Racial/Ethnic Disparities in Substance Use Treatment in Medicaid Managed Care in New York City: The Role of Plan and Geography

Abstract

Objective.

To assess the magnitude of health care disparities in treatment for substance use disorder and the role of health plan membership and place of residence in observed disparities in Medicaid Managed Care plans in New York City.

Data Source.

Medicaid claims and managed care plan enrollment files for 2015–2017 in New York City.

Research Design.

We studied Medicaid enrollees with a substance use disorder diagnosis during their first six months of enrollment in a managed care plan in 2015–2017. A series of linear regression models quantified service disparities across race/ethnicity for five outcome indicators: treatment engagement, receipt of psychosocial treatment, follow-up after withdrawal, rapid readmission, and treatment continuation. We assessed the degree to which plan membership and place of residence contributed to observed disparities.

Results.

We found disparities in access to treatment but the magnitude of the disparities in most cases was small. Plan membership and geography of residence explained little of the observed disparities. One exception is geography of residence among Asian Americans, which appears to mediate disparities for two of our five outcome measures.

Conclusions.

Reallocating enrollees among Medicaid Managed Care plans in New York City or evolving trends in group place of residence are unlikely to reduce disparities in treatment for substance use disorder. System-wide reforms are needed to mitigate disparities.

INTRODUCTION

States have turned to Medicaid Managed Care (MMC) to provide health insurance to Medicaid enrollees.¹ In MMC, the state contracts with private insurers to take financial responsibility for health care for an enrolled population in exchange for a capitation payment. Racial/ethnic disparities exist in substance use disorder (SUD) treatment in MMC, including lower rates of treatment completion among Black and Latino clients compared to Whites.² Disparities in SUD treatment exist in other insurance sectors as well.^3,4 Black and American Indian patients are less likely than White patients to receive SUD services after an initial diagnosis, even after adjusting for socioeconomic factors and place of residence.⁵ Given severe pressures to limit costs and the high prevalence of members of minoritized racial/ethnic groups in Medicaid, MMC is arguably the most important sector in which to study health care disparities by race and ethnicity.

Research on disparities for SUD treatment in MMC rests in a broader literature. Black and Latino Medicaid enrollees have worse access to treatment and continuity of care than non-Latino Whites in every year 1992–1996.⁶ A more recent analysis of a 2014–2015 nationwide survey of Medicaid enrollees assessed disparities in getting needed care, getting care quickly, doctor communication, and customer service. Martino and colleagues⁷ found that, when compared to Whites, American Indian, Alaska Natives, and Asians or Pacific Islander enrollees reported worse experiences with care.

Health insurance can play a role in health care disparities.^8,9 If racial/ethnic minoritized enrollees are disproportionately enrolled in plans of low quality, remedial actions directed to those plans, or reallocation of enrollees among plans, could mitigate disparities. Place of residence can also play a role.^10,11 After matching place of residence of White to Black Medicare beneficiaries, Chandra and colleagues found that 69% of the Black-White disparities in care for myocardial infarction can be explained by place of residence.¹² In terms of SUD treatment, counties with higher percentages of Black or uninsured people are less likely to have at least one SUD facility accepting Medicaid.¹³ The empirical importance of place of residence in explaining any observed disparities in SUD treatment in MMC, however, is not well understood.^14,15

This paper assesses racial/ethnic disparities in SUD treatment services among one of the largest vulnerable populations in the U.S. -- low-income Medicaid enrollees with substance use treatment needs -- in one of the largest MMC programs in the country – New York City (NYC). After quantifying any racial/ethnic disparities within SUD treatment in MMC in NYC as assessed by five quality metrics, we study the role of place of residence and plan membership in explaining these disparities.

METHODS

Study Design

We applied the Institute of Medicine (now the National Academy of Medicine) definition of disparities, following the approach contained in the landmark Unequal Treatment Report.¹⁶ Unfair differences (disparities) in health care are differences not accounted for by differential group clinical need or preferences. Empirical studies applying this approach adjust for variables such as age correlated with health status, but in a more powerful approach to adjust for health status, studies also condition on clinical need prior to group comparision.^8,17 Our study relied primarily on careful conditioning on need to support interpretation of observed differences as disparities. Following most work on health care disparities, we did not adjust for preferences.

