effects on these outcomes.(19–21,23) Critically, many BHH programs in the U.S. are being implemented through the Affordable Care Act’s Medicaid health home waiver program, but none of these programs have been rigorously evaluated to date. The present study evaluating Maryland’s Medicaid Affordable Care Act BHH waiver program fills this research gap.

In 2010, the Affordable Care Act Medicaid health home waiver allowed states to create Medicaid heath homes that would provide care coordination and health promotion services for a designated set of beneficiaries with complex health needs.(24) Medicaid health homes provide services in six core areas: comprehensive care management, care coordination, health promotion, comprehensive transitional care and follow up, individual and family support, and referral to community and social support services.(24) The actual services provided and providers delivering these services differ across health homes, however examples of common services include: 1) comprehensive care management, e.g., developing a care plan within 30 days of BHH enrollment, 2) care coordination, e.g., appointment scheduling and communication with other providers 3) health promotion, e.g. dietary education, 4) comprehensive transitional care and follow up: medication reconciliation re following an acute care incident 5) individual and family support: home visits to support individuals with medication adherence and 6) referral to community and social support services: referrals to housing and transportation services. As of March 2019, nineteen states and DC used this waiver to implement a BHH for individuals with SMI.(25)

Maryland’s BHH program is implemented in community-based psychiatric rehabilitation programs (PRPs) and began in October 2013. Maryland PRPs serve individuals who have functional impairment due to mental illness, a high-need subset of individuals who account for approximately 15% of Medicaid beneficiaries with SMI.(26) PRPs provide psychosocial rehabilitation, case management, vocational services and life-skills training. Prior to the enactment of BHHs, Maryland PRPs were able to provide mental health, SUD and social service coordination through existing Medicaid reimbursement mechanisms.(27) The BHH model enabled PRPs to bill Medicaid for the coordination of physical health services as well.(28)

This study used rigorous non-experimental methods to examine the effects of Maryland’s BHH model on emergency department (ED) and inpatient utilization. Because of the program’s focus on integrating physical health care and behavioral health care, we hypothesized that BHH enrollment would be associated with reduced all-cause inpatient and ED utilization. As the new financing mechanism focuses on physical health integration, we anticipated no effects on behavioral health care utilization.

2. METHODS

2.1. Data

This study used Maryland Medicaid administrative claims data from a sample of 12,232 individuals with meaningful PRP use, defined as having greater than five uniquely dated PRP claims after October 1, 2013 but not limited to the baseline period. As indicated by enrollment in a PRP, all participants had a serious mental illness, defined as significant functional impairment resulting from mental illness.(29,30) Sample inclusion was also restricted to those aged 21 to 64. The baseline period was defined as October 1, 2012 to September 30, 2013 and the intervention period was defined as October 1, 2013 (the first possible date of BHH enrollment) thru December 31, 2016. Participants had to be enrolled in Medicaid during the baseline period and were censored from the sample if they disenrolled from Medicaid during the intervention period.

2.2. Measures

All measures were constructed at the person-three-month level.

2.2.1. Outcome Measures

We measured the effect of BHH enrollment on ED visits and inpatient admission. We examined: 1) the probability of having a visit in a 3-month period, a dichotomous measure of the outcome, and 2) the number of visits in a 3-month period, a count outcome. Visits resulting in hospital admission were excluded from the ED visit measures but were included in the inpatient admission measures. All outcomes excluded maternity- and infant-related utilization. We examined all-cause, physical health, and behavioral health ED and inpatient utilization. Primary diagnosis codes obtained from the Healthcare Effectiveness Data and Information Set were used to measure physical health versus behavioral health utilization.(31)

2.2.2. BHH Exposure Measures

In our main analyses, we measured BHH exposure using a 0/1 dichotomous indicator of enrollment in a BHH in a given person-3-month period. We assumed once an individual was enrolled in a BHH they remained enrolled for the remainder of the study, consistent with the idea of an intent-to-treat analysis.

In a sensitivity analysis, we measured cumulative BHH exposure, defined as total number of 3-month time periods an individual was enrolled in a BHH over the course of the 39-month intervention period.

2.2.3. Confounders

PRPs adopted the BHH model throughout the intervention period. Seventy percent (N=38) of the 54 BHHs operational during the study period began enrolling participants in the first year that Maryland adopted the health home waiver, between October 1, 2013 and September 30, 2014. Eleven percent (N=6) of BHHs enrolled their first participant in year two and 19% (N=10) enrolled their first participant in the final 15 months of the study. All Medicaid beneficiaries at PRPs implementing BHHs were eligible and had to actively consent to participation in the BHH program. In prior research, BHH staff reported attempting to enroll all eligible individuals, though enrollment occurred on a rolling basis. (28) Given this rolling enrollment, we accounted for both time-invariant and time-varying confounding.

