Impact of the 1115 behavioral health Medicaid waiver on adult Medicaid beneficiaries in New York State

Abstract

Objective

To evaluate the impact of the Health and Recovery Plan (HARP), a capitated special needs Medicaid managed care product that fully integrates physical and behavioral health delivery systems in New York State.

Data Sources

2013‐2019 claims and encounters data on continuously enrolled individuals from the New York State Medicaid data system.

Study Design

We used a difference‐in‐difference approach with inverse probability of exposure weights to compare service use outcomes in individuals enrolled in the HARP versus HARP eligible comparison group in two regions, New York City (NYC) pre‐ (2013‐2015) versus post‐ (2016‐2018) intervention periods, and rest of the state (ROS) pre‐ (2014‐2016) versus post‐ (2017‐2019) intervention periods.

Data Collection/Extraction Methods

Not applicable.

Principal Findings

HARPs were associated with a relative decrease in all‐cause (RR = 0.78, 95% CI 0.68‐0.90), behavioral health‐related (RR = 0.76, 95% CI 0.60‐0.96), and nonbehavioral‐related (RR = 0.87, 95% CI 0.78‐0.97) stays in the NYC region. In the ROS region, HARPs were associated with a relative decrease in all‐cause (RR = 0.87, 95% CI 0.80‐0.94) and behavioral health‐related (RR = 0.80, 95% CI 0.70‐0.91) stays. Regarding outpatient visits, the HARPs benefit package were associated with a relative increase in behavioral health (RR = 1.21, 95% CI 1.13‐1.28) and nonbehavioral health (RR = 1.08, 95% CI 1.01‐1.15) clinic visits in the NYC region. In the ROS region, the HARPs were associated with relative increases in behavioral health (RR = 1.47, 95% CI 1.32‐1.64) and nonbehavioral health (RR = 1.17, 95% CI 1.11‐1.25) clinic visits.

Conclusions

Compared to patients with similar clinical needs, HARPs were associated with a relative increase in services used and led to a better engagement in the HARPs group regardless of the overall decline in services used pre‐ to postperiod.

What is already known in this

• Integrating physical and behavioral health within a special needs plan, Medicaid managed care, has been shown to increase access and utilization of health services for adults with serious behavioral health conditions.

• Expanding Medicaid coverage in this population showed improvement in the management of physical conditions such as diabetes, blood pressure, and cholesterol.

• Although there is some evidence of effectiveness of the integrated Medicaid managed care in reducing avoidable hospitalizations, and costs, results are mixed due to the different lengths of enrollment or coverage lapses or both.

What this Study Adds

• We used administrative data to evaluate the impact of a statewide fully integrated special needs Medicaid managed care program for individuals with serious mental illness, a natural experiment that occurred with the New York State's Health and Recovery Plan (HARP) implementation, Medicaid coverage expansion.

• Medicaid managed care recipients who maintained at least 2 years of continuous enrollment in the special needs plan showed significant reductions in hospitalizations and increases in access and utilization of outpatient health services.

1. INTRODUCTION

New York State (NYS) has one of the largest Medicaid programs in the country with roughly 24% of the population enrolled and an estimated spending of almost $78 billion a year (second highest in the nation). ¹ In 2015, the 1115 Medicaid waiver demonstration program was amended to enable qualified Managed Care Organizations (MCOs) throughout the State to comprehensively meet the needs of adults with behavioral health (BH) conditions. ¹ In addition, the Delivery System Reform Incentive Payment (DSRIP) program was another Medicaid waiver program launched in 2014 which allocated funding to 25 provider networks across New York state. This funding enabled health care providers to implement reforms that reduce avoidable hospitalizations, improve care, and reduce costs. ² , ³ Like the NYS OMH Collaborative Care Medicaid Program (CCMP) implemented from 2012 to 2014 and supported through payment reforms in 2015, DSRIP included integration of the physical health and behavioral health delivery systems, following Affordable Care Act (ACA) policies to provide integrated managed care for all Medicaid members.

The focus on clinical integration and payment reforms in ACA drove more states to move toward comprehensive “carve‐in,” or integrated Medicaid managed Care (MMC), to finance and administer different types of services within a single managed care plan. ⁴ A few early studies reported negligible differences between MMC and fee‐for‐service (FFS) in access and utilization of health services for people with behavioral health conditions, ⁵ , ⁶ while an evaluation of Nebraska's MMC showed a significant decline in inpatient mental health treatment. ⁷

Based on previous research, adults with behavioral health conditions account for almost 50% of Medicaid beneficiaries and are one of the most medically needy and high‐cost populations. ⁸ Adults with serious mental illness (SMI) die, on average, 25 years earlier than the general population. These disproportionately high rates of mortality have been linked to cardiovascular and pulmonary disease, based on modifiable risk factors including high rates of smoking and obesity. ⁹ Co‐occurring substance use disorders are also prevalent among SMI individuals. ¹⁰

State Medicaid programs have been expanding the use of MCOs to serve these high need populations. Based on federal incentives, many states began using the Medicaid “health home” option, established by the ACA to advance the integration of physical and behavioral health care for Medicaid beneficiaries with SMI.

Unfortunately, there has been a lack of research on the impact of integrated MMC for people with behavioral health conditions. Missouri was the first state to adopt “health homes” specifically for individuals with SMI. Nardone and Snyder ¹¹ reported Missouri's results enrolling 19,000 Medicaid beneficiaries in health homes. They reported that the percentage of enrollees who had one hospitalization or more declined by 27% between 2011 and 2012. However, a more recent study by Xiang et al ¹² assessed the impact of a new Medicaid managed care model (ICP) in Illinois, investigating health services utilization and costs for adults with behavioral health conditions. Results were mixed but showed some evidence of effectiveness of the integrated MMC (“carve‐in”) for Medicaid beneficiaries with behavioral health conditions. However, the addition of the “SMART Act” (which restricted access to care for certain clients) was detrimental. In fact, the authors argued “absent the SMART Act, it is likely that ICP would have significantly reduced Medicaid spending per person per month”. ¹²

