Defining Prescription Drug Misuse: A Naturalistic Evaluation of National Survey on Drug Use and Health Data from 2012–2014 to 2015–2017

Abstract

Objectives:

Prescription drug misuse (PDM) is a significant public health problem. As research has evolved, the definitions of misuse have varied over time, yet the implications of this variability have not been systematically studied. The objective of this study was to leverage a change in the measurement of PDM in a large population survey to identify its impact on the prevalence and correlates of this behavior.

Methods:

Data from the National Survey on Drug Use and Health were compared prior to and following a change in the definition of PDM from one that restricted the source and motive for use to one that captured any misuse other than directed by a prescriber. Three-year cohorts were constructed, representing a restricted definition of PDM (2012–2014) and a broad definition of PDM (2015–2017).

Results:

Segmented logistic regression models indicated a significant increase in PDM prevalence for all three drug types examined (opioids, tranquilizers and sedatives). Although the magnitude of differences varied somewhat based on drug type, the broader definition was generally associated with older age, higher prevalence of health insurance and higher odds of misusing one’s own prescription. Some worsening of mental health indicators was observed, but results indicated few other clinical or substance use differences.

Conclusions:

Definitions of prescription drug misuse have a substantial impact on the prevalence of misuse and some impact on the characteristics of the population. Further research is needed to understand the optimal strategy for measuring this behavior, based on the scientific or public health question or interest.

Introduction

Several classes of psychotherapeutic prescription medications (opioid analgesics, tranquilizers, sedatives and stimulants) have reinforcing properties, which can result in their misuse (use in a manner inconsistent with the medically directed frequency, dose, or context). Approximately 16.5 million people in the US misused a prescription medication in 2019, making it the fourth most common type of substance used¹. Prescription drug misuse (PDM) is associated with risk for a broad continuum of potential harms, such as negative consequences for health, escalation to a substance use disorder, and overdose^2–4. Understanding of PDM is impeded by a lack of consensus on its definition and measurement, which may contribute to mixed findings across studies, obscure understanding of epidemiology, and hamper the development of interventions to reduce its consequences^5,6.

Definitions of PDM have varied substantially in their scope^5,7,8. The broadest definitions of PDM encompass any medication use at a frequency or dose greater than prescribed or for any reason other than the medical indication for which it was prescribed. More restrictive definitions can require specific sources (e.g., obtaining from a source other than a prescriber for a “legitimate” medical indication), motives (e.g., to get high), or levels of severity (e.g., substance use disorder symptoms).

Achieving greater consistency of assessment and consideration of the complexity of this behavior will require a better understanding of the implications of these variations in the definition of PDM. For example, the implications of variability in definitions for the scope of PDM (e.g., the number of people engaging in this behavior), its correlates and its consequences are unknown. An exemplar of this issue is the definition of PDM in the National Survey on Drug Use and Health (NSDUH), a large, annual survey administered by the Substance Use and Mental Health Services Administration (SAMHSA) that assesses trends in the use of substances across the US. A substantive change in the definition of PDM occurred in 2015 when SAMHSA expanded from a more restrictive definition (use of a prescription medication that was either not prescribed to that person or for a reason other than medically indicated) to a more expansive definition (use of a prescription medication in a way that was not directed by a prescriber). The survey change in the NSDUH provides the opportunity for a naturalistic comparison of the implications of restrictive and broad definitions of PDM.

In this exploratory study, we seek to identify the implications of a change from a restrictive to broad definitions of PDM in the NSDUH, with a focus on opioid analgesics, tranquilizers and sedatives. The objectives of this study are to (1) estimate the magnitude of change in the population prevalence estimate of past-year PDM with the introduction of a broader definition in 2015, (2) compare sociodemographic differences between cohorts who reported PDM using the restrictive (2012–2014 respondents) and broad (2015–2017 respondents) definitions, and (3) compare the clinical severity level in those who reported PDM using the restrictive and broad definitions. Due to the substantial differences across these medication classes, including their reinforcing properties and potential for misuse⁹, we examined each medication class separately.

