Effects of Medicaid Health Homes among People with Substance Use Disorder and Another Chronic Condition on Health Care Utilization and Spending: Lessons from New York State

Abstract

Introduction

New York State implemented a Health Homes (HH) care management program to facilitate access to health services for Medicaid enrollees with multiple chronic conditions. This study assessed the impact of HH on health care utilization outcomes among enrollees who have substance use disorder (SUD).

Methods

Using HH enrollment data and Medicaid claims data 1 year before and after enrollment, this study compared HH enrollees who enrolled between 2012 and 2014 to a statistically matched comparison group created with propensity score methods. Analyses used generalized gamma models, logistic regression models, and difference-in-differences analyses to assess the impact of HH on general (all-cause) health care and SUD-related outpatient, emergency department (ED), hospitalization, and detoxification utilization as well as total Medicaid cost.

Results

The sample consisted of 41,229 HH enrollees and a comparison group of 39,471 matched patients. HH-enrolled patients who had SUD utilized less SUD-related ED services (average marginal effect (AME) = −1.85; 95% CI=−2.45, −1.24), SUD-related hospitalizations (AME=−1.28; 95% CI: −1.64, −0.93), and detoxification services (AME=−1.30; 95% CI=−1.64, −0.96), relative to the comparison group during the 1 year post-HH enrollment. SUD-related outpatient visits did not change significantly (AME=−0.28; 95% CI=−0.76, 0.19) for enrollees, but general health care outpatient visits increased (AME=1.63; 95% CI=1.33, 1.93).

Conclusion

These findings provide preliminary evidence that care management programs can decrease ED visits and hospitalizations among people with SUD.

1. Introduction

Substance use disorders (SUD) are chronic medical conditions that require ongoing disease management.¹ However, currently the U.S. Healthcare system does not regularly integrate behavioral health with other medical care. For instance, continuity of care between hospitalization and outpatient SUD treatment continues to be a major problem.^1,2 Therefore, a minority of patients receive the SUD treatment they need.³ Poor integration is anathema to patient-centered care and leads to wasteful spending due to poor patient self-management and ineffective care.³

SUD is highly prevalent among the costliest Medicaid enrollees.⁴ States that offer Medicaid to their indigent citizens have fiscal incentives to address the specific healthcare needs and service use patterns of these individuals.⁵ Estimates from New York State Medicaid data show that high-spending patients with SUD have a complex set of comorbid medical and mental health conditions that require coordination of care across multiple providers.^6,7 Research shows that care management integrates the health care of high-utilizing individuals with SUD and mental health disorders.⁸

Health Homes (HH) is a provision of the Affordable Care Act (ACA) that established a platform for state Medicaid programs to improve care and reduce spending for treating poorly managed chronic illnesses.⁹ HH are comprised of networks of healthcare, social, and community service providers who collaborate to provide patients with comprehensive, whole-person care.¹⁰ HH are not physical places, rather programs that bring together a group of providers to jointly provide patients with care management for chronic health conditions, including the coordination of care within a complex healthcare system.^3,11,12 HH expands on the Patient-Centered Medical Home (PCMH) model and is designed to assist high-need, high-spending patients in managing their own chronic conditions as well as guiding them through a complicated and fragmented healthcare system.^3,11,13,14 As of 2014, 16 states had implemented the HH program, including New York State.¹⁵

In 2012, New York State launched one of the largest HH programs in the United States to foster better integration, quality, and cost of healthcare. Medicaid recipients with HIV/AIDS, a high-severity mental health disorder, or two or more chronic health conditions, of which one could be an SUD, were eligible to participate in the program.^12,16 Each HH serves patients within a particular region.¹² Networks of providers who apply to be New York HH organizations must already serve Medicaid clients and stipulate procedures to provide comprehensive care management, care coordination, health promotion, comprehensive transitional care, patient and family support, referral to community and support services, quality measures reporting, and have network capacity for integration of health information technology.¹⁷ One of the central functions of the HH is to contract with care management vendors at the behest of the network. The care management vendors conduct outreach, screening, health promotion, and coordination of care for eligible Medicaid members.¹⁷

