The complex relationships among race/ethnicity, social determinants, and opioid utilization

Abstract

Objective

To examine individual‐ and community‐level factors associated with racial/ethnic differences in individuals’ opioid prescription use.

Data Sources

Outpatient opioid prescription utilization and demographic, socioeconomic, and health characteristics from a nationally representative sample of the US noninstitutionalized civilian population obtained from 2013‐2016 Medical Expenditure Panel Survey (MEPS) data and combined with 2012‐2016 American Community Survey data and 2015 Health Area Resources File data.

Study Design

We use the Oaxaca‐Blinder decomposition method to disaggregate racial/ethnic differences in prescription opioid utilization into differences explained by underlying predisposing, enabling and need characteristics, and unexplained differences.

Data Collection/Extraction Methods

We use restricted‐use geographic identifiers to supplement the MEPS data with information on community characteristics and local health care resources.

Principal Findings

The average annual rate of any outpatient opioid prescription use was higher for non‐Hispanic whites (15.8%; standard errors [SE]: 0.3) than for non‐Hispanic blacks and Hispanics by 1.4 percentage points (SE: 0.5) and 6.2 percentage points (SE: 0.4), respectively. The smaller difference between non‐Hispanic blacks and whites is not explained by the differences in the risk factors, while almost all the difference between Hispanics and non‐Hispanic whites can be explained by the differences in the means of the risk factors. The differences in the prevalence of pain, the rate of being United States‐born, and the racial/ethnic composition of the community explain 2.4 (SE: 0.2), 1.4 (SE: 0.3), and 1.9 (SE: 0.4) percentage‐point differences, respectively. Pain prevalence explains the difference regardless of opioid potency, while foreign‐born status and community racial/ethnic composition explain the difference in higher‐potency opioid utilization only.

Conclusions

This study underscores the importance of accounting for both individual and community characteristics when investigating patterns in opioid use. Our results could assist policy makers in tailoring strategies to promote safer and more effective pain management based on individual and community characteristics.

What is Known on This Topic

• Previous studies found that non‐Hispanic whites were more likely to use outpatient opioid prescriptions compared with Hispanics in 2015‐2016 and that the opioid overdose death rate of non‐Hispanic whites in 2017 was higher than those of minority groups.

• The sources of these racial/ethnic differences in prescription opioid use have not been adequately explored by prior research.

What This Study Adds

• The main risk factors that explain the Hispanic vs. non‐Hispanic white difference in outpatient prescription opioid use are (a) prevalence of pain (regardless of opioid potency) and (b) foreign‐born status and community racial/ethnic composition (only for higher‐potency opioids).

• The difference in outpatient prescription opioid use between non‐Hispanic blacks and non‐Hispanics whites is smaller and not explained by the differences in the observed characteristics.

• Understanding the different risk factors of the use of opioid prescriptions across race/ethnicity would help policy makers promote safer and more effective pain management.

1. INTRODUCTION

The opioid epidemic has had devastating effects on families and communities across the United States. In 2017, the number of drug overdose deaths involving an opioid was six times higher than that in 1999, and each day approximately 130 Americans died from an opioid overdose. ¹ Opioid prescriptions for pain management that evolve into nonmedical use are a key contributor to the current crisis, along with increased supplies of heroin and synthetic opioids. ² , ³ This epidemic has had a more profound impact on some population subgroups compared with others. The opioid overdose death rate for non‐Hispanic (hereafter NH) whites was 19.4 per 100 000 population in 2017, while the rate was 12.9 for NH blacks and 6.8 for Hispanics. ⁴ These descriptive statistics highlight the uniqueness of the current opioid epidemic as most drug epidemics in the past have disproportionately affected minority groups. ⁵

Prescription opioid utilization has been found to have the potential to lead to misuse of these medications, as well as addiction, diversion, and the use of heroin and illicitly manufactured synthetic opioids. ⁶ Indeed, racial/ethnic differences in prescription opioid use are qualitatively similar to differences observed for opioid overdoses. In 2015‐16, the use of outpatient opioid prescriptions was higher among NH white nonelderly adults (14.8%) compared with Hispanic nonelderly adults (8.5%), while there was no significant difference in use between NH white and NH black adults (13.0%). ⁷ Yet, the sources of such differences in prescription opioid use have not been adequately explored by prior research. Using prescription drug monitoring program data, Friedman et al show that the racial/ethnic and income patterns of opioid overdoses mirror opioid prescription rate patterns at the county level, suggesting that differential exposure to opioids via the health care system may have contributed to the large, observed racial/ethnic disparities in the opioid epidemic. Using the 2015 National Survey on Drug Use and Health, Han et al ⁹ provide descriptive statistics showing that economic disadvantage and behavioral health problems may be associated with higher rates of prescription opioid misuse. Hollingsworth et al ¹⁰ suggest that community‐level characteristics such as local macroeconomic decline are related to the rise of opioid‐related adverse events. These studies highlight the importance of investigating individual‐level and community‐level factors that may contribute to prescription opioid use.

The purpose of this study is to understand how racial/ethnic differences in individual characteristics and the characteristics of residential communities are associated with the racial/ethnic differences in individual‐level opioid prescription use. Using a nationally representative sample from the Medical Expenditure Panel Survey (MEPS) 2013‐2016 supplemented with community information from the American Community Survey (ACS) and the Health Area Resource File (HARF), we employ an Oaxaca‐Blinder decomposition method to explore factors contributing to racial/ethnic differences in the percentage of people with any outpatient opioid prescriptions per year. We further conduct the decomposition analysis on lower‐ vs. higher‐potency opioids separately to identify risk factors that are unique to higher‐potency opioids, which are more prone to misuse. Our findings could help policy makers more effectively design targeted policies and programs to promote safer pain management.