The Institute of Medicine approach to disparities can be implemented by omitting factors, such as socioeconomic status (SES), that might mediate disparities.¹⁷ Dropping an SES variable implies that disparities mediated through the variable will be absorbed in the estimated coefficient for race and ethnicity. We exploited this property of disparities measurement by estimating regression models with and without plan indicators, and with and without indicators for place residence, to assess the degree to which measured disparities are affected by inclusion of these variables.

Data

We studied disparities in SUD treatment in eight MMC plans in NYC where membership in a MMC plan is mandatory. Our data consisted of six months of claims of adults, aged 18 to 64, with at least six months of enrollment in a plan. Individuals appear in the data once, during their first episode of at least six months of enrollment.

Administrative records from the New York State Department of Health (NYSDOH) supplied data on monthly plan membership, insurance claims covered by Medicaid, and some demographic information, including reason for Medicaid eligibility, and residential zip code. The NYSDOH employs standardized coding of diagnosis and procedures in claims. Berenson-Eggers Type of Service codes are used to assign codes for outpatient data. For the inpatient data, we used the Clinical Classifications Software to assign each inpatient admission to a clinically meaningful category based on the primary diagnosis.

Performance Indicators.

SUD performance indicators appear in research and policy reports.^18–20 We selected five as the most likely to be measured reliably in our data and informative about disparities in SUD treatment. As noted, all measures are conditioned on some prior treatment in order to adjust for health status. For example, our first indicator, receipt of a psychosocial intervention is measured only among those with an SUD diagnosis. Table 1 contains a brief definition of each of the five indicators. The Supplemental Digital Content table provides details on the construction of the five indicators.

Table 1.

Performance Metrics: Substance Use Disorder in New York City Medicaid Managed Care

Performance Indicator	Definition
Receipt of psychosocial interventions	Receipt of at least one psychosocial intervention among those with SUD.
SUD Tx engagement: follow-up w/ in 30 days after ED visit	Any outpatient visit, intensive outpatient encounter, or partial hospitalization indicating alcohol or other drug treatment that occurred within 30 days of an ED visit with a primary diagnosis of SUD.
Follow up after withdrawal (detox) management	Receipt of SUD follow-up care within 14 days after a diagnosis for drug withdrawal or discharge from a withdrawal management episode, including services in outpatient, inpatient, or residential settings.
Rapid readmission to inpatient SUD care 2–30 days post-discharge	Readmission to acute substance use care within 2–30 days of discharge.
Treatment continuation	Among enrollees with SUD treatment (including inpatient/residential and outpatient or methadone services), the share who continued treatment within 14 days of the initial SUD treatment claim for 3 months or more.

Abbreviations: SUD, substance use disorder; Tx, treatment; ED, emergency department

Race/Ethnicity.

Enrollment files contain an imperfect variable with five values: White, Black, Asian, Native American, and Puerto Rican/Hispanic, along with an “Other” and “Unknown” categories. A common challenge in analyzing disparities in Medicaid populations is missing or invalid codes for race/ethnicity information.²¹ Racial/ethnic data are incomplete for approximately 70% of MMC care beneficiaries nationwide,²¹ and between 20–50% in NYS.²² In NYS, this variable is known to assign many Hispanic respondents to the “Other” category.^23–25 In our main specification, we combined the Other and Hispanic categories based on literature identifying this pattern for Hispanic populations.^26–29 We also estimated all our models with Other and Hispanic as separate categories. In addition, we explored dealing with misclassification with a prediction model for race/ethnicity based on supplemental information from another Medicaid project in NYC which collects self-reported race and ethnicity.³⁰ Specifically, using data from the other project, we estimated a multinomial regression model predicting self-reported race/ethnicity as a function of the NYS-designated race/ethnicity, gender, age, and Medicaid enrollment status, all variables available in our data. Applying the prediction model to our data enabled us to re-estimate all our models substituting predicted race/ethnicity for the values in the NYS administrative data.