As of 2014, New York State had 34 HH that contracted with 295 care management agencies and served 128,332 Medicaid members with complex chronic healthcare needs. The State identified potentially eligible Medicaid recipients and provided a list to each HH for outreach and engagement.¹⁶ Additionally, a provider or community member could refer eligible patients to HH, subject to verification of the person meeting criteria. Each HH assigned each enrollee a care manager, who assisted in assessment, development of care plans, and coordinating services. New York State established basic criteria for HH care management services but allowed latitude for the HH lead entities to establish administrative procedures to oversee clinical care. HH received a monthly fee per enrollee.¹⁸

This study evaluates the effects of the New York State HH program among individuals with an SUD from 2012 to 2014 by examining the association between HH enrollment and SUD-related service utilization, general (all-cause) health care utilization, and Medicaid spending for individuals with a primary SUD diagnosis. The conceptual model for the study draws from the Chronic Care Model (CCM).^19,20 The CCM states that individuals with chronic conditions require a different system of care than that provided by traditional medicine. These individuals require consistent healthcare engagement and coordination of services across a diverse team of specialist providers. The model postulates that providing ongoing ambulatory-based care in a manner that addresses the disease management needs of these individuals will reduce the likelihood of acute care episodes, such as emergency department (ED) visits and hospitalizations. Furthermore, individuals with SUD who receive ongoing ambulatory-based addictions treatment will be better engaged in treatment for other chronic medical conditions and have lower likelihoods of acute care episodes that drive high levels of spending.

2. Materials and Methods

2.1. Data and Study Population

New York State Medicaid claims, encounter, and eligibility data formed the basis of the analytical dataset. In addition, we used the New York State Department of Health registry of HH members to identify enrollment and discharge dates. The Center on Addiction Institutional Review Board approved the study protocol (#213).

The analytical sample included 41,229 individuals aged between 18 and 64 years, diagnosed with an SUD and one or more chronic medical conditions (e.g., diabetes, hypertension, asthma, HIV/AIDS), and enrolled in the New York State Medicaid HH program between 2012 and 2014. The only exclusion criterion for our analysis was co-insurance with Medicare due to the incompleteness of healthcare data for these individuals. We derived a comparison group of 39,471 individuals with comparable health and healthcare use characteristics from the Medicaid data using propensity score methods described below.

2.2. Outcomes

The study examined the following outcomes associated with HH enrollment: SUD-related and medical health care utilization before and after HH enrollment for outpatient care, ED visits, hospitalizations, detoxification, and receipt of pharmacotherapy for opioid use disorder (OUD). We considered Medicaid claims to be SUD-related if it also included a primary SUD diagnosis. We treated these measures as binary variables indicating whether there was any use of the service type during the period. Secondary analyses examined general health care utilization, including the use of ED visits (binary), hospitalizations (binary), total Medicaid spending (natural units), and counts of outpatient visits during the period. We defined outpatient care, ED visits, and hospitalizations using a combination of New York State Medicaid-specific payment rate codes, diagnostic related grouping (DRG) codes, and procedure codes (i.e., current procedure terminology (CPT) or healthcare common procedure coding system codes (HCPCS)). We defined detoxifications, outpatient care, and hospitalizations using New York State-specific Medicaid payment codes, procedure codes, and DRG codes. We defined receipt of pharmacotherapy for OUD (methadone, buprenorphine, and naltrexone) using New York State-specific Medicaid payment codes, procedure codes, and national drug codes (NDC).

We derived annual healthcare spending from Medicaid payments for billing paid directly by the State and claims reported by managed care companies (MCO). The MCO-reported data contained approximately 9% of claims with $0 or missing payment data and variation in pricing at the procedure and hospital DRG level that reflected billing reconciliation processes that varied across plans and providers. Drawing from methods used by New York State to negotiate managed care contracts, we standardized pricing for healthcare procedures to reduce distortions related to billing practices.^21-24 To establish prices by type of service, we classified claims by procedure codes, universal billing (UB)-92 revenue codes, all patient refined (APR)-DRG codes, and NDC. We set the pricing floor at the 25^th percentile for procedures or DRG. We set the pricing ceiling at 2.0 times the interquartile range plus the value for the 75^th percentile for inpatient claims and 1.5 times the interquartile range plus the value for the 75^th percentile for non-inpatient procedures. We imputed encounter claims with no reported payment using the mean spending for the associated type of service. We then summed spending for all claims by individual to create total Medicaid spending for each individual.