2. METHODS

2.1. Conceptual framework

This study uses Andersen's behavioral model for health care services utilization, which posits that usage of health services is determined by three types of factors (predisposing, enabling, and need factors) on two levels (individual and community). ¹¹ , ¹² Predisposing factors reflect individuals’ propensity to use health services and include demographic characteristics and beliefs at the individual level, and the demographic and social composition of communities, collective and organizational values, and cultural norms at the community level. Enabling factors are the resources that may facilitate access to services and include income and health insurance at the individual level and health care resources at the community level. Need factors represent potential needs for health service use, including self‐perceived health status and objectively measured health conditions. This study explains racial/ethnic differences in prescription opioid utilization in terms of these three factors, both at the individual and community level.

2.2. Data sources

The main data source is the 2013‐2016 MEPS sponsored by the Agency for Healthcare Research and Quality. The MEPS is a series of nationally representative surveys of the US civilian noninstitutionalized population conducted annually. The MEPS collects information on medical care use (including prescription drug use) and expenditures, as well as detailed data on health and sociodemographic characteristics. In this study, we include opioids that are used for pain relief and are identified using generic drug names for narcotic analgesics and narcotic analgesic combinations in the Multum Lexicon database from Cerner Multum, Inc (see Appendix S1 for details). ¹³

We use restricted‐use geographic identifiers to supplement MEPS with information on community and local health care resources. We obtain community information on racial/ethnic composition and income at the Census block group level from the 2012‐2016 ACS (only a 5‐year average is available). Block groups are the smallest geographic areas for which community‐level statistics are reported, containing between 600 and 3000 people. We obtain information on health care resources at the county level from the 2015 HARF.

We restrict our sample to adults aged 18‐64 years in the United States whose race/ethnicity is NH white, NH black, or Hispanic (N = 72 396). We further restrict the sample to those who responded to the MEPS’ Self‐Administered Questionnaire (SAQ) (N = 64 629) and then to those for whom the explanatory variables listed in Table 1 are not missing (exceptions are the two variables for which a “missing values” category is created) (N = 62 804).

TABLE 1.