To address the unique needs of adults with serious behavioral health conditions, NYS developed a specialty MMC product called a Health and Recovery Plan (HARP). HARPs offer access to an enhanced benefit package comprised of Home and Community‐Based Services (HCBS) designed to provide the individual with a specialized scope of support services not currently covered under the State Plan. Individuals enrolled in HARPs were offered Health Home care management services. Health Homes develop person‐centered plans of care that integrate physical and behavioral health services including HCBS for eligible individuals. The HCBS are recovery oriented and designed to assist individuals with independent daily living and social skills, education and employment support services, peer and family supports, habilitation services, nonmedical transportation, and crises management. Beyond those efforts at coverage expansions, HARPs also have specialized staffing requirements and qualifications along with focused behavioral health performance metrics and incentives to achieve health, wellness, recovery, and community inclusion for enrollees. ¹³

The aim of this study was to examine the impact of the HARPs on Medicaid services utilization. We hypothesized that compared to eligible individuals who did not enroll in the HARP MMC plan, enrollment in the HARP MMC plan would lead to reductions in inpatient and emergency service utilization for behavioral health and for other health conditions and increases in the use of community‐based services for behavioral health and for other health conditions for individuals with significant behavioral health needs who enrolled. This study used a weighted pre‐ and postdifference‐in‐difference longitudinal cohort design (DID) to examine the impact of the HARP on Medicaid service utilization.

2. METHODS

2.1. Study design

Changes in service use between HARP eligible and HARP enrolled individuals were examined with a propensity score adjusted difference‐in‐difference (DID) longitudinal cohort design. This design takes advantage of the natural experiment that occurred with the New York State's HARP implementation. The HARPs were first implemented in New York City residents (NYC) (October 2015) followed by the rest of state (ROS) (July 2016). The exposure group (HARP) was defined as HARP enrolled at some point during the year “rollout phase” of the implementation for each region. The comparison group was defined as HARP eligible but never enrolled during the “rollout phase.” The pre‐HARP period for each phase was defined as two years prior to the implementation start date, from October 2013 to October 2015 for NYC and July 2014 to July 2016 for ROS. It took time to fully implement the HARPs as such a “rollout phase” used as a ramp‐up period was excluded from pre‐post comparison periods. As such, the post‐HARP period was defined as October 2016 to October 2018 for NYC and July 2017 through July 2019 for ROS (Figure 1). Individuals in the HARP exposure group and HARP eligible control group were required to be continuously eligible for Medicaid with no more than a 45‐day gap for the entire study period (pre‐, rollout, and postperiod). Individuals in the HARP exposure group and HARP eligible control group were excluded from the study if they “switched” enrollment, that is, disenrolled from or enrolled in the HARP in the two‐year postperiod. The final study population consisted of Medicaid enrolled adults ages 21‐64 who were enrolled in the NYC phase, HARP (Exposure, N = 35 579) or HARP Eligible (Comparison, N = 10 997) and in the ROS phase, HARP (Exposure, N = 28 200) or HARP Eligible (Comparison, N = 14 241). See Figures S1 and S2 for sample selection process.

FIGURE 1.

Study design. Lines were used to explain the study design and do not represent real data. NYC: New York City, ROS: Rest of the New York State

New York State identified people that were eligible for HARP based on several factors, including past Medicaid use. ¹⁴ The targeting criteria included SMI/SUD diagnosis, receiving Supplemental Security Income; having a current or expired Assisted Outpatient Treatment order; OMH residential program; OMH discharges from correctional facilities; and past use of OMH specialty, inpatient, and emergency room services. HARP eligible enrollees of a Medicaid Managed Care Organization that runs a HARP did not have to do anything to join. These people have received a notice from New York Medicaid Choice. They then had 30 days to opt out or choose to enroll in another HARP. HARP eligible enrollees of a Medicaid Managed Care Organization without a HARP needed to take action to join a HARP. These people received notice from NYS telling them how to join a HARP. Once enrolled in a HARP, members have 90 days to choose another HARP or return to their previous plan. After 90 days, members are locked into their HARP of choice for 9 additional months. Individuals enrolled in an HIV SNP were able to receive HCBS services through the HIV SNP. They were notified of their HARP eligibility by the NYS Enrollment Broker. HARP enrollment was voluntary, and eligible individuals could also choose to remain in a mainstream plan. ¹⁵ , ¹⁶ Although eligible individuals failed to benefit from additional services available through HARP, they had access to existing integrated care through MMC plans and existing OMH recovery‐oriented community‐based services. Approximately 40%‐50% of HARP eligible remained in FFS Medicaid (in the ROS and NYC control samples, respectively).

2.2. Data source

The OMH view of the NYS Medicaid data system was used to identify the study population and to quantify health services used. This database provides information on all individuals with a current or history of behavioral health services in Medicaid and includes demographics, enrollment, and health services claimed during the study period.

2.3. Propensity score weighting criteria

Participant demographic characteristics, behavioral diagnoses, health conditions/clinical risks status, prescribed medications, and number of services (inpatient stays, ER, and outpatient visits) used during the pre‐HARP period were used to adjust for the difference between HARP enrolled and HARP eligible control group. Demographic characteristics were sex, age, race/ethnicity (White non‐Hispanic, Black non‐Hispanic, Asian, American Indian/Alaskan native, Hispanic, and Other‐multiracial), aid category (SSI, safety net, health premium, TANF), and Health Home assignment (yes, no). Principal behavioral health diagnoses included schizophrenia, mood disorders, anxiety, physical, organic, and substance use disorder. Clinical Risk Group (CRG™) status ¹⁷ was used to assign patients to 1 of 9 mutually exclusive CRGs according to the presence, type, and severity of chronic conditions. Behavioral health medications were assessed as having filled a prescription or not in the following categories: antidepressant, anxiolytic, antipsychotic typical, antipsychotic atypical Clozapine, antipsychotic atypical non‐Clozapine, mood stabilizer, stimulant, smoking cessation, and anticraving.

2.4. Utilization outcomes

Medicaid claims and encounters were categorized into three service types: inpatient, emergency, and outpatient. Services were defined using a combination of rate codes, DRG codes, specialty codes, procedure codes, revenue codes, primary diagnosis of mental or substance use disorders, and hospital provider type (see Table S1).