Observed time-invariant confounders included: i) baseline age, ii) sex, iii) race/ethnicity, iv) mental illness diagnosis, and v) the PRP patient population size where the individual received the plurality of PRP services. Observed time-varying variables included: i) eligibility for Medicaid via disability, ii) SUD diagnosis, iii) co-morbidity (measured via Charlson index) iv) region of residence, v) number of PRP services received, vi) past inpatient utilization, vii) past ED utilization, viii) past ambulatory physical health care use and vix) indicator for enrollment in one of 9 possible Medicaid managed care organizations. Time-varying confounders were measured in the baseline period (October 2012 - September 2013) and during each 3-month period throughout the study.

2.3. Analytic Approach

To address potential time-varying confounding, we employed a marginal structural modeling approach. Prior studies have shown that using traditional regression adjustment techniques in these situations or adjusting for time-varying confounders in the intervention period can introduce bias in estimating intervention effects.(32–34)

Marginal structural models build upon the propensity score weighting technique that is widely applied in health policy evaluation. However, instead of applying an inverse probability of treatment weight to each individual, this technique estimates weights for every person-time point in the study period and then fits weighted pooled regression models to estimate effects. These models estimate the effect of an exposure of interest for the entire study population. In our case, the marginal structural modeling approach estimated the effects of BHH enrollment on outcomes using weighting to generate estimates as if the entire population had been enrolled in a BHH versus not enrolled.

First, we generated an enrollment weight. We estimated the inverse probability of BHH enrollment for each person-3-month period, adjusting for time-invariant and time-varying confounders measured up to that 3-month period. The enrollment weight for a given time period was the product of these inverse probabilities up until that point. Second, we calculated a censoring weight. Analogous to the enrollment weights, we calculated the inverse probability of being censored due to Medicaid disenrollment in a time period for every person-3-month observation, adjusting for time-invariant and time-varying variables measured to that point in time. The censoring weight at any given time point is the product of these probabilities up until that time point. The final weight for a given person-3-month was the product of the enrollment and censoring weight, and therefore accounted for a history of enrollment and confounders up until that time point. Detailed weight calculation information is available in Appendix A.

To measure the immediate effect of BHH enrollment on ED utilization and inpatient admissions outcomes, we used a data set with each observation representing a person-3-month period that included an indicator of BHH enrollment in that period. We used weighted logistic regression models to estimate predicted probabilities of ED/inpatient utilization outcomes comparing BHH enrollment to non-enrollment. We used weighted negative binomial regression models to estimate expected counts comparing BHH enrollment to non-enrollment. Standard errors were clustered by individual to account for multiple observations from the same individual.

In a sensitivity analysis, to measure the effect of cumulative BHH enrollment on outcomes, we used a dataset with one observation per person, where the exposure measured an individual’s length of enrollment in a BHH over the entire study period and outcomes were cumulative. We used weighted logistic regressions to estimate predicted probabilities of ED/inpatient utilization outcomes comparing 3 years of BHH enrollment to non-enrollment. We used weighted negative binomial regressions to estimate expected counts of ED/inpatient utilization outcomes comparing 3 years of BHH enrollment to non-enrollment.

3. RESULTS

The final sample included 151,408 person-3-month observations, which were drawn from 12,232 unique individuals. Fourteen percent of individuals were censored due to Medicaid disenrollment at some point during the study, but a majority were censored late in the study period, resulting in a loss of only 5% of potential person 3-month observations. Within the sample, 3,319 individuals were enrolled in a BHH and 8,913 were never enrolled in a BHH. Among those that were enrolled in a BHH, 2,273 (68%) were enrolled in the first year following Maryland’s adoption of the Medicaid health home waiver; 614 (19%) were enrolled in year two; and 432 (13%) were enrolled in final 15 months of the study period.

As detailed in Exhibit 1, the unweighted baseline characteristics between BHH enrollees and non-enrollees differed, further underscoring the need for weighting. Relative to non-BHH enrollees, BHH enrollees were more likely to have schizophrenia as their primary mental health diagnosis (63.3% vs 35.8%, p <0.01), to have qualified for Medicaid because of disability (83.6% vs 53.4%, p<0.01) and to have had at least one inpatient day in the baseline period (22.7% vs 16.3%, p <0.01). BHH enrollees were less likely than the non-enrollees to have at least one ED visit in the baseline period (46.7% vs 55.9%, p <0.01) or to have a diagnosis of a SUD (42.4% vs 49.4%, p<0.01).