Descriptive statistics for the sample of nonelderly adults by race/ethnicity

	NH whites	NH blacks	Hispanics
Dependent variable
Any opioid Rx use	0.158 (0.003)	0.143* (0.005)	0.096* (0.003)
Individual characteristics
Predisposing variables
Age
18‐26 y olds	0.168 (0.004)	0.217* (0.006)	0.237* (0.006)
27‐34 y olds	0.165 (0.005)	0.176 (0.006)	0.203* (0.005)
35‐44 y olds	0.193 (0.005)	0.203 (0.006)	0.239* (0.004)
45‐54 y olds	0.228 (0.005)	0.215* (0.004)	0.196* (0.005)
55‐64 y olds	0.246 (0.006)	0.189* (0.006)	0.126* (0.004)
Female	0.507 (0.003)	0.541* (0.006)	0.497 (0.004)
Married	0.577 (0.006)	0.326* (0.008)	0.474* (0.009)
Born in the United States	0.956 (0.003)	0.898* (0.009)	0.464* (0.012)
Attitude toward health care system: overcome illness without health care system
Disagree strongly	0.326 (0.005)	0.433* (0.008)	0.441* (0.008)
Disagree somewhat	0.239 (0.004)	0.193* (0.005)	0.179* (0.004)
Uncertain	0.134 (0.003)	0.117* (0.004)	0.129 (0.004)
Agree somewhat	0.237 (0.004)	0.163* (0.005)	0.154* (0.004)
Agree strongly	0.040 (0.002)	0.055* (0.002)	0.059* (0.003)
Missing value	0.025 (0.002)	0.039* (0.002)	0.038* (0.002)
Enabling characteristics
Has a usual source of care	0.763 (0.006)	0.681* (0.008)	0.599* (0.009)
Health insurance status
Uninsured	0.089 (0.004)	0.153* (0.005)	0.301* (0.012)
Any private insurance	0.803 (0.006)	0.613* (0.010)	0.511* (0.012)
Public	0.108 (0.005)	0.234* (0.008)	0.188* (0.009)
Education
Less than HS diploma	0.075 (0.003)	0.134* (0.006)	0.318* (0.009)
HS diploma	0.267 (0.006)	0.346* (0.009)	0.288* (0.006)
More than HS diploma	0.658 (0.007)	0.520* (0.009)	0.394* (0.009)
Not employed	0.233 (0.005)	0.302* (0.009)	0.271* (0.007)
Family income
<100% FPL	0.091 (0.003)	0.214* (0.008)	0.179* (0.010)
≥100% FPL, <125%FPL	0.029 (0.002)	0.051* (0.003)	0.061* (0.002)
≥125% FPL, <200%FPL	0.097 (0.003)	0.155* (0.005)	0.194* (0.005)
≥200% FPL, <400%FPL	0.286 (0.005)	0.313* (0.007)	0.336* (0.009)
≥400% FPL	0.497 (0.008)	0.267* (0.010)	0.229* (0.009)
Need characteristics
Pain limits normal work
Not at all	0.597 (0.005)	0.613 (0.008)	0.678* (0.006)
A little bit	0.230 (0.003)	0.187* (0.005)	0.177* (0.004)
Moderately	0.082 (0.002)	0.081 (0.003)	0.062* (0.002)
Quite a bit	0.058 (0.002)	0.070* (0.003)	0.044* (0.002)
Extremely	0.026 (0.001)	0.036* (0.002)	0.024 (0.001)
Missing value	0.007 (0.001)	0.012* (0.001)	0.016* (0.001)
Pain‐related health conditions a
Arthritis	0.227 (0.005)	0.192* (0.006)	0.114* (0.004)
Joint pain	0.476 (0.006)	0.390* (0.008)	0.308* (0.008)
Cancer	0.082 (0.003)	0.033* (0.002)	0.030* (0.002)
Other health conditions a
High blood pressure	0.262 (0.005)	0.349* (0.007)	0.200* (0.004)
Coronary heart disease	0.024 (0.001)	0.025 (0.002)	0.022 (0.001)
Angina	0.015 (0.001)	0.013 (0.001)	0.009*(0.001)
Heart attack	0.020 (0.001)	0.020 (0.002)	0.014* (0.001)
Other heart conditions	0.094 (0.003)	0.071* (0.004)	0.041* (0.003)
Stroke	0.020 (0.001)	0.032* (0.002)	0.014* (0.001)
Emphysema	0.016 (0.001)	0.007* (0.001)	0.003* (0.0005)
Chronic bronchitis	0.027 (0.002)	0.029 (0.002)	0.013* (0.001)
High Cholesterol	0.259 (0.004)	0.205* (0.006)	0.213* (0.005)
Diabetes	0.063 (0.004)	0.096* (0.003)	0.074* (0.003)
Asthma	0.108 (0.003)	0.121* (0.005)	0.071* (0.004)
Perceived physical health status
Excellent	0.270 (0.005)	0.260 (0.007)	0.266 (0.008)
Very good	0.375 (0.005)	0.308* (0.006)	0.285* (0.006)
Good	0.252 (0.005)	0.295* (0.004)	0.311* (0.007)
Fair	0.077 (0.003)	0.111* (0.004)	0.118* (0.003)
Poor	0.026 (0.001)	0.026 (0.002)	0.020* (0.001)
Perceived mental health status
Excellent	0.370 (0.006)	0.429* (0.010)	0.387 (0.008)
Very good	0.330 (0.005)	0.256* (0.006)	0.261* (0.007)
Good	0.230 (0.005)	0.242 (0.007)	0.285* (0.007)
Fair	0.056 (0.002)	0.060 (0.003)	0.057 (0.002)
Poor	0.014 (0.001)	0.014 (0.001)	0.009* (0.001)
Community characteristics
Predisposing variables
Racial composition at the Census block level
% of white residents	0.835 (0.005)	0.431* (0.014)	0.662* (0.012)
% of black residents	0.065 (0.003)	0.448* (0.018)	0.106* (0.006)
% of Hispanic residents	0.097 (0.004)	0.140* (0.010)	0.456* (0.019)
Region
Northeast	0.195 (0.010)	0.153* (0.013)	0.135* (0.013)
Midwest	0.259 (0.010)	0.171* (0.014)	0.089* (0.012)
South	0.348 (0.014)	0.584* (0.023)	0.367 (0.031)
West	0.199 (0.011)	0.092* (0.012)	0.409* (0.027)
MSA	0.826 (0.015)	0.916* (0.014)	0.943* (0.015)
Enabling characteristics
Median HH income in Census block group (in 10 000 dollars)	6.593 (0.097)	4.830* (0.089)	5.349* (0.095)
Primary care HPSA
Whole county	0.044 (0.010)	0.041 (0.013)	0.021* (0.005)
Part of county	0.833 (0.017)	0.883 (0.020)	0.934* (0.010)
No part of county	0.123 (0.016)	0.076* (0.015)	0.045* (0.008)
Dental care HPSA
Whole county	0.024 (0.007)	0.027 (0.007)	0.028 (0.009)
Part of county	0.826 (0.019)	0.871 (0.026)	0.914* (0.014)
No part of county	0.151 (0.018)	0.102 (0.024)	0.058* (0.010)
MDs per 1000 people in county	2.632 (0.073)	3.355* (0.087)	2.753 (0.082)
Log of MDs per 1000 people in County	1.172 (0.015)	1.362* (0.022)	1.247* (0.024)
N	26 794	14 030	21 980

N = 62 804. Standard errors in parentheses. Estimates are weighted and adjusted for the complex survey design of MEPS.

Abbreviations: FPL, federal poverty level; HH, household; HPSA, health professional shortage area; HS, high school; MD, medical doctor; MSA, metropolitan statistical areal; NH, non‐Hispanic; Rx, prescription.

^aPain‐related and other health conditions are the conditions that the individual was ever diagnosed with except that chronic bronchitis and joint pain are the conditions experienced in the past twelve months.

^*Statistically different from NH whites at the 5% level.

Sources: 2013‐2016 Medical Expenditure Panel Survey, 2012‐2016 American Community Survey, 2015 Health Area Resources File.

2.3. Outcome and determinant measures

Our outcome is an indicator variable that takes a value of one if the individual had any outpatient opioid pain prescription fills in a given year and zero otherwise. Determinants at the individual level include predisposing factors such as age, sex, marital status, and whether the individual was born in the United States. We also include a measure of attitudes toward the health care system obtained from a question in the MEPS SAQ. The question asks the extent to which one agrees that an illness can be overcome without help from a medically trained person. Enabling factors at the individual level are whether the individual has a usual source of care or not, health insurance status, family income as a share of the federal poverty level, as well as an individual's education and employment status. We have two sets of need factors. The first set includes a measure of pain that assesses the degree to which pain interfered with normal work outside the home and housework during the past four weeks. This measure is obtained from the SAQ, is validated, ¹⁴ and was used in a previous study. ¹⁵ We also include three pain‐related conditions: whether an individual was ever diagnosed with arthritis and cancer, and whether an individual had joint pain in the past twelve months. The second set of variables, included to capture overall health status, consist of other health conditions available in MEPS (see Table 1 for a complete list) and perceived physical and mental health status. We refer to these two sets as “pain prevalence” and “health status,” respectively, when discussing the decomposition results.