Inpatient was categorized as all‐cause, behavioral, and nonbehavioral health‐related stays at licensed general or private hospitals. Behavioral health inpatient stays included treatment for mental (psychiatric care) and substance use (detox, rehabilitation, and diagnosis‐based) disorders. Nonbehavioral inpatient comprised stays that did not meet the conditions for mental and substance use disorder treatment. Three measures were created to evaluate inpatient use, a binary indicator for ever use or not, total number of inpatient stays, and total length of stay (days).

ER was also categorized as all‐cause, behavioral, and nonbehavioral health‐related visits to general hospitals and comprehensive psychiatric emergency programs. Behavioral health ER comprised of mental health and substance use disorders‐related visits. Nonbehavioral ER use comprised visits that were related neither to mental health nor substance use disorders. Two measures, a binary indicator for at least one visit (yes, no) and total number of ER visits, were used to evaluate ER services.

Outpatient was defined as behavioral, nonbehavioral clinic visits, and OMH specialty services. Behavioral health clinic comprised psychiatry‐related or substance use disorder‐related daily visits at an OMH licensed clinic or Office of Addiction Services and Supports (OASAS) outpatient setting/clinic. Nonbehavioral clinic use comprised visits that were related neither to mental health nor substance use. OMH specialty services comprised ACT and PROS units per patient per month in community settings. Two measures, a binary indicator for at least one outpatient visit (yes, no) and total number of visits, were used to evaluate outpatient services.

2.5. Statistical analysis

Separate analyses were conducted in parallel for the two HARP rollout phases (NYC and ROS). Each phase had a specific rollout and follow‐up period (Figure 1). The analyses for each phase used a difference‐in‐difference (DID) approach ¹⁸ with propensity score weighting (PS).

Generalized boosted models (GBM) in the package “twang” (Toolkit for Weighting and Analysis of Nonequivalent Groups) were used for the estimation and evaluation of propensity scores and the associated weights. ¹⁹ We modeled the propensity score to predict group assignment (HARP enrolled vs eligible controls) using information on demographic characteristics, CRG™ status, principal diagnosis, Health Home assignment, aid category, prescribed medications, and number of services used during pre‐HARP period. GBM is a machine learning technique that builds the propensity score model iteratively by starting from a globally constant model, then adding a regression tree at each step, subsequently creating an increasingly complex piecewise constant function. It fits both linear and nonlinear models using regression trees and then combines/merges the predictions computed by each model. ²⁰ , ²¹ The absolute standardized mean difference (SMD) of the effect size and the Kolmogorov‐Smirnov statistic were used to assess the balance in covariate distribution. ¹⁹ The PS method achieved excellent balance in baseline covariates between HARP enrolled and eligible control group. In addition, the PS method via the GBM balances the groups in terms of missingness of covariates. This is a substantial advantage over the standard logistic regression traditionally used in this field. However, a recent study suggests that imputing missing covariates prior to fitting a PS model results in less bias than GBM, but this may be affected by several factors including trimming of weights and magnitude of missingness. ²² Inverse probability of exposure (HARP insurance coverage) weighting using the propensity score method was used to mitigate the effect of confounding. Compared to PS matching and stratification, PS weighting was attractive because it is robust to the misspecification of the regression model and retains the entire sample for optimal power. ²³

To examine the differences in service use before and after each phase, adjusted odds and rate ratios were calculated by using a generalized estimating equation (GEE) model (accounting for multiple observations per individual) with a binomial or negative binomial distribution and an independent working correlation structure. ²⁴ , ²⁵ For binary outcomes (having used or not having used a service), a GEE logistic regression model was used. For count outcomes (number of inpatient stays, length of stays, ER, or clinic visits), a GEE negative binomial regression model was used. The GEE models were specified for the service use outcomes using inverse probability of exposure weights, indicator for exposure (HARP or not), binary indicator for time period, and the interaction between time period and indicator for exposure. A statistically significant interaction between time period and indicator for exposure (DID estimate of HARPs effect postmanage care implementation) is interpreted as the impact of the HARP plan. Odds ratios (ORs) and rate ratios (RRs) with 95% confidence intervals are presented.

To test the robustness of the study findings, outcome models using estimated propensity scores without pre‐utilization variables were conducted. In addition, alternative model specifications for the outcome variables were examined using a standard normal regression. To rigorously test the dependence of the findings on time (duration in HARP), this research also conducted weighted analyses limited to individuals who were continuously enrolled in HARP for only a year. A review by Nathan S. Kline Institute's Institutional Review Board deemed this study to not involve human subjects’ research.

3. RESULTS

Table 1 reports the unweighted and propensity weighted sample characteristics of HARP enrolled (intervention) and HARP eligible (control) group by intervention phase (NYS and ROS) during the pre‐HARP intervention period. PS weighting resulted in an excellent balance of all confounders (SMD < 0.1) between intervention and control groups.

TABLE 1.

Baseline characteristics of HARP enrolled and HARP eligible recipients in New York City and the Rest of the NY State