2.3. Covariates

Analyses controlled for patient-level sociodemographic variables such as age, gender, race/ethnicity, residence status, and number of months eligible for Medicaid in our models. We also controlled for various substance use disorders (OUD, alcohol use disorder (AUD), stimulant use disorder (cocaine or meth/amphetamine use disorder), and cannabis use disorder), severe mental illness (schizophrenia, major depression, bipolar disorder, and other psychosis), HIV, and other chronic physical conditions (heart disease, diabetes, asthma, chronic obstructive pulmonary disease, hypertension, and hepatitis). We identified HIV/AIDS status by using ICD-9-CM diagnoses codes, HIV-specific procedure codes, and NDC codes for antiretroviral treatments. We defined the remaining conditions using diagnoses codes (International Classification of Diseases (ICD-9-CM and ICD-10-CM)).

2.4. Statistical Analyses

We generated a statistically matched comparison group based on propensity score in the month of enrollment for each HH beneficiary.^25-27 In brief, we modeled the monthly likelihood of enrolling in a HH among 2,811,174 adults in Medicaid with multiple chronic health conditions. For the propensity score model, we included sets of demographic (e.g., age, gender), chronic condition diagnoses (e.g., diabetes, HIV/AIDS, bipolar disorder), healthcare utilization (e.g., counts of healthcare visits by service type over the prior month, 3 months and year), and Medicaid spending over the prior 12 months. To select a match for each HH participant, we computed monthly propensity scores for the probability of enrollment within each county: Pr[y_it∣ D_i, U_it], where y_it is an indicator of whether person i enrolled in the HH program in month t, D_i is a matrix of time invariant variables at the person level (e.g., gender, race), and U_it is a matrix of time-varying variables assessing different dimensions of medical, severe mental illness, and SUD service utilization of the 1 year prior to month t (e.g., frequency of outpatient, ED use). We derived a matched comparison for each HH client in the month of HH enrollment by selecting the closest propensity score for that specific month t among all individuals who never enrolled in the HH program. We conducted matching with replacement and stratified by gender and county of residence. We conducted propensity score modeling iteratively, introducing or removing variables (e.g., healthcare service utilization over varying periods) to maximize matching balance. We also examined the use of inverse probability of treatment weighting across the full population sample but the matching method provided better balance across covariates between enrolled and comparison individuals.²⁸

We used difference-in-differences (DID) models to estimate the effects of HH across all outcomes. Since HH enrollment occurred from 2012 to 2014, we defined 2 periods (pre-enrollment and post-enrollment) to reflect 1 year before or after the start of participation in HH. We employed logistic regression to examine changes in healthcare service utilization from baseline to follow-up for SUD-related outpatient, ED, detoxification visits, and the receipt of pharmacotherapy for OUD. Due to differences in distribution that included having extreme outliers and small proportions of zeros, we modeled costs and outpatient utilization (where over 90% of the sample had at least one all-cause OP visit with large variance in number of visits) using generalized gamma models, which offer flexibility in modeling non-Gaussian outcomes by computing three parameters to fit the observed distribution: location, scale, and shape.²⁹ We employed logistic regression to examine changes in general ED visits and hospitalizations. We conducted all modeling in Stata 16 using commands for longitudinal data (StataCorp, College Station, TX, USA). Each analytical dataset contained two observations per individual, one for the year before enrollment and another for the year after enrollment. Models included indicators for time, treatment, and interactions between time and treatment to estimate the difference in difference. All models adjusted for the above covariates and accounted for clustering by Medicaid ID and month. We used the margins post-estimation command in Stata to generate difference-in-difference HH effect estimates.³⁰

3. Results

Table 1 shows the baseline (pre-HH enrollment) demographic and clinical characteristics of our sample, including the statistically matched comparison group. The standardized differences between HH enrollees and comparisons show good matching (standardized difference less than 0.10) across variables, with small exceptions for prevalence of AUD and rate of hospitalization. HH enrollees were mostly men (57%) and individuals over 40 years old (65%). Nearly half lived in New York City (55%). A quarter of enrollments were Latinx (25%), and approximately a third were black (35%) and white (30%) each. HH enrollees had an average baseline annual Medicaid cost of $26,890, which falls within the top 10% of all Medicaid patients.⁷

Table 1.