Analytic Plan

Three linear regression models were estimated for each outcome indicator. All regressions include age, gender, and race/ethnicity with White the reference category. Model 1 does not include indicators for plan or geography whereas Models 2 and 3 add fixed effects of plan membership and zip code (i.e., place of residence) respectively. In Model 1, the race/ethnicity variable is a full estimate of racial/ethnic disparities in relation to Whites, including any effect mediated through plan or geography (or any other factors). In Model 2 adding plan membership, the race/ethnicity variable measures disparities not mediated through plans. For example, if the coefficient on Black race in Model 1 indicates a 20% lower quality than Whites, and inclusion of a plan membership variable reduces this estimate to 15% lower, results imply that differential group membership in health plans explains 5% of the racial/ethnic disparity. Model 3’s race/ethnicity variable measures disparities not mediated either through plan or geography.

RESULTS

The total sample consists of 559,638 enrollees aged 18 to 64 from NYC whose first six months of enrollment in one of the MMC plans started during 2015–2017 (see Table 2). Among these, 35,069 qualified for inclusion in at least one of the indicators listed in Table 1, essentially meaning that at some point during their initial six months of enrollment they had a claim indicating a substance use treatment. Approximately 38% of the entire sample is female, and the average age is 40 years. The racial/ethnic distributions are roughly equal among the three major racial/ethnic groups (27.6–33.2%) as reported in the data, with a smaller number of individuals designated as Asian (6.2%). There were 53 aid category codes in the data.³¹ In Table 2, we report the four aid category variables with at least 4% of the sample that in total cover about 85% of the entire sample.

Table 2.

Sample Characteristics, NYC Medicaid Population in Eight MMC Plans, 2015–2017, N=35,069

	Mean/percent
Demographics
Age when first enrolled	39.6
Female	37.6%
NYS Medicaid Designated Race/Ethnicity
White	33.2%
Black	27.6%
Asian	6.2%
Hispanic/Other*	33.0%
Aid Category **
Low-income family with deprivation	8.0%
Singles and childless couples with income ≤ 100% FPL	57.5%
Singles and childless couples with income 100%−138% of FPL	7.9%
Adult caretaker with income ≤ 133% of the FPL	3.9%
Other	22.7%
Performance Indicators
Psychosocial treatment	31.4% (N = 35,066)
SUD Tx Engagement	19.5% (N = 2,151)
Follow up after withdrawal	49.3% (N = 1,890)
Rapid readmission	27.3% (N = 6,469)
Treatment continuation	13.1% (N = 35,066)

Abbreviations: NYC, New York City; MMC, Medicaid Managed Care; NYS, New York State; FPL, Federal Poverty Level

^*

Membership in the Other Category was as follows:

• OTHER 9,869 unique members
• PUERTO RICAN – HISPANIC, 5,291 unique members
• AMERICAN INDIAN (NATIVE AMERICAN) – OR ALASKAN NATIVE, 2,175 unique members
• UNKNOWN, 18,636 unique members

^**

Aid Categories were defined as follows:

• 32 – Low-income family with deprivation
• 90 – Singles and childless couples with income ≤ 100% of the federal poverty level
• H0 – Singles and childless couples with income 101–138% of the federal poverty level
• H1 – Adult (19–64) parent and caretaker relatives with income < 133% of the federal poverty level, or age 19–20.

For more information on the definition of aid categories, see https://www.health.ny.gov/health_care/medicaid/reference/mrg/category.pdf

Table 2 also shows the sample mean for each quality indicator, as well as the number of observations for which the indicator applies. For the psychosocial treatment and treatment continuation indicators, all observations with an SUD are included in the denominator of the group rates. The other three indicators are conditioned on a prior event: an ED visit, detox treatment, or a discharge from inpatient care, respectively.

Variation in plan enrollment by race/ethnicity leaves the potential open for plans to play a role in disparities. Plan enrollment ranges from 14% to 27% for Black enrollees, 9% to 30% for White enrollees, 10% to 26% for Asians, and 35% to 48% for Hispanic/Other enrollees. Zip codes are characterized by substantial variation in group composition. For example, among the zip codes with 100+ enrollees in our data, the mean percent Black is 36.7% with a two-standard deviation of 20.7%.

Regression Results

Column (1) of Table 3 compares the quality indicator for the three minoritized race/ethnicity groups to Whites in a regression adjusting for age/gender. With respect to the presence of some psychosocial treatment, Blacks are significantly more likely to receive this than Whites, indicating no disparity unfavorable to Blacks, whereas Asian and Hispanic/Other enrollees receive psychosocial treatment at rates significantly lower than Whites, indicating a disparity. All groups are less likely than Whites to receive treatment engagement and to receive follow-up after withdrawal. All three groups, however, have lower rates of rapid readmission than Whites indicating no disparity for this negative indicator of quality. Finally, with respect to treatment continuation, Asian and Other/Hispanic enrollees, but not Blacks, are treated favorably in relation to Whites.