New York City	Before weighting				After weighting
Baseline covariates	HARP		Non‐HARPs		HARPs	Non‐HARPs
	(N = 35 579)		(N = 10 997)		(N = 35 579)	(N = 10 997)
	N	%	N	%	%	%	SMD
Sex (male)	17 256	48.5	6510	59.2	50.5	50.6	0.002
Race/ethnicity
Hispanic	7080	19.9	1089	9.9	17.8	17.4	0.011
White	9215	25.9	3750	34.1	27.5	27.6	0.002
Black	15 477	43.5	4674	42.5	43.4	44	0.011
American Indian/Alaskan native	569	1.6	132	1.2	1.5	1.4	0.009
Asian	1032	2.9	495	4.5	3.4	3.3	0.005
Other	2241	6.3	847	7.7	6.4	6.4	0.001
Health Home assignment	8681	24.4	1924	17.5	22.9	22.5	0.009
Clinical risk groups (health condition)
Healthy	1103	3.1	583	5.3	3.7	3.9	0.008
History of significant acute disease	391	1.1	154	1.4	1.2	1.1	0.008
Single minor chronic disease	463	1.3	132	1.2	1.3	1.2	0.011
Minor chronic disease in multiple organ systems	249	0.7	66	0.6	0.7	0.5	0.021
Single dominant or moderate chronic disease	5337	15	2100	19.1	15.8	15.5	0.009
Disease in chronic multiple organ systems	18 715	52.6	6180	56.2	53.3	54.1	0.016
Dominant chronic disease in three or more organ systems	4234	11.9	1254	11.4	11.8	11.9	0.002
Dominant and metastatic malignancies	249	0.7	77	0.7	0.7	0.7	0.008
Catastrophic conditions	4803	13.5	451	4.1	11.5	11.2	0.009
Aid category
SSI	30 278	85.1	5026	45.7	76.7	75.9	0.019
Safety net	4412	12.4	1738	15.8	13.5	13	0.015
Health premium	605	1.7	3783	34.4	8.4	9.6	0.042
TANF	285	0.8	440	4	1.5	1.5	0.003
Diagnosis
Physical	18 572	52.2	6257	56.9	53.3	54	0.014
Substance use disorder	4910	13.8	2034	18.5	14.5	14.2	0.009
Schizophrenia	4661	13.1	1034	9.4	12.4	12.3	0.006
Mood disorder	5764	16.2	1221	11.1	15.3	15	0.007
Organic mental disorder	71	0.2	44	0.4	0.2	0.2	0
Anxiety	1352	3.8	330	3	3.5	3.7	0.007
Medication adherence
Antidepressants	19 746	55.5	4982	45.3	53.5	53.5	0.001
Anxiolytics	13 876	39	3431	31.2	37.4	37.1	0.006
Antipsychotics typical	3451	9.7	891	8.1	9.5	9.6	0.005
Mood stabilizers	11 670	32.8	3145	28.6	31.8	32.4	0.013
Stimulants	1281	3.6	451	4.1	3.7	3.5	0.01
Smoking cessation	3558	10	880	8	9.5	9.3	0.007
Anticraving	1032	2.9	616	5.6	3.5	3.6	0.009
Antipsychotic atypical Clozapine	640	1.8	121	1.1	1.6	1.5	0.012
Antipsychotic atypical non‐Clozapine	15 370	43.2	3629	33	41.3	41.8	0.01
	N	Mean (SD)	N	Mean (SD)	Mean	Mean	SMD
Age	35 579	47.42 (11.40)	10 997	45.48 (13.56)	47.09	46.97	0.01
Emergency room visit
All cause	30 568	4.45 (10.22)	9630	6.79 (18.77)	4.76	4.87	0.01
Behavioral health related	14 238	1.17 (4.81)	5708	2.51 (10.17)	1.39	1.46	0.012
Nonbehavioral health related	29 127	3.28 (7.82)	9124	4.28 (13.19)	3.37	3.41	0.005
Inpatient stay
All cause	20 279	1.51 (3.70)	7609	2.94 (6.24)	1.73	1.79	0.016
Behavioral health related	11 783	0.87 (2.75)	5565	2.00 (4.89)	1.05	1.11	0.019
Nonbehavioral health related	14 205	0.64 (2.14)	4852	0.93 (3.45)	0.68	0.69	0.003
Outpatient visit
Behavioral health clinic	33 887	25.67 (28.99)	9783	18.60 (27.15)	24.18	24.06	0.004
Nonbehavioral health clinic	34 956	29.51 (31.53)	10 623	23.11 (29.78)	28.09	28.08	0
OASAS Opioid‐related	4815	40.03 (125.10)	1816	33.27 (106.27)	38.05	36.56	0.012
PROS	1822	6.50 (37.99)	430	4.01 (28.49)	5.98	5.69	0.008
ACT	1232	1.72 (11.41)	333	0.86 (7.58)	1.53	1.42	0.01

Rest of the State	Before weighting				After weighting
Baseline covariates	HARPs		Non‐HARPs		HARPs	Non‐HARPs
	(N = 28 200)		(N = 14 241)		(N = 28 200)	(N = 14 241)
	N	%	N	%	%	%	SMD
Sex (male)	15 512	43.6	6136	55.8	46.7	47.2	0.011
Race/ethnicity
Hispanic	4412	12.4	594	5.4	10.4	10.3	0.005
White	21 347	60	8237	74.9	64.6	64.6	0.001
Black	8824	24.8	1815	16.5	22.4	22.2	0.005
American Indian/Alaskan native	427	1.2	110	1	1.1	1.2	0.005
Asian	569	1.6	110	1	1.4	1.4	0.005
Other	0	0	121	1.1	0.1	0.4	0.064
Health Home assignment	10 247	28.8	1826	16.6	25.6	25.3	0.007
Clinical risk groups (health condition)
Healthy	1672	4.7	935	8.5	6.2	6	0.006
History of significant acute disease	889	2.5	341	3.1	2.7	2.6	0.009
Single Minor chronic disease	854	2.4	242	2.2	2.4	2.3	0.004
Minor chronic disease in multiple organ systems	498	1.4	154	1.4	1.4	1.4	0.004
Single Dominant or Moderate Chronic disease	6831	19.2	3035	27.6	21.1	21.8	0.017
Disease in chronic multiple organ systems	19 711	55.4	5334	48.5	53.3	53.1	0.006
Dominant chronic disease in three or more organ systems	3736	10.5	671	6.1	9.4	9.3	0.003
Dominant and metastatic malignancies	285	0.8	77	0.7	0.8	0.8	0
Catastrophic conditions	1103	3.1	209	1.9	2.8	2.9	0.005
Aid category
SSI	27 609	77.6	3310	30.1	62.9	61.8	0.022
Safety net	5657	15.9	1847	16.8	16.5	16	0.012
Health premium	1459	4.1	4751	43.2	16.1	17.3	0.031
TANF	854	2.4	1089	9.9	4.6	4.9	0.015
Diagnosis
Physical	21 312	59.9	6565	59.7	59.6	60	0.007
Substance use disorder	3344	9.4	2155	19.6	12.3	12.7	0.012
Schizophrenia	3273	9.2	539	4.9	8.2	8.1	0.005
Mood disorder	5088	14.3	1078	9.8	13	12.7	0.009
Organic mental disorder	36	0.1	11	0.1	0.2	0.2	0.003
Anxiety	2064	5.8	495	4.5	5.4	5.3	0.006
Medication adherence
Antidepressants	22 130	62.2	5685	51.7	59.3	58.7	0.012
Anxiolytics	14 374	40.4	3717	33.8	38.9	38.8	0.002
Antipsychotics typical	2704	7.6	506	4.6	6.9	7	0.005
Mood stabilizers	13 947	39.2	3497	31.8	37.5	36.9	0.013
Stimulants	3238	9.1	1320	12	10.1	10.4	0.01
Smoking cessation	4839	13.6	1221	11.1	12.9	12.5	0.012
Anticraving	2455	6.9	1628	14.8	9.1	9.2	0.005
Antipsychotic atypical Clozapine	640	1.8	99	0.9	1.5	1.5	0
Antipsychotic atypical non‐Clozapine	13 057	36.7	2826	25.7	34	33.6	0.009
	N	Mean (SD)	N	Mean (SD)	Mean	Mean	SMD
Age	28 200	43.75 (11.77)	14 241	39.07 (13.04)	42.61	42.29	0.026
Emergency room visit
All cause	25 320	5.54 (10.33)	12 935	5.41 (9.74)	5.42	5.25	0.018
Behavioral health related	12 022	1.15 (4.02)	7820	1.60 (4.52)	1.24	1.27	0.006
Nonbehavioral health related	24 478	4.39 (8.39)	12 298	3.81 (7.50)	4.18	3.98	0.026
Inpatient stay
All cause	15 287	1.14 (2.45)	9698	1.59 (2.56)	1.26	1.26	0.001
Behavioral health related	8791	0.60 (1.62)	7243	1.06 (1.98)	0.72	0.74	0.008
Nonbehavioral health related	10 611	0.53 (1.65)	5553	0.53 (1.50)	0.54	0.53	0.008
Outpatient visit
Behavioral health clinic	26 575	18.72 (22.16)	12 611	12.34 (18.39)	16.99	16.65	0.016
Nonbehavioral health clinic	27 672	26.98 (28.33)	13 624	17.36 (23.35)	24.29	23.46	0.031
OASAS Opioid related	1177	7.65 (56.56)	1206	8.97 (57.04)	8.49	7.83	0.011
PROS	3277	11.66 (46.59)	1195	6.76 (33.85)	10.54	9.95	0.014
ACT	732	1.07 (8.82)	175	0.32 (4.40)	0.9	0.65	0.034