Baseline characteristics of the patients enrolled in the NYS HH program and comparison group, 2012-2014

		Group,%
Variable		HH Enrolled n=41,229	Comparison N=39,471	Standardized differences
Age category	18-29	16.1	17.3	0.03
	30-39	18.9	20.2	0.03
	40-49	28.1	27.3	0.02
	50-59	30.5	28.4	0.05
	60-65	6.3	6.8	0.02
Sex	Male	57.3	57.8	0.01
Race/ethnicity	Non-Hispanic White	30.3	33.1	0.06
	Black	34.5	35.1	0.01
	Latinx	25.3	24.3	0.02
	Other	5.2	3.9	0.06
	Unknown	4.8	3.7	0.05
Residence, NYC vs. ROS	NYC	55.0	53.9	0.02
Substance Use Disorder	Opioid use disorder	38.4	37.0	0.03
	Alcohol use disorder	50.1	46.6	0.12
	Stimulant use disorder	51.6	47.5	−0.08
	Cannabis use disorder	24.2	20.9	0.08
	Hallucinogen use disorder	1.1	1.0	0.01
Medical Condition	Severe mental illnesses	38.2	40.9	0.05
	Heart disease	33.5	31.8	0.03
	Diabetes	18.3	17.2	0.03
	Asthma	10.8	9.6	0.04
	HIV/AIDS	18.3	18.1	0.00
	COPD	10.2	9.4	0.02
	Hypertension	35.8	33.8	0.04
	Hepatitis	13.8	11.8	0.06
General Medicaid utilization	Total Medicaid cost, $	$26,890 (36,838)	$26,524 (39,720)	−0.01
	Outpatient visit	90.1	91.5	0.05
	ED visit	59.1	56.2	0.08
	Hospitalization	42.3	37.8	0.12
SUD Outcomes	Outpatient visit	28.7	25.3	0.08
	ED visit	20.1	17.8	0.06
	Hospitalization	16.1	13.8	0.06
	Detoxification episode	15.0	13.0	0.06
	Pharmacotherapy for opioid use disorder*	30.4	31.4	0.02

Abbreviations: NYC, New York City; ROS, rest of the state; HIV/AIDS, human immunodeficiency virus/acquired immunodeficiency syndrome; COPD, chronic obstructive pulmonary disease; ED, emergency department; SUD, substance use disorder.

^*Among patients who have opioid use disorder diagnosis

Overall, the population had a notable prevalence of chronic medical conditions and high levels of healthcare services utilization. Both psychiatric and medical conditions were highly prevalent, speaking to some of the complex healthcare needs and challenges for coordinated care. Thirty-eight percent had OUD, 50% had AUD, 51% had stimulant use disorder, and 24% had cannabis use disorder. Fifty-nine percent of enrollees had at least one ED episode, 20% of which were related to SUD. Forty-two percent had at least one hospitalization in the prior year, 16% of which were related to SUD. Another 15% of enrollees had a detoxification episode. Only a quarter of individuals had a visit with an SUD-related outpatient treatment provider, and only 30% of individuals with an OUD diagnosis received pharmacotherapy (methadone, buprenorphine, or naltrexone).

Table 2 presents DID findings for SUD-related healthcare outcomes. While the probability of having an outpatient treatment for SUD event did not significantly change (average marginal effect (AME) = −0.28; 95% CI= −0.76, 0.19), ED visits decreased by −1.85 percentage points (95% CI=−2.45, −1.24) and hospitalizations decreased by 1.30 percentage points (95% CI: −1.64, −0.93). The probability of detoxification decreased by 1.30 percentage points (95% CI= −1.64, −0.96). The probability of receipt of pharmacotherapy for OUD did not significantly change (AME=0.21; 95% CI=−0.39, 0.81). Among the model control variables, it was notable that having AUD was associated with a higher probability of SUD-related ED visits (10.48 percentage points), hospitalizations (16.30 percentage points), and detoxification (14.28 percentage points).