Table 3.

Plan and Place Mediation in Disparities in SUD Treatment

Psychosocial treatment	Full Disparity (1)	95% CI	Adding Plan (2)	95% CI	Adding Place (3)	95% CI	Disparity	Plan Mediation	Place Mediation
Black	0.030	[0.019, 0.041]	0.026	[0.014, 0.037]	0.014	[0.0015, 0.025]	no	N/A*	N/A*
Asian	−0.174	[−0.192, −0.156]	−0.173	[−0.191, −0.155]	−0.151	[−0.170, −0.132]	yes	no	yes/NS
Other / Hispanic	−0.054	[−0.064, −0.043]	−0.058	[−0.068, −0.047]	−0.057	[−0.068, −0.046]	yes	no	no
SUD Tx engagement
Black	−0.025	[−0.066, 0.016]	−0.031	[−0.072, 0.010]	−0.028	[−0.072, 0.016]	no	N/A*	N/A*
Asian	−0.123	[−0.210, −0.033]	−0.110	[−0.200, −0.020]	−0.103	[−0.199, −0.006]	yes	no	no
Other / Hispanic	−0.063	[−0.105, −0.021]	−0.066	[−0.108, −0.024]	−0.068	[−0.112, −0.023]	yes	no	no
Follow up after withdrawal
Black	−0.063	[−0.082, −0.043]	−0.066	[−0.086, −0.047]	−0.069	[−0.091, −0.049]	yes	no	no
Asian	−0.179	[−0.211, −0.147]	−0.179	[−0.211, −0.147]	−0.163	[−0.197, −0.130]	yes	no	yes/NS
Other / Hispanic	−0.125	[−0.144, −0.106]	−0.128	[−0.147, −0.109]	−0.122	[−0.141, −0.102]	yes	no	no
Rapid readmission
Black	−0.035	[−0.057, −0.014]	−0.046	[−0.068, −0.025]	−0.058	[−0.080, −0.035]	no	N/A*	N/A*
Asian	−0.129	[−0.169, −0.089]	−0.131	[−0.171, −0.091]	−0.104	[−0.146, −0.063]	no	N/A*	N/A*
Other / Hispanic	−0.099	[−0.119, −0.078]	−0.105	[−0.126, −0.085]	−0.097	[−0.119, −0.075]	no	N/A*	N/A*
Treatment continuation
Black	0.023	[0.016, 0.031]	0.0205	[0.013, 0.028]	0.005	[−0.004, 0.013]	no	N/A*	N/A*
Asian	−0.091	[−0.104, −0.079]	−0.091	[−0.104, −0.079]	−0.084	[−0.098, −0.071]	yes	no	no
Other / Hispanic	−0.015	[−0.023, −0.008]	−0.018	[−0.026, −0.011]	−0.023	[−0.031, −0.016]	yes	no	no

Notes:

Italicized entries are significant at p < .01; bold entries are significant at p < .001.

Equation (1) no plan or geography fixed effects; equation (2) adds plan effects to (1); equation (3) adds zip code fixed effects to (2).

Non-Hispanic White is the omitted race/ethnicity category.

^*N/A represents items with no disparities.

Estimated coefficients can be normed against sample means to identify which effects are large in relation to population averages. For example, an estimated coefficient for Asian enrollees of −.174 for psychosocial treatment means Asians are estimated to receive psychosocial treatment at a rate 17.4 percentage points lower than Whites whereas the sample mean reported in Table 2 is 31.4%, implying that Asians receive psychosocial treatment at about only half the rate of the average enrollee. By contrast, while the Hispanic/Other coefficient in the psychosocial treatment regression is negative and significant, the estimated effect, a 5.4% lower rate, is smaller. The Asian estimate for treatment continuation of −.091 implies that Asians continue treatment at a rate 9.1 percentage points less frequently than Whites, scaled against the sample mean reported in Table 2 of 13.1%. Asians continue treatment at lower rates than all other racial/ethnic groups. Asians are subject to large disparities for SUD treatment engagement and follow up after withdrawal as well.

