Biopsychosocial factors associated with opioid misuse in a 2017-2018 United States national survey: A comprehensive multivariate model

Background Few studies have comprehensively and contextually examined the relationship of variables associated with opioid misuse. Our purpose was to fill a critical gap in comprehensive risk models of opioid misuse in the United States by identifying the most salient predictors. Methods A multivariate logistic regression was used on the 2017 and 2018 National Survey on Drug Use and Health, which included all 50 states and the District of Columbia of the United States. The sample included all noninstitutionalized civilian adults aged 18 and older (N=85,580; weighted N=248,008,986). The outcome of opioid misuse was based on reported prescription pain reliever and/or heroin dependence or abuse. Biopsychosocial predictors of opioid misuse in addition to sociodemographic characteristics and other substance dependence or abuse were examined in our comprehensive model. Biopsychosocial characteristics included socioecological and health indicators. Criminality was the socioecological indicator. Health indicators included self-reported health, private health insurance, psychological distress, and suicidality. Sociodemographic variables included age, sex/gender, race/ethnicity, sexual identity, education, residence, income, and employment status. Substance dependence or abuse included both licit and illicit substances (i.e., nicotine, alcohol, marijuana, cocaine, inhalants, methamphetamine, tranquilizers, stimulants, sedatives). Results The comprehensive model found that criminality (adjusted odds ratio [AOR]=2.58, 95% confidence interval [CI]=1.98-3.37, p<0.001), self-reported health (i.e., excellent compared to fair/poor [AOR=3.71, 95%CI=2.19-6.29, p<0.001], good [AOR=3.43, 95%CI=2.20-5.34, p<0.001], and very good [AOR=2.75, 95%CI=1.90-3.98, p<0.001]), no private health insurance (AOR=2.12, 95%CI=1.55-2.89, p<0.001), This study provides the most recent and comprehensive risk assessment of possible biopsychosocial characteristics indicative of opioid misuse. Findings provide the population-level risk factors to improve risk assessments and to tailor future interventions to stem and ameliorate the opioid epidemic. For instance, at-risk individuals had a history of criminality, serious psychological distress, suicidality, no private health insurance, and substance dependence or abuse. Individuals, however, are not variables representative of risk factors on an outcome to opioid misuse. At a population-level analysis, we must acknowledge that results of a person-centered approach such as this work only represent findings based on a population average. More specialized approaches, such as variable-centered ones, are necessary to study specific at-risk groups. Thus, these findings serve as a population-level risk profile using the most recent US nationally-representative data to inform epidemiological trends and possible large-scale interventions.

3 2.14, p=0.004), and other substance dependence or abuse were significant predictors of opioid misuse. Substances associated were nicotine (AOR=3.01, 95%CI=2. 30 Conclusions Biopsychosocial characteristics such as socioecological and health indicators, as well as other substance dependence or abuse were stronger predictors of opioid misuse than sociodemographic characteristics.