Table 2.

Average Marginal Effects (AME) for SUD-related Health Care Utilization ^{a, b}

SUD-related Outcome Measures	Outpatient visit with an SUD diagnosis AME (95% CI)	Any ED visits with an SUD diagnosis AME (95% CI)	Hospitalization with an SUD diagnosis AME (95% CI)	Detoxification visit AME (95% CI)	Pharmacotherapy for opioid use disorderc AME (95% CI)
DID Treatment Effect for HH	−0.28 (−0.76, 0.19)	−1.85 (−2.45, −1.24)	−1.28 (−1.64, −0.93)	−1.30 (−1.64, −0.96)	0.21 (−0.39, 0.81)
Model Covariates
Age category
18-29	Ref	Ref	Ref	Ref	Ref
30-39	1.69 (0.93, 2.45)	−0.08 (−1.04, 0.87)	0.55 (−0.02, 1.11)	0.63 (0.08, 1.19)	5.02 (3.26, 6.78)
40-49	2.36 (1.61, 3.10)	−1.72 (−2.64, −0.79)	1.10 (0.54, 1.66)	0.86 (0.33, 1.40)	5.47 (3.80, 7.14)
50-59	1.66 (0.88, 2.44)	−3.14 (−4.10, −2.18)	−0.10 (−1.56, −0.40)	−1.24 (−1.79, −0.69)	7.53 (5.77, 9.30)
60-65	−1.11 (−2.24, 0.03)	−4.11 (−5.43, −2.78)	−2.96 (−3.74, −2.17)	−3.35 (−4.07, −2.62)	10.67, 8.37, 12.98)
Male	3.25 (2.77, 3.72)	3.25 (2.72, 3.79)	4.32 (3.98, 4.66)	3.87 (3.54, 4.20)
Race/ethnicity
Non-Hispanic White	Ref	Ref	Ref	Ref	Ref
Black	0.37 (−0.26, 1.00)	−4.86 (−5.58, −4.14)	−2.83 (−3.28, −2.38)	−3.03 (−3.46, −2.60)	−10.45 (−11.76, −9.14)
Latinx	−2.86 (−3.57, −2.16)	−2.57 (−3.34, −1.79)	−3.94 (−4.44, −3.45)	−3.34 (−3.81, −2.87)	5.04 (3.74, 6.34)
Other	2.32 (1.15, 3.48)	−1.06 (−2.45, 0.34)	−1.36 (−2.23, −0.49)	−1.71 (−2.54, −0.88)	−5.54 (−8.06, −3.03)
Unknown	3.26 (2.03, 4.48)	−0.64 (−2.02, 0.73)	1.14 (0.32, 1.97)	1.34 (0.56, 2.13)	−3.01 (−5.32, −0.71)
NYC	2.44 (1.88, 2.99)	4.38 (3.74, 5.01)	3.82 (3.42, 4.22)	4.73 (3.54, 4.20)	18.18 (16.99, 19.37)
Opioid use disorder	−11.88 (−12.5, −11.22)	8.59 (7.84, 9.35)	4.87 (4.44, 5.31)	5.33 (4.90, 5.75)
Alcohol use disorder	15.27 (14.80, 15.74)	10.48 (9.81, 11.16)	16.30 (15.91, 16.70)	14.28 (13.89, 14.66)	−14.63 (−15.63, −13.63)
Stimulant use disorder	8.24 (7.60, 8.88)	3.34 (2.61, 4.07)	6.38 (5.94, 6.83)	3.37 (2.95, 3.80)	32.34 (30.62, 34.05)
Cannabis use disorder	12.86 (12.30, 13.42)	−2.00 (−2.63, −1.36)	−1.94 (−2.33, −1.54)	−3.12 (−3.50, −2.73)	−18.12 (−19.29, −16.96)
Severe mental illnesses	1.07 (0.57, 1.57)	2.34 (1.79, 2.89)	0.19 (−0.16, 0.54)	1.06 (0.73, 1.39)	−6.15 (−7.07, −5.23)
Heart disease	−1.42 (−1.96, −0.89)	7.39 (6.81, 7.96)	2.78 (2.42, 3.14)	3.40 (3.06, 3.75)	−2.42 (−3.37, −1.46)
Diabetes	−0.88 (−1.54, −0.22)	−1.41 (−2.15, −0.67)	−2.29 (−2.77, −1.81)	−1.94 (−2.39, −1.49)	−1.98 (−3.16, −0.81)
Asthma	−1.40 (−2.20, −0.60)	1.87 (1.03, 2.71)	−1.01 (−1.57, −0.46)	−0.67 (−1.20, −0.15)	−0.45 (−1.78, 0.87)
HIV/AIDS	−2.22 (−2.89, −1.56)	−0.48 (−1.18, 0.22)	0.53 (0.08, 0.09)	1.12 (0.70, 1.55)	−2.56 (−3.65, −1.48)
COPD	−0.79 (−1.63, 0.04)	−0.29 (−1.17, 0.59)	−0.05 (−1.10, 0.04)	−0.48 (−1.02, 0.07)	−0.56 (−2.00, −0.88)
Hypertension	−2.37 (−2.94, −1.80)	0.47 (−0.14, 1.09)	−0.07 (−1.10, −0.31)	−0.54 (−0.92, −0.17)	−0.25 (−1.27, 0.76)