Calculated proportional and mean differences using chi‐square or t test. In the NYC phase, all variables were significantly different at alpha = 0.05, except race‐Black, History of significant acute disease, Single Minor chronic disease, Minor chronic disease in multiple organ systems, Dominant chronic disease in three or more organ systems, and Dominant and metastatic malignancies. In the ROS phase, all variables were significantly different at alpha = 0.05, except race‐Asian, Single Minor chronic disease, Minor chronic disease in multiple organ systems, Dominant and metastatic malignancies, physical and organic diagnoses, number of behavioral health inpatient stays, and number of all‐cause ER visits.

Abbreviations: NY, New York; SD, Standard deviation; SMD, Standardized mean difference.

Table 2 presents the unadjusted pre‐ and postperiod estimates for each service utilization measure (ever use, average rate of use, and inpatient length of stay). Service use for HARP and non‐HARP participants generally decreased in the pre‐ to postperiod comparisons. Notable patterns (from baseline rate) of acute service (ever use) measures included a 7% (26.4) and 14.9% (29.9) decrease for behavioral health inpatient service use for HARP pre‐post and non‐HARP pre‐post, respectively, in the NYC phase. Similarly, a decrease of 8.1% (24.5) and 17.6% (42.8) in behavioral health inpatient service use was observed for HARP pre‐post and non‐HARP pre‐post in the ROS phase. In terms of average all‐cause inpatient length of stay, a decrease of 1.5 (12.5) and 5.5 (22.8) days were observed for HARP pre‐post and non‐HARP pre‐post in the NYC phase. Notable patterns for outpatient service (at least one visit) included a 5.9% (89.0) and 8.2% (78.3) decrease for behavioral clinic visit for HARP pre‐post and non‐HARP pre‐post in the NYC phase. Regarding average nonbehavioral clinic visits, notable patterns included an observed decrease of 2.7 (29.5) and 4.5 (23.1) visits for HARP pre‐post and non‐HARP pre‐post in the NYC phase.

TABLE 2.

Service use measures before and after Medicaid managed care implementation among HARP enrolled and eligible control group in New York City and the Rest of the State