Exhibit 1.

Unadjusted Baseline Characteristics of Sample

	Total Sample (n=12,232)		Ever Enrolled in Health Home (n=3,319)		Never Enrolled in Health Home (n=8,913)		P-Value
	Mean	SD	Mean	SD	Mean	SD
Age	42.2	11.3	44.6	11.2	41.3	11.2	<0.01
Charlson Index	0.8	1.5	0.9	1.5	0.8	1.5	0.01
	%	N	%	N	%	N
Female	56.4	6896	44.7	1482	60.7	5414	<0.01
Race
Black	56.3	6886	46.1	1530	60.1	5356	<0.01
White	38.3	4684	47.0	1560	35.0	3124	<0.01
Other	5.4	662	6.9	229	4.9	433	<0.01
Region of Residence
Baltimore City	36.6	4473	20.9	693	42.4	3780	<0.01
Baltimore Surrounding	25.5	3121	31.2	1037	23.4	2084	<0.01
Eastern	8.3	1010	19.6	652	4.0	358	<0.01
National Capital Area	16.6	2030	18.0	598	16.1	1432	0.01
Northwest	8.0	983	10.0	333	7.3	650	<0.01
Southern	5.0	609	0.1	4	6.8	605	<0.01
Other	0.0	6	0.1	2	0.0	4	0.75
Qualified for Medicaid via Disability	61.6	7536	83.6	2776	53.4	4760	<0.01
Primary Serious Mental Illness Diagnosis
Schizophrenia	43.2	5289	63.3	2100	35.8	3189	<0.01
Bipolar Disorder	33.1	4043	24.0	797	36.4	3246	<0.01
Major Depressive Disorder	22.7	2776	11.9	396	26.7	2380	<0.01
Other	1.0	124	0.8	26	1.1	98	0.09
Substance Use Disorder Diagnosis	47.5	5812	42.4	1407	49.4	4405	<0.01
No. Inpatient Days in Baseline Year
0 Days	81.9	10024	77.3	2566	83.7	7458	<0.01
1–5 Days	9.4	1150	11.0	366	8.8	784	<0.01
>5 Days	8.6	1058	11.7	387	7.5	671	<0.01
No. ED Visits in Baseline Year
0 Visits	46.6	5697	53.3	1769	44.1	3928	<0.01
1–5 Visits	45.4	5553	39.6	1313	47.6	4240	<0.01
>5 Visits	8.0	982	7.1	237	8.4	745	0.02
No. Outpatient Physical Health Visits in Baseline Year
0 Visits	11.6	1423	6.5	216	13.5	1207	<0.01
1–3 Visits	16.2	1985	8.8	291	19.0	1694	<0.01
>3 Visits	72.1	8824	84.7	2812	67.5	6012	<0.01

Note: Categories in primary mental health diagnosis are mutually exclusive. Other includes individuals without schizophrenia, bipolar disorder, and major depressive disorder but other serious psychiatric conditions.

The effectiveness of weighting at balancing these characteristics between the BHH-enrolled and non-enrolled participants was assessed at every 3-month time period using the absolute standardized mean difference, or the absolute difference in means between the BHH enrollees and non-enrollees divided by the standard deviation of the entire sample. As seen in Exhibit 2, weighting improved covariate balance: after weighting the average absolute standardized mean difference of baseline covariate values between the BHH and non-BHH groups was near or below 20% – a suggested cut-point used to mark good covariate balance(34) – for all covariates, with the exception of number of PRP services, Medicaid eligibility via disability and one region of residence, all of which were greatly improved. Appendix B shows the absolute standardized mean differences for each time period separately.

Exhibit 2. Average Absolute Standardized Mean Difference of Baseline Covariates in Weighted and Unweighted Samples.

Note: The absolute standardized bias difference for each baseline covariate was calculated separately using weights for each 3-month time period. This figure reports the average of this value across all 13 time periods. The absolute standardized bias difference is defined as the absolute difference in means between the BHH-enrolled and non-BHH-enrolled groups standardized by the standard deviation of the entire sample. Vertical line at 20%, a suggested threshold for good balance given in Griffin et al.