Column (2) of Table 3 reports results from regressions including indicators for health plan of enrollment. Evidence that plan membership contributes to disparities would consist of a smaller estimate for the race/ethnicity variable when controlling for plan. After controlling for plan, the race/ethnicity coefficient picks up only disparities remaining after adjusting for a plan effect. We find no evidence of plan mediation. Where we find evidence of significant disparities in relation to Whites (indicated by a “yes” in Column (4)), including plan effects has no significant negative effect on the estimate of the magnitude of disparities. Throughout Table 3, the estimated coefficients in Column 2 are very close to those in Column 1.

Column (3) of Table 3 adds zip code fixed effects to the models with plan effects to investigate the role of geography. There is modest evidence in our data for place of residence playing a role in contributing to disparities. For Asians, the estimated disparity in relation to Whites is slightly lower for our psychosocial treatment and follow-up after withdrawal performance indicators, implying that place of residence (e. g. geography) explains part of the disparities for this group, possibly because of low access to treatment in these neighborhoods.

Robustness Checks

We conduct three sets of robustness checks on our results, with detailed results available from the authors. In a first check, we reran all the regressions reported in Table 3 adding indicators for the aid codes listed in Table 2. The results changed very little. There were no changes in the group-indicator pattern of evidence for disparities.

Two robustness checks were directed towards misclassification of race and ethnicity. In a first check, we separated the Hispanic and Other categories and reran all regressions. For follow up after withdrawal and rapid readmission, the separated groups were subject to disparities to about the same degree. For the other three indicators however, all of the disparities reported in Table 3 fall on the Other category. Hispanics identified in the data as Hispanics may differ systematically from those misclassified as Other. Separating these two groups had no meaningful effect on the estimate of disparities for Asians or Blacks. Our second robustness check on race/ethnicity was based on predictive models for self-reported race/ethnicity. Unfortunately, the predictive models were not powerful in distinguishing group membership. Ideally, such predictive models would yield high likelihood of membership in a particular group for each individual, with low probabilities for other groups. Ours did not yield sharp predictions, limiting the validity of disparity estimates based on predicted race/ethnicity. Subject to this caveat, the disparities estimates based on models including predicted race/ethnicity were generally higher than those in Table 3 using Medicaid-reported race/ethnicity. Similar to Table 3, there is little evidence that plan or geography play a role.

DISCUSSION

We find a pattern of disparities in receipt of psychosocial treatment and treatment engagement for SUD care for Asians and Hispanics but not for Black enrollees. The finding for Asians and Hispanics could be partly explained by a shortage of culturally and linguistically matched services for this population.^32,33 These findings were highlighted in our earlier work detailing how non-English speaking patients, mostly Asians, who required residential services were limited to outpatient clinic visits due to inconsistent language supports across inpatient levels of care.³⁴ For treatment engagement, the presence of a small Hispanic disparity (6.3% lower for Hispanics) but not a Black-White disparity is consistent with Pinedo’s analysis of the 2015–2017 National Survey on Drug Use and Health.³⁵ When controlling for insurance, Hispanic but not Black adults were less likely than White enrollees to use specialty treatment for substance use.³⁵ Pinedo’s study did not include Asians, and unfortunately many similar disparities studies also do not include Asians as a separate category.³⁶ Because in NYS many Hispanic respondents classify themselves in the “Other” category^23–25 particularly Black Latinos, our findings suggest that disparities might be even larger for those with double minoritized status. Compounded disparities by language and race for those in need of SUD care might put them at a greater disadvantage than their White non-Hispanic counterparts. In our main specification, we combine the Other and Hispanic categories and confirm the disparities.

The finding that Asian enrollees were 12.1% less likely than Whites to engage in SUD treatment is consistent with a nationwide study of Medicaid beneficiaries’ experiences with their healthcare.⁷ Asians report worse experiences than White beneficiaries, while Black beneficiaries report better experiences than White beneficiaries, and the results for Hispanic-White differences are mixed.⁷ Interpretation of disparities results needs to be made recognizing that treatment engagement for SUD is low across all racial/ethnic groups. While identification of SUD in Medicaid plans doubled nationwide during the period of this study, rates of SUD treatment initiation and engagement remained mired at low levels (15.8% and 12.1%, respectively).¹⁴

With all racial/ethnic groups combined, half of enrollees in our Medicaid sample got follow-up services within 14 days of detoxification; these rates are superior to those from 2008 for Medicaid enrollees in other states – when only a third of people received follow-up services within 14 days.³⁷ This is an important indicator because, after inpatient treatment or detox, follow-up services, especially medication-assisted treatment or residential treatment for SUD, may help prevent readmissions.