Predisposing community factors are the ratios of residents who are white, black, and Hispanic, respectively, and median household income (all at the Census block group level). Note that the proportion of residents who are white or black available at the Census block group level includes both Hispanic and non‐Hispanic residents. We also include indicators for residence in the US Census regions and for residence in a Metropolitan Statistical Area (MSA). The enabling factors at the community level are availability of local health care resources, as measured by a county's primary care and dental Health Profession Shortage Areas (HPSAs) status (as designated by the Health Resources and Services Administration) and the number of medical doctors per 1000 people in a county.

2.4. Analytic approach

We begin the analysis by describing the differences in the outcome and explanatory variables across racial/ethnic groups, using z‐score tests to test for significant differences. To explain the observed racial/ethnic differences in prescription opioid utilization, we use a regression‐based decomposition method developed by Oaxaca and Blinder. ¹⁶ , ¹⁷ Having originated in the field of labor economics to study wage disparities, the Oaxaca‐Blinder decomposition method is increasingly applied in health disparity studies. ¹⁸ , ¹⁹ , ²⁰ , ²¹ We run stratified (by race/ethnicity) linear probability models to disaggregate differences in prescription opioid utilization into differences that are explained by the underlying predisposing, enabling, and need characteristics (observed means of the explanatory variables) as well as the differences that are not explained by these underlying characteristics (unexplained differences). Our estimates of explained differences can be interpreted as a change in the minority group's outcome mean if the minority group's means of the explanatory variables were the same as those of NH whites. The details of our decomposition model and estimation are presented in Appendix S1. All estimates were weighted and adjusted for the complex survey design of the MEPS, and the differences mentioned in the text were statistically significant at the P ≤ 5% level.

We further investigate whether the decomposition results differ when the outcomes are defined based on the potency of drugs because the potency differs substantially across opioids, and stronger opioid prescriptions are associated with a higher risk of addiction and death due to overdose. ²² To determine the potency of each opioid, we use the morphine milligram equivalent (MME) conversion factor which is used to standardize the dose of a given opioid into the equivalent dose of morphine (ie, an MME conversion factor equal to 1 indicates that 1 unit of the drug has the same potency as 1 unit of morphine). ²³ We use the MME conversion factor to define two types of opioids: (1) low‐equivalency opioids (MME factor < 1) and (2) high‐equivalency/equivalent opioids (MME factor ≥ 1). ²⁴ We estimate separate models for any use of low and high equivalency/equivalent opioids during the year.

3. RESULTS

3.1. Observed differences by race/ethnicity

Table 1 shows that NH white nonelderly adults were on average more likely than NH black or Hispanic nonelderly adults to have at least one outpatient opioid pain prescription filled per year in 2013‐2016 (15.8%, 14.3%, and 9.6%, respectively). The remainder of the table shows that there are marked differences in predisposing, enabling, and need characteristics both at the individual and community levels by race/ethnicity. Minorities in the sample are, on average, younger and less likely to be married than NH whites. While most NH whites and NH blacks are born in the United States, more than half of Hispanics are foreign born. Minorities are also less likely than NH whites to believe that they can overcome illness without help from a medically trained person. In the enabling domain, minorities are less likely than NH whites to have a usual source of care or to be insured. They also tend to have less education and lower family incomes and are less likely to be employed compared with NH whites. In the need domain, the NH white vs. NH black differences are complicated—while NH blacks are less likely to report pain‐related conditions, heart conditions other than heart attack, emphysema, and high cholesterol, they are more likely to report high blood pressure, stroke, diabetes, and asthma. Also, NH blacks are more likely to report pain limiting normal work activities “quite a bit” or “extremely,” and to rate perceived physical health status as “Good” or “Fair” than NH whites. In contrast, Hispanics are uniformly less likely to report almost all health conditions and are more likely to report pain not limiting their normal work activities “at all” than NH whites though they are more likely to report perceived physical health status being “Good” or “Fair.”

In the predisposing domain at the community level, we observe substantial residential segregation: all three groups tend to concentrate in communities where their race/ethnicity is predominant. NH blacks are about 20 percentage points more likely than NH whites to live in the South, while Hispanics are also about 20 percentage points more likely than NH whites to live in the West. Minorities are more likely to live in urban areas compared with NH whites. Finally, in the enabling domain at the community level, NH whites tend to live in neighborhoods with higher median income and to live in a county where no part is designated as a primary care or dental care health professional shortage area (as compared to minorities). The number of medical doctors per 1,000 people is lower for NH whites than for minorities, possibly reflecting the tendency of minorities to live in denser urban areas.

3.2. Linear probability models

Table 2 shows the results for linear probability models on having any outpatient opioid pain prescription fills per year by race/ethnicity. Many individual‐level enabling and need factors are statistically significant among all three racial/ethnic groups. On the other hand, many individual‐level predisposing variables and community‐level variables have the anticipated sign, but their significance levels vary across the different groups.

TABLE 2.