As with any naturalistic study, observed changes may be attributable to other factors (e.g., changes in prescribing practices or perceptions of drug harms). To mitigate the potential confounding of other factors and population trends in the prevalence of PDM, we combined data into two 3-year periods (2012–2014 and 2015–2017), each encompassing one of the two definitions.

Methods

This secondary data analysis was preregistered on the Center for Open Science Open Science Framework (https://osf.io/m3pw2). We used publicly available data from the NSDUH, an annual survey of people aged 12 and older across the US, including civilian, non-institutionalized persons in all 50 states and the District of Columbia. The NSDUH uses a complex survey design that allows for estimation of state and national prevalence of substance use.¹⁰

Prior to 2015, the NSDUH survey defined PDM as taking a prescription drug that “was not prescribed for you” or taking it “only for the experience or feeling it caused”¹¹. Starting in 2015, the definition of PDM was broadened to: “use of the drug in any way a doctor did not direct you to use it, including: using it without a prescription of your own, using it in greater amounts, more often, or longer than you were told to take it, using it in any other way a doctor did not direct you to use it”¹¹. This modification provides the opportunity for a naturalistic test of the implications of two definitions on the observed prevalence of PDM. Respondents included in each of these survey years will be referred to as the “Restrictive Definition Cohort” (2012–2014) and “Broad Definition Cohort” (2015–2017). All participants from the six years of survey data included a total of 336,018 people, including 165,699 participants in the Restrictive Definition Cohort and 170,319 in the Broad Definition Cohort. The analysis was exempt from Institutional Review Board review as secondary analysis of publicly available de-identified data.

Measures

Sociodemographic variables of interest included gender, health insurance, age, and race, as well as variables representing gender by age and gender by race. Although we initially planned to control for urbanicity, we excluded this variable from analysis due to changes in the NSDUH survey in 2015, which SAMHSA specified would preclude comparison of this variable across years. To compare the clinical characteristics of the two PDM cohorts, we included several indicators of clinical severity, including other drug use, the most recent source of a misused prescription drug, overall physical health, psychiatric severity, functional impairment and past-year suicidal thinking, suicide plans, and attempted suicide. Although the NSDUH collects data on methamphetamine, hallucinogens, and inhalants, we excluded these drugs due to measurement changes in these variables in 2015 that may confound our analyses.

The NSDUH includes questions about the misuse of the following classes of prescription medications: opioid analgesics, stimulants, tranquilizers and sedatives. Due to a change in the categorization of illicit methamphetamine in 2015, we excluded prescription stimulants from our analysis. Diagnostic and Statistical Manual of Mental Disorders, 4^th Edition (DSM-IV)¹² diagnosis of abuse or dependence for each prescription drug was included to determine the conditional dependence for each prescription drug (i.e., the proportion of people with past-year misuse who also met criteria for a substance use disorder for that drug). We utilized an indicator of combined DSM-IV abuse and/or dependence, which we will refer to as “use disorder.”

Psychiatric severity was measured using the Kessler-6 Distress Scale¹³. Possible scores range from 0–24, with higher scores reflecting greater psychological distress. This measure has high internal consistency reliability and construct validity, as it accurately discriminates respondents with and without serious mental illness¹³.

The World Health Organization Disability Assessment Schedule (WHODAS)¹⁴ was used to measure functional impairment. Scores ranged from 0–24, with higher scores reflecting greater impairment. This measure has high internal consistency reliability and retest reliability, and strong concurrent validity in the general population¹⁴.

Data Analysis

We combined data from NSDUH survey years 2012–2017 and created two “cohorts” representing the years with the restrictive (2012–2014) and broad (2015–2017) definitions of PDM.

We utilized a segmented logistic regression to examine differences in prevalence between the Restrictive and Broad Definition Cohorts. This regression accounted for study year by assuming linear trends over the three survey years but with potentially different slopes between the two cohorts; a dummy-coded variable for cohort (reference: 2012–2014) was included to allow for discontinuity, or a shift, in the prevalence between the two cohorts. This discontinuity or shift parameter can be interpreted in terms of the difference in prevalence (expressed as log odds) in 2015 following the change in definition from restrictive to broad.