Utilization measure	New York City						Rest of the State
	HARPs (N = 35 579)			Non‐HARPs (N = 10 997)			HARPs (N = 28 200)			Non‐HARPs (N = 14 241)
	Pre (%)	Post (%)	Post‐ Pre	Pre (%)	Post (%)	Post‐Pre	Pre (%)	Post (%)	Post‐ Pre	Pre (%)	Post (%)	Post‐Pre
Ever use of service
Emergency room visit
All cause	72.84	71.45	−1.39	76.59	71.05	−5.55	79.46	75.15	−4.30	81.28	72.88	−8.40
Behavioral health related	29.33	25.41	−3.93	41.54	33.70	−7.84	32.52	24.90	−7.62	44.84	30.31	−14.53
Nonbehavioral health related	67.12	66.94	−0.18	69.40	65.72	−3.68	74.98	71.30	−3.67	73.63	67.81	−5.81
Inpatient stay
All cause	44.12	37.99	−6.12	60.13	46.40	−13.73	41.42	34.05	−7.37	56.93	40.69	−16.24
Behavioral health related	26.42	19.41	−7.00	44.80	29.87	−14.93	24.45	16.32	−8.13	42.69	25.15	−17.55
Nonbehavioral health related	27.02	25.89	−1.13	31.69	28.10	−3.59	24.96	23.55	−1.40	25.95	23.90	−2.05
Outpatient visit
Behavioral health clinic	88.99	83.11	−5.89	78.32	70.17	−8.15	88.21	79.73	−8.48	76.75	70.00	−6.75
Nonbehavioral health clinic	95.29	93.61	−1.68	90.70	86.01	−4.68	95.02	92.89	−2.12	88.69	85.46	−3.24
OASAS Opioid related	11.95	12.19	0.24	13.33	13.31	−0.02	2.82	3.57	0.76	4.24	7.08	2.84
PROS	4.13	3.03	−1.10	3.08	2.06	−1.03	9.34	5.66	−3.69	6.59	4.09	−2.49
ACT	2.56	2.85	0.29	1.74	2.36	0.63	1.83	2.04	0.20	0.74	0.84	0.09
	Pre (mean)	Post (mean)	Post‐Pre	Pre (mean)	Post (mean)	Post‐Pre	Pre (mean)	Post (mean)	Post‐Pre	Pre (mean)	Post (mean)	Post‐Pre
Rate of use
Emergency room visit
All cause	4.45	4.87	0.42	6.79	6.97	0.18	5.54	4.76	−0.78	5.41	4.47	−0.94
Behavioral health related	1.17	1.24	0.07	2.51	2.67	0.16	1.15	0.93	−0.23	1.60	1.22	−0.38
Nonbehavioral health related	3.28	3.63	0.35	4.28	4.30	0.02	4.39	3.83	−0.56	3.81	3.24	−0.57
Inpatient stay
All cause	1.51	1.38	−0.13	2.94	2.37	−0.57	1.14	0.97	−0.17	1.59	1.21	−0.38
Behavioral health related	0.87	0.68	−0.19	2.00	1.50	−0.51	0.60	0.42	−0.19	1.06	0.71	−0.35
Nonbehavioral health related	0.64	0.70	0.06	0.93	0.87	−0.06	0.53	0.55	0.02	0.53	0.51	−0.03
Inpatient length of stay
All cause	12.50	10.96	−1.54	22.85	17.33	−5.52	8.99	7.22	−1.78	13.33	9.63	−3.71
Behavioral health related	9.45	6.99	−2.46	17.99	12.09	−5.90	6.47	4.16	−2.31	10.43	6.56	−3.87
Nonbehavioral health related	3.05	3.97	0.92	4.86	5.24	0.38	2.52	3.06	0.54	2.91	3.07	0.16
Outpatient visit
Behavioral health clinic	25.67	24.12	−1.55	18.60	16.00	−2.59	18.72	16.62	−2.10	12.34	12.23	−0.11
Nonbehavioral health clinic	29.51	26.78	−2.73	23.11	18.57	−4.54	26.98	21.85	−5.13	17.36	14.64	−2.72
OASAS Opioid related	40.02	18.86	−21.17	33.27	20.29	−12.98	7.65	4.46	−3.18	8.97	7.46	−1.51
PROS	6.50	5.26	−1.25	4.01	2.88	−1.13	11.66	6.66	−5.00	6.76	4.62	−2.14
ACT	1.72	1.53	−0.19	0.86	1.12	0.26	1.07	1.05	−0.02	0.32	0.36	0.05

Table 3 presents unadjusted and propensity score adjusted estimates of the difference‐in‐difference impact of HARP plan on Medicaid service utilization for HARP enrolled group relative to the comparison group. In terms of having an emergency room service, individuals in the HARPs relative to the comparison group were more likely to have an ER visit for any reason (OR = 1.03, 95% CI 1.02‐1.05) and (OR = 1.03, 95% CI 1.02‐1.05) in NYC and ROS phases, respectively. Individuals in the HARPs relative to the comparison group were also more likely to have a nonbehavioral health‐related ER visit (NYC; OR = 1.03, 95% CI 1.02‐1.05, ROS; OR = 1.03, 95% CI 1.02‐1.05) in NYC and ROS phases, respectively. In terms of average rate of all‐cause ER visits, HARPs were associated with a relative decrease in all‐cause ER (RR = 0.88, 95% CI 0.78‐0.98) in the NYC phase. The HARPs were also associated with a relative decrease in ER visits for behavioral health reasons (NYC; RR = 0.68, 95% CI 0.54‐0.86, ROS; RR = 0.86, 95% CI 0.75‐0.99) in both phases.

TABLE 3.

Unadjusted and Propensity Score adjusted difference‐in‐difference estimates of MC impact on service use (pre‐ and post‐HARP implementation)