Coefficient estimates from race/ethnicity‐specific linear probability models on any opioid prescription use

	NH whites	NH blacks	Hispanics
Individual characteristics
Predisposing variables
Age (reference = 18‐26 y old)
27‐34 y olds	0.013 (0.009)	0.023* (0.011)	0.013 (0.008)
35‐44 y olds	−0.004 (0.009)	0.016 (0.009)	0.016 (0.008)
45‐54 y olds	−0.026* (0.010)	0.005 (0.010)	0.006 (0.009)
55‐64 y olds	−0.047* (0.009)	−0.036* (0.012)	−0.013 (0.012)
Female	0.022* (0.005)	0.024* (0.007)	0.008 (0.005)
Married	0.007 (0.005)	−0.011 (0.008)	−0.002 (0.006)
Born in the United States	0.012 (0.011)	0.020* (0.009)	0.029* (0.006)
Attitude toward health care system: overcome illness without health care system (reference = disagree strongly)
Disagree somewhat	−0.019* (0.007)	−0.013 (0.009)	−0.013 (0.008)
Uncertain	−0.031* (0.008)	−0.020* (0.009)	−0.013 (0.008)
Agree somewhat	−0.026* (0.007)	−0.019* (0.009)	−0.022* (0.007)
Agree strongly	−0.039* (0.011)	−0.028 (0.015)	−0.014 (0.009)
Missing value	−0.037* (0.016)	0.011 (0.018)	−0.012 (0.012)
Enabling characteristics
Usual source of care	0.028* (0.005)	0.026* (0.006)	0.025* (0.005)
Health insurance (reference = uninsured)
Any private insurance	0.052* (0.009)	0.032* (0.008)	0.028* (0.005)
Public	0.082* (0.011)	0.056* (0.009)	0.044* (0.007)
Education (reference = less than HS diploma)
HS diploma	0.004 (0.010)	0.009 (0.010)	0.001 (0.006)
More than HS diploma	−0.009 (0.011)	0.027* (0.010)	0.005 (0.006)
Not employed	0.016* (0.007)	0.019* (0.008)	0.014* (0.006)
Family income (reference = <100%FPL)
≥100% FPL, <125%FPL	−0.027 (0.016)	−0.005 (0.016)	−0.008 (0.010)
≥125% FPL, <200%FPL	−0.007 (0.011)	−0.009 (0.011)	−0.010 (0.007)
≥200% FPL, <400%FPL	−0.020 (0.010)	−0.013 (0.009)	−0.015* (0.007)
≥400% FPL	−0.020 (0.011)	0.002 (0.011)	−0.015 (0.008)
Need characteristics
Pain limits normal work (reference = not at all)
A little bit	0.039* (0.006)	0.023* (0.010)	0.044* (0.008)
Moderately	0.142* (0.013)	0.099* (0.014)	0.067* (0.013)
Quite a bit	0.272* (0.016)	0.198* (0.017)	0.173* (0.021)
Extremely	0.330* (0.024)	0.262* (0.027)	0.222* (0.026)
Missing value	0.053 (0.030)	0.059 (0.037)	0.004 (0.014)
Pain‐related health conditions a
Arthritis	0.084* (0.009)	0.071* (0.012)	0.081* (0.012)
Joint pain	0.031* (0.006)	0.062* (0.007)	0.034* (0.007)
Cancer	0.055* (0.011)	0.084* (0.023)	0.047* (0.020)
Other Health conditions a
High blood pressure	0.019* (0.007)	0.018* (0.009)	0.009 (0.007)
Coronary heart disease	−0.021 (0.022)	−0.002 (0.033)	−0.0014 (0.025)
Angina	0.042 (0.026)	−0.075 (0.039)	−0.050 (0.033)
Heart attack	0.008 (0.025)	0.044 (0.030)	0.090* (0.032)
Other heart conditions	0.012 (0.010)	0.005 (0.019)	0.032 (0.020)
Stroke	0.029 (0.024)	0.002 (0.021)	0.038 (0.035)
Emphysema	0.020 (0.029)	−0.018 (0.040)	0.110 (0.075)
Chronic bronchitis	0.023 (0.019)	−0.004 (0.027)	0.018 (0.033)
High cholesterol	0.005 (0.007)	0.004 (0.010)	−0.001 (0.009)
Diabetes	−0.011 (0.013)	0.013 (0.014)	−0.030* (0.015)
Asthma	−0.003 (0.009)	0.025 (0.010)	0.013 (0.014)
Perceived physical health status (reference = excellent)
Very good	0.021* (0.007)	0.005 (0.009)	0.013 (0.007)
Good	0.040* (0.008)	0.019 (0.011)	0.019* (0.008)
Fair	0.099* (0.015)	0.081* (0.016)	0.038* (0.010)
Poor	0.162* (0.024)	0.117* (0.032)	0.109* (0.027)
Perceived physical health status (reference = excellent)
Very good	−0.008 (0.006)	−0.009 (0.008)	−0.004 (0.008)
Good	−0.017* (0.008)	−0.027* (0.011)	−0.006 (0.008)
Fair	−0.031* (0.013)	−0.027 (0.018)	−0.007 (0.015)
Poor	−0.070* (0.027)	−0.110* (0.041)	−0.013 (0.036)
Community characteristics
Predisposing variables
Racial composition at the Census block level
% of white residents	−0.022 (0.030)	0.060* (0.029)	0.007 (0.014)
% of black residents	0.020 (0.040)	0.046 (0.028)	−0.003 (0.023)
% of Hispanic residents	−0.017 (0.020)	−0.031 (0.022)	−0.050* (0.009)
Region (reference = Northeast)
Midwest	0.032* (0.008)	0.016 (0.012)	0.027* (0.013)
South	0.033* (0.008)	0.023* (0.010)	0.018 (0.009)
West	0.027* (0.009)	0.037* (0.013)	0.019* (0.009)
MSA	0.011 (0.009)	0.015 (0.013)	−0.011 (0.013)
Enabling characteristics
Median HH income in Census block group (in $10 000)	0.00002 (0.001)	−0.004* (0.002)	−0.003* (0.001)
Primary care HPSA (reference = whole county)
Part of county	0.025* (0.013)	0.026 (0.015)	0.013 (0.021)
No part of county	0.025 (0.013)	0.026 (0.018)	−0.010 (0.033)
Dental care HPSA (reference = whole county)
Part of county	0.004 (0.014)	−0.026* (0.013)	−0.002 (0.016)
No part of county	−0.004 (0.014)	−0.014 (0.016)	0.015 (0.022)
Log of MDs per 1000
People in county	−0.014* (0.007)	−0.008 (0.010)	−0.0003 (0.009)
N	26 794	14 030	21 980

N = 62 804. Standard errors in parentheses. Regressions are run separately for each racial/ethnicity group. The dependent variable is an indicator variable that takes a value of one if the individual had any outpatient opioid pain prescription fills in a given year and zero or otherwise. Estimates were weighted and adjusted for the complex survey design of MEPS.