Abbreviations: ED, emergency department; DID, difference-in-differences; CI, Confidence Interval

^aAll models were adjusted for age, sex, race/ethnicity, chronic conditions (HIV/AIDS, severe mental illness, diabetes, hepatitis, hypertension, heart disease, asthma), prior care management enrollment status, duration of Medicaid eligibility, and year of enrollment.

^bLogistic regression model results, presenting AME in percentage point units

^cAmong patients who have opioid use disorder diagnosis

^*Statistically significant at p<0.05

^**p<0.01

^***p<0.001

Table 3 presents DID findings for all-cause healthcare cost and service utilization. The number of outpatient visits increased (AME=1.63; 95% CI=1.33, 1.93), while total Medicaid healthcare costs decreased (AME=−848.67; 95% CI=−1196.24, −501.08). The probability of ED visits for HH enrollees decreased by 2.16 percentage points (95% CI=−2.85, −1.45) and hospitalizations decreased by 2.53 percentage points (95% CI=−2.93, −2.14). Among the model control variables, it was notable that individuals with OUD experienced substantial differences in Medicaid spending (AME= 2094.80; 95% CI=1752.48, 2437.12) and the number of outpatient visits was substantial. On average, HH enrollment led to 12.74 additional visits for OUD patients (95% CI=12.19, 13.29). For individuals with AUD, the probability of having ED visits and hospitalizations increased by 12.78 and 12.65 percentage points, respectively.

Table 3.