Utilization measure	New York City		Rest of the State
	Unadjusted	PS Adjusted	Unadjusted	PS Adjusted
	ORs (95% CI)
Ever use of service
Emergency room visit
All cause	1.04 (1.03, 1.05)***	1.03 (1.02, 1.05)***	1.04 (1.03, 1.05)***	1.03 (1.02, 1.05)***
Behavioral health related	1.04 (1.03, 1.05)***	1.01 (1.00, 1.03)	1.07 (1.06, 1.08)***	1.02 (1.00, 1.03)*
Nonbehavioral health related	1.04 (1.02, 1.05)***	1.03 (1.02, 1.05)***	1.02 (1.01, 1.03)***	1.03 (1.02, 1.05)***
Inpatient Stay
All cause	1.08 (1.07, 1.09)***	1.00 (0.98, 1.01)	1.09 (1.08, 1.11)***	1.00 (0.99, 1.02)
Behavioral health related	1.08 (1.07, 1.09)***	1.00 (0.99, 1.01)	1.10 (1.09, 1.11)***	1.00 (0.99, 1.02)
Nonbehavioral health related	1.02 (1.01, 1.04)***	0.99 (0.97, 1.01)	1.01 (1.00, 1.02)	0.99 (0.98, 1.00)
Outpatient visit
Behavioral health clinic	1.02 (1.01, 1.03)***	1.06 (1.04, 1.07)***	0.98 (0.97, 0.99)***	1.03 (1.02, 1.05)***
Nonbehavioral health clinic	1.03 (1.02, 1.04)***	1.05 (1.04, 1.06)***	1.01 (1.00, 1.02)**	1.03 (1.02, 1.04)***
OASAS Opioid related	1.00 (1.00, 1.01)	1.01 (1.00, 1.01)**	0.98 (0.98, 0.98)***	1.00 (0.99, 1.00)
PROS	1.00 (1.00, 1.00)	1.00 (1.00, 1.01)	0.99 (0.98, 0.99)***	1.00 (0.99, 1.01)
ACT	1.00 (0.99, 1.00)*	1.00 (1.00, 1.01)	1.00 (1.00, 1.00)	1.01 (1.00, 1.01)***
	RR (95% CI)
Rate of use
Emergency room visit
All cause	1.07 (1.01, 1.12)*	0.88 (0.78, 0.98)*	1.04 (1.00, 1.08)*	1.01 (0.94, 1.08)
Behavioral health related	1.00 (0.91, 1.09)	0.68 (0.54, 0.86)**	1.05 (0.98, 1.13)	0.86 (0.75, 0.99)*
Nonbehavioral health related	1.10 (1.05, 1.16)***	0.99 (0.91, 1.07)	1.03 (0.99, 1.06)	1.06 (0.99, 1.13)
Inpatient stay
All cause	1.13 (1.08, 1.18)***	0.78 (0.68, 0.90)***	1.12 (1.07, 1.17)***	0.87 (0.80, 0.94)***
Behavioral health related	1.04 (0.99, 1.11)	0.76 (0.60, 0.96)*	1.03 (0.97, 1.09)	0.80 (0.70, 0.91)***
Nonbehavioral health related	1.17 (1.10, 1.24)***	0.87 (0.78, 0.97)*	1.09 (1.02, 1.16)**	0.99 (0.89, 1.09)
Inpatient length of stay
All cause	1.16 (1.10, 1.22)***	0.66 (0.52, 0.84)***	1.11 (1.05, 1.18)***	0.89 (0.75, 1.05)
Behavioral health related	1.10 (1.04, 1.17)**	0.64 (0.46, 0.90)**	1.02 (0.95, 1.10)	0.93 (0.74, 1.17)
Nonbehavioral health related	1.21 (1.11, 1.32)***	0.86 (0.66, 1.12)	1.15 (1.04, 1.26)**	1.01 (0.82, 1.25)
Outpatient visit
Behavioral health clinic	1.09 (1.06, 1.13)***	1.21 (1.13, 1.28)***	0.90 (0.87, 0.93)***	1.47 (1.32, 1.64)***
Nonbehavioral health clinic	1.13 (1.10, 1.16)***	1.08 (1.01, 1.15)*	0.96 (0.94, 0.98)***	1.17 (1.11, 1.25)***
OASAS Opioid related	0.77 (0.72, 0.82)***	1.42 (1.04, 1.93)*	0.70 (0.61, 0.80)***	1.24 (0.89, 1.71)
PROS	1.12 (0.97, 1.30)	1.24 (1.01, 1.51)*	0.84 (0.75, 0.93)***	1.20 (0.90, 1.61)
ACT	0.69 (0.59, 0.80)***	2.11 (1.39, 3.22)***	0.85 (0.68, 1.07)	2.52 (1.61, 3.93)***

*, **, and *** denote significant difference from zero at the .05, .01, and .001 level respectively. Unadjusted models accounted for within‐person correlation but not propensity score (PS) weights. Adjusted models accounted for within‐person correlation and PS weights.

There was no statistically significant difference in the likelihood (ever use) of having an inpatient stay in the HARPs relative to the comparison group in both phases. However, for average rate of inpatient stays, HARPs were associated with a relative decrease in all‐cause (RR = 0.78, 95% CI 0.68‐0.90), behavioral health‐related (RR = 0.76, 95% CI 0.60‐0.96), and nonbehavioral health‐related (RR = 0.87, 95% CI 0.78‐0.97) stays in the NYC phase. In the ROS phase, HARPs were associated with a relative decrease in all‐cause (RR = 0.87, 95% CI 0.80‐0.94) and behavioral health‐related (RR = 0.80, 95% CI 0.70‐0.91) stays. Regarding average length of inpatient stay, HARPs were associated with a relative shorter all‐cause (RR = 0.66, 95% CI 0.52‐0.84) and behavioral health‐related (RR = 0.64, 95% CI 0.46‐0.90) stays in the NYC phase. Similar relative shorter stays were observed in the ROS phase but were not statistically significant at alpha = 0.05.

Individuals in the HARPs relative to the comparison group were more likely to have behavioral health (OR = 1.06, 95% CI 1.04‐1.07) and nonbehavioral health (OR = 1.05, 95% CI 1.04‐1.06) clinic visits, or OASAS opioid‐related services (OR = 1.01, 95% CI 1.00‐1.01) in the NYC phase. In the ROS phase, individuals in the HARPs relative to the comparison group were more likely to have behavioral health (OR = 1.03, 95% CI 1.02‐1.05) and nonbehavioral health (OR = 1.03, 95% CI 1.02‐1.04) clinic visits, and ACT services (OR = 1.01, 95% CI 1.00‐1.01). As to average outpatient visits, the HARPs were associated with relative increases in behavioral health (RR = 1.21, 95% CI 1.13‐1.28) and nonbehavioral health (RR = 1.08, 95% CI 1.01‐1.15) clinic visits, OASAS opioid‐related services (RR = 1.42, 95% CI 1.04‐1.93), and PROS (RR = 1.24, 95% CI 1.01‐1.51) and ACT (RR = 2.11, 95% CI 1.39‐3.22) services in the NYC phase. In the ROS phase, the HARPs were associated with relative increases in behavioral health (RR = 1.47, 95% CI 1.32‐1.64) and nonbehavioral health (RR = 1.17, 95% CI 1.11‐1.25) clinic visits, and ACT services (RR = 2.52, 95% CI 1.61‐3.93).

Results from the analyses using estimated propensity scores without pre‐utilization variables and the alternative specifications using standard normal models were mostly consistent with the main findings with slight differences in effect estimates or statistical significance (see Tables S2 and S3). Also, the results from the restricted analyses differed in the level of statistical significance but not the main findings (see Table S4). In addition, all main results were estimated without the use of propensity score weights. Although the estimation accounted for within‐person correlation, the results of the unweighted analyses should be interpreted with caution since there are possible systematic differences between the HARP enrolled and the HARP eligible control group.

4. DISCUSSION

The HARP MMC “carve‐in” was designed to include BH populations not previously enrolled in MMC and integrate physical and behavioral health within a special need managed care plan. The system goals were to decrease ER and inpatient services, increase outpatient and recovery services, and improve the quality of health care. The preliminary findings reported here provide evidence of a positive impact of the HARPs on recipient outcomes in terms of increasing outpatient service utilization and decreasing inpatient service utilization. NYS will conduct a comprehensive statewide evaluation using an independent evaluator to document the overall impact of MMC implementation.