Background
Estimates indicate that up to 29% of persons misuse prescription pain relievers for chronic pain, 1 and between 8 to 12% develop a misuse disorder. 2,3 The United States (US) Department of Health and Human Services (2019) declared a public health emergency in 2017, although the first wave of the epidemic can be traced to the 1990s. 3 In 2016 alone, the record numbers of opioid misuse and overdose death provided a stark realization of how the epidemic has become a public health crisis. 2 For instance, opioid related deaths increased from 345% in 2001 to 2016. 4 Subsequently, between July 2016 and September 2017 deaths due to illicit opioids overdose deaths increased by 30% leading to emergency declaration in 45 states. 4 Projections revealed if current prevention and intervention strategies do not change by 2025, the rate of misuse and overdose death will rise by 61%. 5 In response to the epidemic, multiple federal, state, and local agencies have implemented various strategies to address the opioid crisis. Current interventions such as increasing availability of naloxone are projected to result in approximately a 4% reduction in opioid-related 4 deaths. 6 Similar interventions like reduced prescribing for pain patients and excess opioid management can increase life years and quality-adjusted life years, but overdose deaths would increase among those with opioid dependence due to a move from prescription opioids to heroin. 6 Overall, these strategies are found to have minimal impact preventing only 3.0-5.3% of overdose deaths. 5 Studies by Chen and colleagues 5 and Pitt and colleagues 6 have further revealed that current universal interventions are not enough to address the multidimensional and dynamic aspects of the opioid epidemic. Improving universal opioid prevention strategies to more tailored approaches has been suggested. 7 Non-Hispanic whites, for instance, have become a primary focus for multiple prevention programs and strategies as they have been found to misuse opioid at greater rates. However, multiple racial/ethnic groups have been found to be affected by opioid misuse and are at differential risk. 8− 10 Other racial/ethnic groups found to experience high disparities in misuse and related outcomes include American Indian/Alaska Natives 8 , Asians 11 , and Hispanics. 12 To ameliorate the effect of the opioid epidemic, we must identify the risk factors associated with the etiology of misuse to curb dependence and abuse. Secondly, it is crucial to understand biopsychosocial characteristics in the presence of multiple sociodemographic factors and other substance dependence or abuse that underpin the risk profiles of misuse at the population-level in order to stem overdose deaths.
Biopsychosocial characteristics for our research purposes include socioecological (e.g., criminality) and health factors (e.g., self-reported general health; mental health, suicidality; access to health services). Therefore, to understand what factors are contributing to the increasing opioid epidemic, we comprehensively examined the effects of biopsychosocial characteristics on opioid misuse using four domains: (1) 5 sociodemographic factors; (2) socioecological factors; (3) health factors; and (4) other substance dependence or abuse. We took this approach to determine the most salient risk factors for opioid misuse in a representative, noninstitutionalized US adult sample.
We hypothesized that sociodemographic factors, while crucial to the comprehensive risk model, would not be critical predictors when included with socioecological and health factors, or other substance dependence or abuse. The purpose of this study was to add to a critical gap in the literature to improve population-level prevention strategies by identifying the most salient predictors of opioid misuse.

Methods
We used multivariate logistic regression on the combined 2017 13 and 2018 14 National Survey on Drug Use and Health (NSDUH) to examine the relationship of biopsychosocial characteristics on opioid misuse; measured as opioid dependence or abuse.
Biopsychosocial characteristics, as well as sociodemographic and other substance dependence or abuse were tested independently in unadjusted models. Adjusted models were then built using a block entry method to test biopsychosocial characteristics on opioid misuse in the following order: (Model 1) sociodemographic indicators; (Model 2) socioecological indicator; (Model 3) health indicators; and (Model 4) other substance dependence or abuse. All variables were retained as controls and covariates in subsequent models. We accounted for the complex survey design of the NSDUH by the strata and clusters provided, as well as adjusting the analytical weights to account for two years. All year was also ascertained. Dependence and abuse in the past year for the following substances were also determined: marijuana, cocaine, hallucinogens, inhalants, methamphetamine, tranquilizers, stimulants (i.e., independent of methamphetamine), and sedatives. 18 Opioid misuse was characterized by dependence or abuse in past year of those that used prescription pain relievers and/or heroin.

Statistical Analysis
We performed descriptive analyses to detail the characteristics of NSDUH sample participants. We checked the data for normality of the residuals, homoscedasticity, multicollinearity, outliers and influence. After the data were found to be adequate for the logistic regression model, four weighted multivariate models were built using Stata survey 8 procedure. All models were weighted and accounted for clustering and stratification of the complex survey design. All findings are reported in odds ratios (ORs) or adjusted odds ratios (AORs) using a 95% confidence interval (CI) and p-value for significance criteria.

Sample Characteristics
The sample consisted of 85,580 individuals (weighted N = 248,008,986) over the age of  Table 1 for a detailed breakdown of the sample's characteristics.  Table 2 for a complete report of the sample's substance dependence and abuse profile.

Logistic Regression
Independent unadjusted models. All sociodemographic and biopsychosocial characteristics, as well as other substance dependence or abuse were tested independently in unadjusted models to examine the relationship of each characteristic on opioid misuse. All characteristics tested with exception of residence at some level were found to be a significant factor predictive of opioid misuse. See Table 3 for all associations.  Table 4 for more detail.