Average Marginal Effects (AME) for General Health Care Cost and Utilization^a

Outcome Measures	Total Medicaid cost of care per patientb AME (95% CI)	All-cause outpatient visitsb AME (95% CI)	All-cause ED visitsc AME (95% CI)	All- cause hospitalizationsc AME (95% CI)
DID Treatment Effect for HH	−848.67 (−1196.25, −501.08)	1.63 (1.33, 1.93)	−2.16 (−2.85, −1.45)	−2.53 (−2.93, −2.14)
Model Covariates
Age category
18-29	Ref	Ref	Ref	Ref
30-39	−505.49 (−957.60, −53.37)	3.36 (2.87, 3.86)	−4.43 (−5.47, −3.39)	−6.34 (−7.24, −5.44)
40-49	−93.04 (−532.09, 346.01)	6.13 (5.63, 6.63)	−10.22 (−11.24, −9.20)	−10.20 (−11.08, −9.33)
50-59	675.11 (217.15, 1133.07)	8.29 (7.76, 8.83)	−15.30 (−16.38, 14.21)	−12.34 (−13.26, −11.43)
60-65	2821.36 (2189.13, 3453.58)	8.13 (7.36, 8.89)	−18.74 (−20.43, −17.04)	−9.92 (−11.24, −8.60)
Gender
Male	161.30 (−97.55, 420.15)	−2.72 (−3.04, −2.41)	−1.84 (−2.51, −1.16)	1.57 (1.02, 2.13)
Race/ethnicity
Non-Hispanic White	Ref	Ref	Ref
Black	−2495.07 (−2853.25, −2136.89)	−1.82 (−2.21, 1.42)	−0.60 (−1.50, 3.00)	−4.97 (−5.70, −4.24)
Latinx	−1677.73 (−2048.98, −1306.47)	3.87 (3.40, 4.34)	−2.95 (−3.93, −1.98)	−6.48 (−7.28, −5.67)
Other	−673.92 (−1379.89, 32.06)	−0.53 (−1.39, 0.33)	2.69 (0.90, 4.48)	−0.22 (−0.02, 1.28)
Unknown	−4596.95 (−5309.28, −3884.61)	−2.27 (−3.13, −1.42)	2.80 (1.11, 4.51)	2.28 (0.85, 3.72)
NYC	4178.78 (3860.65, 4496.90)	9.57 (9.14, 9.99)	−1.52 (−2.31, −0.73)	5.85 (5.21, 6.50)
Opioid use disorder	2094.80 (1752.48, 2437.12)	12.74 (12.19, 13.29)	3.08 (2.12, 4.05)	2.78 (2.00, 3.57)
Alcohol use disorder	400.49 (140.90, 660.08)	−2.24 (−2.56, −1.92)	12.78 (11.93, 13.64)	12.65 (12.10, 13.19)
Stimulant use disorder	140.79 (−207.18, 488.77)	3.20 (2.80, 3.61)	1.59 (0.69, 2.49)	12.65 (12.10, 13.19)
Cannabis use disorder	−1958.44 (−2275.54, −1641.34)	−0.23 (−0.60, 0.15)	0.33 (−0.46, 1.13)	−4.34 (−5.02, −3.67)
Severe mental illnesses	3521.56 (3245.17, 3797.95)	0.43 (−0.75, −0.11)	4.16 (3.46, 4.85)	2.54 (1.96, 3.13)
Heart disease	5801.14 (5514.50. 6087.79)	1.67 (1.32, 2.01)	14.12 (13.40, 14.84)	12.50 (11.90, 13.11)
Diabetes	4558.88 (4227.78, 4889.97)	2.95 (2.55, 3.35)	1.47 (0.52, 2.42)	3.03 (2.27, 3.79)
Asthma	3152.99 (2808.37, 3497.60)	3.65 (3.17, 4.13)	6.16 (5.04, 7.27)	0.47 (−0.45, 1.38)
HIV/AIDS	13712.28 (13314.23, 14110.33)	0.38 (−0.08, 82.80)	1.07 (0.19, 1.96)	6.36 (5.59, 7.12)
COPD	5009.50 (4619.59, 5399.41)	1.62 (1.12, 2.12)	1.93 (0.79, 3.07)	6.36 (5.59, 7.12)
Hypertension	3190.36 (2895.69, 3485.02)	2.89 (2.52, 3.27)	1.64 (0.86, 2.42)	2.68 (2.01, 3.34)

Abbreviations: ED, emergency department; DID, difference-in-differences; CI, Confidence Interval

^aAll models were adjusted for age, sex, race/ethnicity, chronic conditions (HIV/AIDS, severe mental illness, diabetes, hepatitis, hypertension, heart disease, asthma), prior care management enrollment status, and year of enrollment.

^bGamma model results, presenting AME

^cLogistic regression model results, presenting AME in percentage point units

4. Discussion

The enactment of the ACA and funding for Medicaid HH provided comprehensive care management for individuals with complex healthcare needs. This study examined the impact of HH on healthcare services for individuals with SUD. Individuals enrolled in the program had a high prevalence of comorbid chronic medical conditions and high levels of acute care service utilization and spending at baseline. The HH program reduced the use of acute care services associated with SUD, with a modest drop in hospitalizations for detoxification. We also observed a moderate decline in acute care services and spending for all-cause healthcare services utilization.