Although several studies have examined the impact of integrated MMC “carve‐in” for Medicaid beneficiaries with behavioral health conditions, the findings have been mixed with some evidence of effectiveness. ¹² , ²⁶ , ²⁷ Nevertheless, much of the findings seem to suggest that MMC leads to reductions in medical cost, ER visits, fewer inpatient stays, and increased outpatient use. ¹² , ²⁸ , ²⁹

While the findings of this study suggest that HARPs were associated with a relative increase in outpatient services, a relative reduction in inpatient stays and length of stays, and a relative decline in ER services, it is of note that service use in both groups generally declined in the pre‐ to post‐HARP period comparisons. This may in part be due to the various health care reforms (ACA, DSRIP, CCMP, and MMC) in the health care delivery system which likely led to new and improved services. ² , ³ As described in the introduction, these reforms may have impacted changes in service utilization in addition to the HARPs. For example, DSRIP required providers to collaborate by forming a Performing Provider System to implement innovative projects focusing on system transformation, clinical improvement, and population health improvement. Also, the decline in services use may partly reflect individuals having care coordination, learning about their health, making informed choices while avoiding inappropriate services.

Individuals in the HARPs relative to comparison group were no different in the likelihood of using an inpatient service but were more likely to use ER and clinic services for behavioral and nonbehavioral reasons. The observed increases in the likelihood of using a service, especially clinic services, suggest that the HARPs led to some improvements in accessing outpatient services. This may be due to the initial plan announcements (advertising and education), and possibly to the initial outreach and enrollment efforts. These preliminary findings seem to suggest that the initial rollout of the HARP MMC benefit package was somewhat successful in meeting this goal.

The most important finding of this study was the relative impact of the HARPs on the average rate of service use. HARPs were associated with significant increases in average outpatient service use, decreases in average inpatient stays and length of stays, and decreases in average ER use between the two groups before and after implementation. The study showed evidence of relative reductions in inpatient stays for all‐cause and behavioral health‐related stays among the HARP enrolled compared to the eligible control group. Similar findings have been reported in studies that examine the effects of behavioral carve‐in for individuals with serious mental conditions. ¹² , ²⁹ Likewise, the study showed reductions in the average length of behavioral and nonbehavioral inpatient stays for individuals in the HARPs relative to comparison group. These reductions were not statistically significant in the ROS phase. The difference in outcomes between ROS residents (suburban‐rural) and those in NYC (urban) may reflect population, economic, and cultural differences. Individuals in ROS areas tend to be older, more unwell, less wealthy, have poorer health literacy, and transportation challenges than those in urban environments. ³⁰

This study found evidence of significant increases in behavioral and nonbehavioral outpatient clinic visits in both phases following the implementation. In contrast, a study by Kern et al ³¹ found an overall decrease in ambulatory visits for individuals switching from Medicaid fee‐for‐service to MMC. Although the study used a matched control DID approach, it did not differentiate between the types of visits, had a shorter follow‐up period of 12 months, and did not employ a more robust PS analysis in mitigating the baseline differences between groups. This study used a PS adjusted DID approach, and the findings are consistent with studies that used ¹² , ²⁹ similar approaches in evaluating the impacts of the integrated (carve‐in) plan in the states of Illinois and Washington. There was also evidence of increases in the use of OMH specialty services (ACT and PROS), although PROS was only statistically significant in the NYC phase. This is particularly important since gaining access and utilizing these specialty services are critical to the recovery and stability of patients with serious mental illness.

NYS collects bi‐weekly reports on claims information as well as monthly inpatient and quarterly outpatient utilization reviewed denials from health plans described elsewhere. ³¹ The reports highlight the complaints and appeals filed by consumers with MMC plans regarding service access, use, and quality of care. Although the preliminary analysis of the HARPs indicates positive gains, these complaints show that the plan has not always guaranteed health insurance coverage, has experienced variation in implementation, and effectiveness in different settings for different populations. In addition, the effects of Medicaid expansion (clinical integration and payment reforms) through the ACA on service delivery systems remain to be determined. Also, HCBS was not fully implemented during the time frame of the study. Therefore, rather than viewing the HARPs as a successful or failed plan, we should try to understand the complexity of the areas in which the HARPs offer the greatest value.

The strengths of this preliminary analysis included the ample observation time for the HARP enrolled or eligible controls, the full two years pre‐ and follow‐up periods, and the rigorous analysis that used PS adjusted DID framework. Therefore, the findings of this study provide a comprehensive view on the early evaluation of MMC implementation and can serve as the foundation for future studies that will examine the long‐term effects of the MMC and newer initiatives.

This analysis had some limitations. First, the ramp‐up time for HARP implementation may have been greater than a year. We accounted for this by allowing a full‐year enrollment period followed by a two‐year post observation period. Second, while the study attempted to alleviate bias associated with baseline differences between groups by using a strong quasi‐experimental design with a robust set of individual characteristics, it is always possible that unmeasured differences may confound the impact estimates. Third, the impact of HARPs on service use before and after the MMC implementation does not necessarily reflect changes in quality of care or improvement in clinical outcomes. Future studies will investigate whether HARPs lead to improvements in quality of care and long‐term clinical outcomes.

The study findings support two assertions. First, compared to patients with similar clinical needs, the HARPs were associated with a relative increase in the likelihood of using ER and clinic services for various reasons. Access to comprehensive quality health care is important for managing disease, reducing disability, and premature death. Second and most importantly, the relative increases in average outpatient service use and reductions in average inpatient stays suggest that the HARPs led to a better engagement in the HARP group. This may also signal a shift toward gains in a fully integrated behavioral health service system.

CONFLICT OF INTEREST

The authors have no conflicts of interest to disclose.

Supporting information

Author matrix

Supplementary Material

Frimpong EY, Ferdousi W, Rowan GA, Radigan M. Impact of the 1115 behavioral health Medicaid waiver on adult Medicaid beneficiaries in New York State. Health Serv Res. 2021;56:677–690. 10.1111/1475-6773.13657