Exhibit 3 shows that, among the study population with SMI, in a given three-month time period BHH enrollment was associated with an overall reduction in the odds of having an all-cause ED visit compared to no BHH enrollment (OR: 0.87, p<0.01). This corresponded to a 3-percentage point reduction in the predicted probability of having an all-cause ED visit in a 3-month period (PP: 0.23 BHH enrollment vs. 0.26 No BHH enrollment, p<0.01), or an 12 percentage point reduction in the probability of having and ED visit over a year. This reduction in all-cause ED utilization was driven by a reduction in the predicted probability of having a physical health ED visit in a 3-month period among BHH enrollees (PP: 0.21 BHH enrollment vs. 0.24 No BHH-enrollment, p<0.01).

Exhibit 3.

Effect of Health Home on Probability of Having Emergency Department and Inpatient Utilization Event in a 3-month Period (n=151,408)

	Odds Ratio OR [95%CI]	Predicted Probability in 3 Months if Enrolled in Health Home PP (95% CI)	Predicted Probability in 3 Months if Not Enrolled in Health Home PP (95% CI)
Any Emergency Department Visit
All-Cause	0.87* [0.81,0.93]	0.23^ [0.22, 0.24]	0.26 [0.25, 0.26]
Physical Health	0.86* [0.80,0.92]	0.21^ [0.20, 0.22]	0.24 [0.23, 0.24]
Behavioral Health	1.04 [0.90,1.19]	0.04 [0.03, 0.04]	0.04 [0.04, 0.04]
Any Inpatient Admission
All-Cause	0.97 [0.88,1.07]	0.07 [0.07, 0.08]	0.08 [0.07, 0.08]
Physical Health	0.94 [0.82,1.07]	0.04 [0.03, 0.04]	0.04 [0.04. 0.04]
Behavioral Health	1.01 [0.88,1.15]	0.04 [0.04, 0.05]	0.04 [0.04, 0.05]

Notes: This table shows results from an analysis at the person 3-month level, for the following logistic regression: Pr(Outcome Event_ij)= B₀ + B₁(HealthHome_ij) + B₂(Time), where HealthHome_ij represents any enrollment in a given person-3 month period. Behavioral Health outcomes include both mental health and substance use disorder outcomes.

^*OR is not equal to 1.0 at p<0.05

^{^}Predicted probability (marginal effect) different from Not Enrolled in Health Home at p<0.05

There was no effect of BHH enrollment on the probability of having an inpatient utilization in a 3-month period. Exhibit 4 shows that there was no effect of BHH enrollment on the number of ED visits or inpatient admissions.

Exhibit 4.

Effect of Health Home on Number of Emergency Department and Inpatient Utilization Events in a 3-month Period (n=151,408)

Outcome	Incidence Rate Ratio IRR [95% CI]	Expected Count in 3 Months if Enrolled in Health Homes Count [95% CI]	Expected Count in 3 Months if Not Enrolled in Health Homes Count [95% CI]
Emergency Department Visits
All-Cause	0.93 [0.79,1.10]	0.46 [0.39, 0.54]	0.50 [0.47, 0.52]
Physical Health	0.90 [0.78,1.04]	0.39 [0.33. 0.44]	0.43 [0.41, 0.45]
Behavioral Health	1.13 [0.67,1.91]	0.08 [0.04, 0.11]	0.07 [0.06, 0.08]
Inpatient Admissions
All-Cause	0.96 [0.85,1.07]	0.09 [0.08. 0.10]	0.10 [0.09, 0.10]
Physical Health	0.96 [0.81,1.13]	0.04 [0.04, 0.05]	0.04 [0.04, 0.05]
Behavioral Health	0.95 [0.82,1.11]	0.05 [0.04, 0.06]	0.05 [0.05, 0.06]

Notes: This table shows results from an analysis at the person 3-month level, for the following negative binomial regression: Number of Outcome Events= B₀ + B₁(HealthHome_ij) + B₂(Time), where HealthHome_ij represents any enrollment in a given person-3 month period. Behavioral Health outcomes include both mental health and substance use disorder outcome

^*IRR is not equal to 1.0 at p<0.05 ^ Expected count (marginal effect) is different from not enrolled in HH at p<0.05

Sensitivity analysis results (Appendix C) showed that longer BHH enrollment was associated with stronger effects on ED visits: the predicted probabilities of having an ED visit after three years of BHH enrollment was 0.75, compared to 0.79 no BHH enrollment over the same three years (p<0.01). As in our main analysis, this overall reduction was driven by a reduction in physical health ED visits among BHH enrollees. While our main analyses showed no immediate effects of BHH enrollment on inpatient utilization, sensitivity analysis results showed that every additional 3-months of BHH enrollment was associated with a small increase in the odds of having an inpatient visit (OR: 1.01, p<0.01) over the 39-month study period (Appendix C); this small increase was driven by an increase in behavioral health admissions. As in the main analysis, the sensitivity analyses showed no effect of length of enrollment on the number of ED or inpatient utilizations (Appendix D).