Our results indicated that HH enrollees moderately increased outpatient medical visits and lowered use of acute care services compared to non-enrollees. According to the study conceptual model, increased receipt of outpatient care may prevent avoidable hospitalizations or ED visits. Contrary to our conceptual model, we did not observe a significant increase in outpatient SUD care or pharmacotherapy for OUD that might have explained better engagement in outpatient medical care and reduced acute care visits. The care management services were successful in improving care for chronic medical conditions despite no evidence of increase in treatment for SUD. While the data do not directly speak to why, plausible explanations could be that the care managers may have been addressing the SUD needs of these individuals or that better engagement with primary care included some level of intervention for the SUD. The moderate decline in the probabilities of ED use, detoxification, and hospitalization related to SUD speaks to some benefits of care management for individuals with SUD. The results are not adequate to conclude that care management services suffice for addressing the SUD care needs of these individuals. We observed small effects on SUD specific outcomes. Further research should identify whether there are mechanisms within a generalist care management model to promote better engagement in SUD treatment.

Promoting engagement in primary care may be easier than with SUD treatment, particularly within a generalist care management model. The lack of an effect on SUD outpatient treatment may be due to a combination of multiple personal and structural barriers to accessing SUD treatment and the increased accessibility of primary and other healthcare through HH. Commonly cited personal barriers include the stigma of substance use and associated help-seeking, disbelief in the effectiveness of treatment, and readiness to seek help.^31-33 There are additional structural barriers that hinder the accessibility of SUD treatment programs compared to primary care.³⁴ In New York State, 87% of opioid treatment programs are operating at more than 80% capacity.³⁵ The few available SUD treatment programs may also be out of reach for patients due to distance, a lack of transportation, or no respite from other responsibilities such as employment or caregiving.^31-33 Additionally, prior research has shown that the average wait-time for SUD treatment varies between 24 days to 54 days.³⁶ Individuals who have a comorbid chronic condition may be more likely to receive SUD treatment directly through their primary care providers if they have frequent contact with the primary care system. Primary care providers can directly provide federally approved medications for alcohol and opioid use disorders³⁷ and may not necessarily find it appropriate to refer to a specialty setting. Finally, our study examines short-term outcomes after 1 year of enrollment, which may not be enough to observe the effects on outpatient SUD care utilization.

HH draws from the conceptual underpinnings of the CCM^19,20 and are mechanisms for assessing and addressing the needs of individuals with complex healthcare needs that require coordination across diverse sets of providers. The CCM emphasizes the importance of addressing the multi-faceted needs of individuals, including their behavioral health.³⁸ This aspect of HH may be driving the benefits seen for enrollees with SUD. As codified in the ACA and, particularly as implemented in New York State, the Medicaid HH program is a policy approach that delivers many of the elements of the CCM to a large number of individuals with complex chronic medical conditions.

The effects of the New York State HH program are notable because the state supported a ‘generalist’ model that funded care management agencies to provide services to a broad array of chronic health conditions rather than targeting one or fewer specific conditions. Some states have developed more targeted models that provide services for a more narrowly defined group of individuals. For example, other state HH programs have increased the use of primary care and reduced the use of acute services for individuals with severe mental illnesses. Studies of these programs have found improved health and healthcare services outcomes.^39,40 The effects observed in this study indicate that ‘generalist’ models can also benefit individuals with SUD.

While promising, we should understand the findings of this study in light of its limitations. We could not conduct a more robust study method that would have included randomization to address unobserved confounding. We employed rigorous analytical methods to derive a comparison group that would represent the counterfactual for the HH enrollees; however, these methods cannot fully protect against unmeasured confounding. In addition, these analyses rely on Medicaid billing data to assess individual characteristics and outcomes. We may have not captured all of the factors associated with the experiences and outcomes of individuals. Despite these limitations, the study employs a rigorous quasi-experimental method that can inform the decisions of policymakers on the benefits of similar programs.

5. Conclusion

This study found that the New York State HH program reduced acute care service utilization among individuals with SUD. These findings are notable given that the program employed a generalist model that made care management available for a diverse set of chronic medical conditions.

Highlights.

• Health Home (HH) enrollees with SUD had high levels of acute care services and costs

• HH program reduced acute care services associated with SUD

• HH program reduced acute care services and costs for all-cause healthcare services