Abbreviations: FPL, federal poverty level; HH, household; HPSA, health professional shortage area; HS, high school; MD, medical doctor; MSA, metropolitan statistical area; NH, non‐Hispanic.

^aPain‐related and other health conditions are the conditions that the individual was ever diagnosed with except that chronic bronchitis and joint pain are the conditions experienced in the past 12 mo.

^*Denotes significance at the 5% level.

Sources: 2013‐2016 Medical Expenditure Panel Survey, 2012‐2016 American Community Survey, 2015 Health Area Resources File.

Starting from predisposing factors at the individual level, being born in the United States is associated with a higher probability of having any opioid pain medication fills among minorities. Agreeing with the “overcome illness without the health care system” statement is negatively associated with the outcome, especially among NH whites. Enabling factors, such as having a usual source of care and health insurance, are positively correlated with the outcome among all racial/ethnic groups. Not being employed is associated with a higher probability of having filled opioid pain medication while family income is not a significant risk factor among any of the racial/ethnic groups. Need factors, such as self‐reported pain and pain‐related health conditions, are strong predictors in all racial/ethnic groups. While most other health conditions are not significant risk factors, perceived physical health status is a significant predictor across racial/ethnic groups. Negative association with worse perceived mental health status and the outcome among NH whites and NH blacks likely reflect the fact that combining opioids and benzodiazepines, which are prescribed for anxiety and sleeping problems, can be unsafe.

At the community level, the percentage of residents who are white is positively associated with the likelihood of NH blacks having any opioid pain prescriptions. Among Hispanics, the percentage of Hispanic residents is negatively associated with the likelihood of Hispanics filling opioid pain prescriptions. For example, for every 10 percentage‐point increase in the ratio of residents who are Hispanic, the probability that Hispanic nonelderly individuals have any opioid pain prescriptions decreases by 0.5 percentage points. Regardless of race/ethnicity, living in the West Census region is associated with a higher propensity to fill opioid pain prescriptions (reference group is Northeast). The results on enabling characteristics at the community level suggest that higher median household income is negatively associated with the outcome among minorities, while the availability of local health care resources is significantly associated with the outcome only among NH whites and NH blacks.

3.3. Decomposition analysis

Table 3 displays the results of our decomposition analysis. The first two rows show that NH blacks are 1.4 percentage points less likely than NH whites to use prescription opioids and that if NH blacks had the same characteristics as NH whites, the differences in opioid utilization would not change significantly (the numbers might not add up due to rounding). We present the extent to which the difference in prescription opioid use can be attributed to the mean differences of specific sets of explanatory variables in the remainder of the rows for ease of interpretation (the results for each explanatory variable available in Table S3). ¹⁸ , ¹⁹ The difference in several sets of explanatory variables are positively or negatively associated with the outcome difference between NH blacks and NH whites. For example, if the rate of NH blacks with a usual source of care increases to equal the rate among NH whites, the difference in the ratio of those with any opioid prescription fills would decrease by 0.2 percentage points because having a usual source of care is associated with a higher rate of the outcome. Likewise, if NH blacks had the same pain prevalence as NH whites, the difference in the percentage of those having prescription opioid use compared with NH whites would narrow by 0.8 percentage points, while if they had the same health status as those of the NH whites, the difference would widen by 0.6 percentage points. The effects of these two sets of need factors cancel out, and the overall association between need factors and the outcome is not statistically significant (not shown in table). If NH blacks had the same demographic characteristics and beliefs as NH whites, the difference in outcome would widen by 0.6 and 0.2 percentage points, respectively. At the community level, if NH blacks lived in a neighborhood with the same median household income as NH whites, the differences in the outcome would widen by 0.7 percentage points.

TABLE 3.

Decomposition results (in percentage points)

	NH blacks versus NH whites	SE	Hispanics versus NH whites	SE
Actual differences with any opioid prescription use	1.430*	0.540	6.190*	0.440
Amount of difference attributable to differences in means	0.043	0.653	6.108*	0.635
Amount of difference attributable to differences in means of specific variables
Individual characteristics
Predisposing variables
Age, sex, and marital status	−0.604*	0.193	−0.275*	0.132
Born in the United States	0.114*	0.056	1.433*	0.287
Attitude, health care system	−0.205*	0.089	−0.228*	0.087
Enabling characteristics
Has a usual source of care	0.213*	0.058	0.404*	0.078
Health Insurance	−0.104	0.142	0.457*	0.156
Education, employment, family Income	0.307	0.232	−0.138	0.218
Need characteristics
Pain prevalence	0.759*	0.249	2.376*	0.230
Health status	−0.579*	0.154	0.416*	0.207
Community characteristics
Predisposing variables
Race composition	0.799	0.600	1.941*	0.382
Region, rural/urban (MSA)	−0.148	0.252	0.148	0.235
Enabling characteristics
Median income	−0.702*	0.278	−0.372*	0.145
Health care resources	0.192	0.170	−0.054	0.138

N = 62 804. Estimates were weighted and adjusted for the complex survey design of MEPS.