4. DISCUSSION

The Maryland BHH program created through the Affordable Care Act Medicaid health home waiver was associated with small reductions in all-cause ED visits, driven by reductions in physical health ED visits. This finding suggests that the program’s focus on physical health care coordination may be preventing enrollees from ever experiencing a physical health ED visit or re-directing individuals with non-emergent physical health conditions away from the ED, but the size of these effects is small. Further research is needed to disentangle if reductions in ED use are driven by these types of positive events or are driven by reductions in access needed emergency care. Future research should examine if these reductions in physical health ED usage are driven by reductions in ED visits for ambulatory care sensitive conditions and are accompanied by increases in ambulatory care usage, which would suggest a shift in service use towards more appropriate settings of care.

While BHH enrollment reduced the probability of having any physical health ED visit, it was not associated with the overall number of physical health ED visits. This may suggest that the effect of BHHs is concentrated among healthier individuals who are prevented from making any visit to the ED, rather than reducing the number of ED visits in a given time period experienced by frequent ED users. It could suggest that more targeted intervention is needed to reduce ED use for high utilizers. Future research should examine if BHH effects differ across other subpopulations, such as individuals with SMI and multiple co-morbid medical conditions or older adults with SMI, and examine other outcomes relevant to high utilizers of the ED, such as follow-up after an ED visit.

While the main analysis found no association between BHH enrollment and inpatient admissions within a 3-month period, sensitivity analysis findings suggest that longer duration of BHH enrollment may be associated with a small increase in the probability of having a behavioral health inpatient admission. While this finding was small in magnitude and should be interpreted with caution, it may signal that the reductions in physical health ED usage associated with BHH enrollment may be coming at the expense of behavioral health service coordination in the psychiatric rehabilitation programs implementing BHHs in Maryland. Prior to BHH implementation, PRPs already engaged in mental health care coordination, including psychotropic medication management and home visiting, and some PRPs also performed substance use disorder care coordination.(28) A case study of the Maryland BHH program by McGinty et al. found that some BHH providers reported being unable to complete all required BHH services by themselves and therefore relied on existing PRP staff to assist in providing these services.(28) This task-shifting may result in PRP staff spending more time on physical health care coordination, and less time on the behavioral health care coordination activities for which they have been and continue to be responsible. Alternatively, these reductions may be a result of BHH staff identifying previously unmet behavioral health needs, resulting in increases in behavioral health care use. Further exploration of this policy’s effects on staffing and PRP services is needed to test these hypotheses and could inform policy decisions on the sufficiency of the BHH monthly reimbursement rate, staffing and service requirements.

These findings add to a growing body of literature on BHH impacts on individuals with SMI. While the recent HOME RCT study by Druss et al. and an evaluation of two SAMHSA Primary and Behavioral Health Care Integration (PBHCI) grantees found no changes in ED usage associated with BHH enrollment,(16,21) other BHH evaluations have found decreases in ED utilization.(19,23) Mixed findings are also seen when looking at hospitalizations, with some BHHs reporting decreases(19,21,22) and other studies reporting no effect.(16,19,21) These mixed findings may reflect the variation across different models and the populations they serve.

4.1. Limitations

This study should be considered in the context of several limitations. First, this analysis used administrative claims data only reflecting services paid through Maryland Medicaid. To address this, we conducted a sensitivity analysis excluding the 32% of individuals who were dually enrolled in Medicare; results were consistent. Second, the use of marginal structural modeling assumes no unobserved confounding. Third, this analysis did not examine whether BHH improved quality of care, provider-reported outcomes or health outcomes; future research in these areas is needed. Finally, while these results are illustrative of the potential effects of a Medicaid BHH based in community mental health programs, it is an evaluation of a specific state’s program, and therefore generalizability to other states may be limited.

4.2. Conclusion

This study used rigorous quantitative methods applying strong controls for time-varying confounding to evaluate the effects of a BHH implemented through the Medicaid health home waiver. Maryland Medicaid BHH enrollment was associated with small reductions in physical health ED visits and no effect on inpatient admissions in a three-month period. This finding suggests that by financing physical health care coordination in community mental health organizations may reduce costly ED visits, but further evidence is needed to translate these efforts to meaningful magnitudes.

HIGHLIGHTS.

Health home associated with small reductions in all-cause emergency department use.

Reductions in all-cause emergency department use driven by somatic ED reductions.

Health home not associated with inpatient utilization in the short term.