Discussion
Opioid misuse prevention strategies and programs must focus on multiple associated misuse factors in the context of the person and their environment to ameliorate the ongoing epidemic. Epidemics do not occur in a vacuum, as such we accounted for the biopsychosocial characteristics associated with opioid misuse in context of sociodemographic factors and substance use. Analyses revealed sociodemographic, socioecological, and health factors, as well as other substance dependence or abuse, were significant biopsychosocial risk factors for opioid misuse. Specifically, we found that socioecological indicators like criminality and health status factors, including serious psychological distress and suicidality, as well as private health insurance were significant risk characteristics.
Nicotine, alcohol, marijuana, cocaine, methamphetamine, tranquilizer, and stimulant substance dependence or abuse were also significant predictors of opioid misuse.
Sociodemographic factors have generally been identified as a definitive risk factors in opioid misuse, and overdose death. 8, 9, 19− 21 In the presence of biopsychosocial factors and other substance abuse we found that sociodemographic characteristics were no longer significant predictors but served as controls for our comprehensive opioid misuse model.  24 found that individuals with opioid misuse disorder who had a severe mental illness were at an increased risk of criminality and suicidality. The risk increased between those using only heroin, both heroin and prescription opioids, and all other substance use disorders, in that order.
Other substance dependence or abuse has been associated with opioid misuse based on varying risk factors. 19, 25− 28 In this study, we specifically found that nicotine, 25,29 marijuana, 25 cocaine, 28 methamphetamine, 30 tranquilizers, 31− 33 and stimulants 34 increase the probability toward opioid misuse. Although the present study revealed an increased association of opioid misuse with marijuana compared to non-marijuana users, the relationship in the literature has been mixed. A more recent review found that marijuana use may decrease the probability of opioid misuse. 35 Campbell et al. 35 further revealed that medical cannabis laws/use decreases opioid overdose deaths in states that allow marijuana use compared to states that do not have medical marijuana laws.
Alcohol has been another substance with mixed associations for opioid misuse. For instance, Fernandez et al. 36 reported that alcohol dependence or abuse was not associated with opioid misuse.
We found, however, in our comprehensive adjusted model that alcohol dependence or abuse was associated with a higher probability for opioid misuse, in line with the findings of Witkiewitz et al. 37 Overall, prevention strategies and prevention programs must focus on both the combined use of legal and illicit substances.
Our study took a comprehensive approach to understand how multiple biopsychosocial variables combine to predict opioid misuse. Individuals are influenced by a constellation of factors, and any research should account for this miscellany when considering causes, effects, and treatment.
Although comprehensive models can be cumbersome, they provide the ability to examine multiple risk factors in context to assemble profiles of misuse at a population level.

Limitations
To our knowledge, this is the first US population-level study to comprehensively address risk profiles of opioid misuse using the latest national survey data available. Like most surveys of this kind, there are limitations to the NSDUH. The most prominent limitation is the use of self-reported data. These data are subject to the individual participant's bias, truthfulness, recollection, and knowledge.
Second, although the data are nationally representative, it is cross-section and exclude some subsets of the population. The NSDUH targets only noninstitutionalized US citizens, so active-duty military members and institutionalized groups (e.g., prisoners, hospital patients, treatment center patients, and nursing home members) are excluded. Thus, if substance use differs between US noninstitutionalized and institutionalized groups by more than 3%, data may be problematic for the total US population. 18 Another issue that may have introduced bias is participant knowledge or lack thereof concerning opioids and other substances. 38 Moreover, heroin is a less commonly used opioid and there are issues in accounting for the true prevalence of this substance use. 38,39 Finally, the 20 opioid misuse data does not fully account for synthetic opioids like fentanyl.

Conclusion
This study provides the most recent and comprehensive risk assessment of possible biopsychosocial characteristics indicative of opioid misuse. Findings provide the population-level risk factors to improve risk assessments and to tailor future interventions to stem and ameliorate the opioid epidemic. For instance, at-risk individuals had a history of criminality, serious psychological distress, suicidality, no private health insurance, and substance dependence or abuse. Individuals, however, are not variables representative of risk factors on an outcome to opioid misuse. At a population-level analysis, we must acknowledge that results of a person-centered approach such as this work only represent findings based on a population average. More specialized approaches, such as variablecentered ones, are necessary to study specific at-risk groups. Thus, these findings serve as a population-level risk profile using the most recent US nationally-representative data to inform epidemiological trends and possible large-scale interventions.
Abbreviations AOR Adjusted odds ratio CBSA Core-based statistical areas OR Odds ratio

Ethics approval and consent to participate
The study received exemption from the Texas A&M University Institutional Review Board as no human participants were involved in this research.

Consent for publication
Not applicable.