Abbreviations: MSA, metropolitan statistical area; NH, non‐Hispanic; SE, standard errors.

^*Denotes significance at the 5% level.

Data Sources: 2013‐2016 Medical Expenditure Panel Survey, 2012‐2016 American Community Survey, 2015 Health Area Resources File.

Our Hispanic vs. NH white decomposition results are very different from those for NH black vs. NH whites. Hispanics are 6.2 percentage points less likely than NH whites to use prescription opioids, and the decomposition results show that if Hispanics had the same characteristics as NH whites, the proportion of Hispanics having an opioid prescription filled would be almost the same as NH whites. If Hispanics had the same demographic characteristics and attitude toward the health care system as NH whites, the difference in outcome would widen by 0.3 and 0.2 percentage points, respectively. For Hispanics, if the rate of those born in the United States increases to equal that of NH whites (ie, a 49 percentage point increase), the difference in the outcome would decrease by 1.4 percentage points (ie, 24% of the explained difference). The difference in two of the individual‐level enabling factors, usual source of care and health insurance, explains 0.4 and 0.5 percentage points of the difference (about 7% and 8% of the explained difference) in the outcome. The difference in the need characteristics at the individual level accounts for nearly half of the explained difference (2.8 percentage points) between Hispanics and NH whites (pain prevalence and health status explain a 2.4 and 0.4 percentage‐point difference, respectively). Moving Hispanics into the communities that have the same racial/ethnic composition as average NH whites will lead them to have more opioid prescriptions filled, thus closing the gap between Hispanics and NH whites by 1.9 percentage points (32% of the explained difference). Accounting for the difference in median household income at the community level increases the gap in the outcome by 0.4 percentage points.

3.4. Decomposition analysis by low‐ vs high‐equivalency opioids

NH whites are 1.0 percentage point (28 percent) more likely than Hispanics to have at least one low‐equivalency opioid prescription per year (Table 4). The difference attributable to the differences in means is 1.1 (fourth row on the first column), which is slightly higher than the actual difference, indicating that if Hispanics had the same observable characteristics as NH whites, the mean of the outcome would be slightly higher for Hispanics than for NH whites. The analysis by specific characteristics shows that if Hispanics had the same level of access to health insurance as NH whites, the difference in the outcome would decrease by 24.1% (2.3 percentage points). Likewise, if Hispanics had the same level of pain prevalence as NH whites, the outcome would be higher for Hispanics than for NH whites (note that another individual need factor, health status, is not significant). Contrary to our main findings, being born in the United States and the racial composition of the neighborhood do not explain the racial/ethnic difference in low‐equivalency opioid use.

TABLE 4.

Decomposition results (in percentage points), hispanics versus NH whites: outcome is any low vs. equivalent/high‐equivalency opioid prescriptions

Outcome	Any Low‐Equivalency (MME factor < 1) Opioid Rx	SE	Any Equivalent (MME factor = 1) or High‐Equivalency (MME factor > 1) Opioid Rx	SE
Mean among NH whites	4.400*	0.179	12.728*	0.307
Mean among Hispanics	3.438*	0.165	6.840*	0.288
Actual differences	0.961*	0.225	5.888*	0.417
Amount of difference attributable to differences in means	1.082*	0.335	5.803*	0.579
Amount of difference attributable to differences in means of specific variables
Individual characteristics
Predisposing variables
Age, sex, and marital status	−0.049	0.094	−0.224*	0.111
Born in the United States	0.171	0.212	1.516*	0.244
Attitude, health care system	−0.143*	0.049	−0.106	0.072
Enabling characteristics
Has a usual source of care	0.081	0.045	0.327*	0.074
Health Insurance	0.232*	0.109	0.206	0.126
Education, employment, family income	−0.164	0.144	−0.049	0.207
Need characteristics
Pain prevalence	1.006*	0.139	1.670*	0.194
Health status	0.184	0.141	0.434*	0.188
Community characteristics
Predisposing variables
Race composition	−0.317	0.249	2.334*	0.363
Region, rural/urban (MSA)	0.117	0.134	0.088	0.236
Enabling characteristics
Median income	−0.044	0.105	−0.375*	0.128
Health care resources	0.008	0.092	−0.017	0.146

N = 48 774. Estimates were weighted and adjusted for the complex survey design of MEPS.

Abbreviations: MME factor, Morphine Milligram Equivalent conversion factor; MSA, metropolitan statistical area; NH, non‐Hispanic; SE, standard errors.

^*Denotes significance at the 5% level.

Data Sources: 2013‐2016 Medical Expenditure Panel Survey, 2012‐2016 American Community Survey, 2015 Health Area Resources File.

The difference in the use of high‐equivalency/equivalent opioid prescriptions is starker. NH whites are 5.9 percentage points (86%) more likely than Hispanics to have at least one high‐equivalency/equivalent opioid prescription filled per year and almost all of the actual difference (5.8 percentage points) can be explained by the differences in the means of explanatory variables. The set of explanatory variables that explain the difference in the high‐equivalency/equivalent opioid prescriptions are very similar to the ones discussed in the main analysis. The differences in the three sets of explanatory variables (ie, the rate of those born in the United States, pain prevalence, and community‐level racial/ethnic composition) account for 1.5, 1.7, and 2.3 percentage points of the difference (25%, 28%, and 38% of the observed difference), respectively. We only performed the analysis for the difference between Hispanics and NH whites because the observed difference in any outpatient opioid prescription use between NH blacks and NH whites is smaller.