With respect to treatment continuation, the fact that almost 90% of enrollees do not continue care for at least three months for a chronic and treatable condition signals the need to redesign the substance use treatment system, care management, and the net value they offer delivering care.

Managed care poses peril and promise for SUD treatment and disparities. Barriers to treatment entry and continuation imposed by state policies and managed care plans include narrow medical necessity criteria, inadequate information available to patients seeking treatment, lack of standards for wait time or network adequacy.³⁸ Strict limitations (e.g., pre-authorization requirements)³⁸ disrupt care continuity and contribute to poor SUD outcomes.^39,40 On the other hand, MMC can improve matters over fee-for-service Medicaid which controls costs by limiting levels of provider reimbursement, also a cause of access problems. Although disparities were not eliminated altogether, Massachusetts Medicaid transition to MMC care improved the quality of treatment for Black and Hispanic clients and reduced racial/ethnic service disparities in behavioral health care.⁶

Improving performance of health care providers or developing and disseminating low-cost interventions such as non-invasive and generic treatments (i.e. use of beta blockers as first line treatment instead of cardiac catheterization for treating acute myocardial infarction) has the potential to reduce health care disparities.¹² The same logic applies to MMC plans. Policies to improve the quality of treatment for substance use disorder may, at the same time, reduce disparities.

Our results indicate that while some disparities exist in MMC in NYC, plan membership or geography of residence do not seem to exacerbate or ameliorate SUD treatment disparities. We found that place of residence mediated disparities for two of the five outcome measures for Asian Americans, which may be indicative of low access to adequate treatment in these neighborhoods. Data compiled through the NYC Neighborhood Health Atlas reports place-based differences in health and healthcare outcomes by zip code, highlighting marked differences in health access and outcomes based on place of residence.⁴¹ Most Asian Americans in NYC reside in Chinatown (62%) and Flushing (68.8%), many of whom report significant limited English proficiency.⁴¹ While there are some SUD treatment centers in areas where most Asians live⁴², many clinics in NYC lack the resources to adequately work with linguistic minorities and meet their specific treatment needs.³⁴

Health plan membership could ameliorate disparities if members of minoritized racial/ethnic groups were disproportionately enrolled in plans of higher quality, but we find no evidence for plan effects either way. The practical import of our finding is that reallocating enrollees among plans to reduce racial/ethnic disparities would have little effect on system-wide disparities. Changes at the system level are needed to make plans accountable for their performance in SUD treatment. Nonetheless, our finding of no apparent “plan effects” on disparities does not absolve plans from responsibility to address disparities. Policies directed at health plans, for example, enforcing standards for adequate access to SUD treatment, remain good policy candidates for reducing disparities.

Our study has limitations. First, MMC members with untreated SUD during their first six months of enrollment do not appear in our data at all. We miss racial/ethnic disparities in care that may exist in initial recognition and treatment. Second, as in virtually all the literature measuring disparities, statistical adjustments and conditioning on treatment events are imperfect adjustments for health status. Third, our study shares research limitations imposed by missing or miscoded data on race and ethnicity. Research on disparities among the privately insured is hampered by the unavailability of race/ethnicity data, so most disparities research has relied on data from public payers – Medicare, Medicaid, the Veterans’ Administration – or on survey data. Medicaid data predominantly include race/ethnicity indicators, but measurement error can impair inferences about the magnitude of disparities.^43,44

Fourth, it is important to keep in mind that an association of plan membership or zip code of residence with disparities may not be causal. As in any non-experimental cross-sectional empirical analysis, the observed association of an outcome result with plan indicators could be due to something the plan does or with unmeasured variables associated with a plan: for example, unmeasured dimensions of case mix of enrollees. Similarly, and more obviously, zip code of residence does not itself cause disparities. It is some factor related to zip code (e.g., access to public transportation) that is ultimately responsible. We believe these findings offer us opportunities to chart the way for addressing these disparities.