3.5. Robustness checks

First, we performed our main decomposition analysis using a logit model. ²⁵ Although not directly comparable, the logit model results are similar to our main results, with the exception that none of the specific characteristics explain the NH white‐NH black difference (Table S4). Second, we divide the study period into two spans (2013‐2014 and 2015‐2016) because the factors associated with the racial/ethnic differences could change over time, especially before and after the recent health care reforms. We find that the results are very similar between the two periods for the Hispanic vs. NH white decomposition analysis while the actual outcome difference is significant only in 2015‐2016 between NH blacks and NH whites (Table S5). The results are also robust to using different definitions of heavy opioid prescription use, including having four or more opioid prescriptions per year (to capture repeated uses) and having total annual MMEs greater than 700 (to capture high‐dose uses). ⁷ , ²⁶ We also excluded perceived physical and mental health from a model as a robustness check because reporting can differ across race/ethnicity. Finally, to confirm that the results are not solely driven by those treated for cancer, we excluded those who were ever diagnosed with cancer (information on whether they are currently treated for cancer or not is unavailable in the data for most years) and also added indicators for the types of cancer that they were ever diagnosed with. We found that the results of these robustness checks are similar to the main results of the analysis (Tables S6 and S7).

4. DISCUSSION

This study documents a large difference in the use of prescription opioids between NH white and Hispanic nonelderly adults during the period 2013‐2016 and shows that almost all the difference can be explained by observable characteristics. Three risk factors—pain prevalence (individual need factor), the rate of being United States born, and the racial/ethnic composition of the community (predisposing factors at the individual and community levels, respectively)—explain most of the difference. Other individual‐level predisposing and enabling factors, such as demographics, beliefs, access to care, and community‐level income, are significantly associated with the difference in opioid use but to a lesser extent. Other community characteristics, including region, rural/urban (a predisposing factor), and availability of health care sources (an enabling factor), were not significantly associated with the difference in opioid prescription use.

The difference in prescription opioid use between NH white and NH black nonelderly adults during the same period was relatively small. Very little of the difference is explained by the differences in observable characteristics, which is consistent with the previous research decomposing racial/ethnic differences in other prescription drugs. ¹⁸ , ²⁷ , ²⁸ The gap in prescription opioid utilization between NH black and NH white would narrow if NH black had the same characteristics as the NH white on some dimensions, while it would widen on other dimensions. Within the individual need domain, the difference in pain prevalence explains more than a half of the outcome difference, yet closing gaps in health status would widen the difference in the outcome. Likewise, less access to a usual source of care among NH blacks (individual enabling factor) significantly explains the outcome difference, yet if NH blacks had the same community median income (community enabling factor) as NH whites, the outcome difference would be wider. Lastly, within individual predisposing factors, the differences in the rate of United States born explains the outcome difference, while if NH blacks had the same demographic characteristics and belief as NH whites, the outcome difference would be larger.

We further find that the predictors of the Hispanic vs. NH white difference differ significantly between higher‐ and lower‐potency opioids. Unlike our overall results, the difference in lower‐potency opioid prescription use is explained mainly by the difference in pain prevalence and access to insurance. These findings are consistent with the literature on racial/ethnic differences in prescription drug use ¹⁸ , ²⁸ and may indicate underutilization of low‐potency prescription opioids among Hispanics due to less access to care. In contrast, our results on higher‐potency opioid prescription are unique and differ from an existing study that found no significant effects of community characteristics on the use of other prescription drugs. ²⁸ Lack of access to a usual source care among Hispanics explains the difference in higher‐potency prescription opioid use, yet its contribution is much smaller than the three main risk factors (ie, pain prevalence, community racial/ethnic composition, and rate of being United States born).

Our findings regarding the importance of foreign‐born status for Hispanics are consistent with prior research on “the immigrant paradox,” a well‐documented phenomenon wherein new immigrants fare better than their native counterparts on a variety of health measures, despite having disadvantaged socioeconomic status. ²⁹ The community/social context also plays an important role in explaining the difference in Hispanic‐NH white high‐potency prescription opioid use. Our finding that the percentage of Hispanics at the neighborhood level impacted individual use patterns for Hispanics beyond individual race/ethnicity is consistent with studies of the importance of social context for both health and health care outcomes for immigrant populations in the United States. Research on immigrant enclaves provides numerous examples of how social networks and community social support enhance employment, health care, and social outcomes. ³⁰

A limitation of this study is that, as in other decomposition analysis studies, we identify an association but not causality between the outcome and the risk factors. Another limitation of the decomposition analysis is that the source of the “unexplained” difference cannot be identified, which is relevant in interpreting our finding that the difference between NH white and NH black remain largely unexplained. We also note that our self‐reported pain measure can contain measurement errors that can systematically differ across race/ethnicity. We tried to mitigate this concern by controlling for an extensive list of health‐related, socioeconomic, and demographic characteristics. The outcome difference explained by the self‐reported pain measure is relatively smaller (about one‐third in magnitude) compared with the difference explained by pain‐related conditions. Lastly, our model focuses on patient utilization behavior and does not account for provider characteristics. Further studies investigating the mechanism behind the association identified in the current study, including the role of providers, could lead to more actionable insights.

Although the rate of opioid prescribing in the United States has decreased steadily from 2012 to 2018, it remains very high in some areas in the country. ³¹ As the policy makers continue to promote safer and more effective pain management, the baseline difference in the use of opioid prescriptions across race/ethnicity, individual and community characteristics identified in the current study could be informative in tailoring programs, such as clinician education on prescribing practices, patient family and care giver education, and collaborating efforts with communities.

Supporting information

Author matrix

Appendix S1

Moriya AS, Xu L. The complex relationships among race/ethnicity, social determinants, and opioid utilization. Health Serv Res. 2021;56:310–322. 10.1111/1475-6773.13619