We conducted chi-square analyses comparing the sociodemographic characteristics of people who reported past-year misuse of each prescription drug type between the two cohorts. This subgroup analysis only included people who reported past-year misuse of each prescription medication class. We adjusted for multiplicity using the Sidak correction accounting for the number of tests conducted within each drug class (8 tests; adjusted alpha = .006).

To compare the clinical severity levels between the two cohorts, we conducted a series of regression analyses with cohort dummy-coded (reference: 2012–2014) among those with past-year misuse of each prescription drug type. Binary outcomes were analyzed using logistic regression, continuous outcomes using linear regression, and ordinal outcomes using ordinal regression. These analyses controlled for age, gender, and presence of health insurance coverage. We adjusted for multiplicity using the Sidak correction accounting for the number of tests conducted within each drug class (16 tests, adjusted alpha = .003).

All analyses accounted for the complex sampling design of the NSDUH, using published guidance from SAMSHA. The selection of weighting variables accounted for combining multiple years of NSDUH data. Analyses were conducted in SPSS Version 20.

Results

Prevalence of PDM Between the Restrictive (2012–2014) and Broad (2015–2017) Definition Cohorts

The estimated population prevalence of misuse of each prescription drug type over time is presented in Figure 1. The estimated prevalence increased for all three drug types from 2014 to 2015 (i.e., the year of the definition change).

Figure 1. Prevalene of Prescription Drug Use 2012–2017.

Black filled shapes denote prevelance rates of substance misuse each year of the NSDUH survey from 2012 to 2017. Squares indicate prevelance rates or predicted prevalence rates using the restrictive definition of misuse (i.e., 2012–14 definition). Triangles indicate prevelance rates using the broad definition (i.e., 2015–17 definition). The white filled squares are prevelance rates that have been predicted using parameters from the segmented logistic regressions for each substance respectively. These estimates represent the predicted prevelance rates for each substance for 2015 with the restrictive definition. These estimates represent the predicted 2015 prevelance rates as if there was no definition change. Error bars are two times the standard error to increase visibility. The standard errors for the sedative estimates were too small to be plotted.

Results of the segmented regression analyses indicated a significant increase in the odds of misuse for 2015–2017 respondents (Broad Definition Cohort) compared to 2012–2014 respondents (Restrictive Definition Cohort) for all three prescription drug types. Specifically, the relative increase in the odds of misuse in 2015 following the change in definition from restrictive to broad was 38% for opioids (adjusted odds ratio [aOR] = 1.38, 95% CI = 1.26, 1.51, p < .001), 36% for tranquilizers (aOR = 1.36, 95% CI = 1.18, 1.55, p < .001), and 89% for sedatives (aOR = 1.89, 95% CI = 1.30, 2.74, p < .001). Specifically, the estimated weighted prevalence prevalence increased from 3.9% to 4.7% for opioids, 1.9% to 2.3% for tranquilizers and 0.3% to 0.6% for sedatives. These results suggest that after controlling for yearly trends in the prevalence of each drug type, the prevalence was significantly higher in 2015 following the change from a restrictive to broad definition of misuse.

Sociodemographic Differences Between the Restrictive (2012–2014) and Broad (2015–2017) Definition Cohorts

Table 1 provides an overview of the cohort differences in sociodemographic composition. For all three substances, the proportion of people with health insurance was greater in the Broad compared to the Restrictive Definition Cohort. Differences in age across cohorts were observed for opioid analgesics and sedatives, characterized generally by older age among the Broad Definition Cohort.

Table 1:

Sociodemographic Characteristics of People Who Report Past-Year Prescription Drug Misuse in the 2012–2014 (Restrictive Definition) and 2015–2017 (Broad Definition) Cohorts (weighted, unadjusted estimates; % [SE])

	Past-Year Opioid Misuse			Past-Year Tranquilizer Misuse			Past-Year Sedative Misuse
	2012–2014	2015–2017	p	2012–2014	2015–2017	p	2012–2014	2015–2017	p
Gender			0.77			0.53			0.56
Female	46.0% (1.0%)	45.6% (0.8%)		50.5% (1.3%)	49.5% (0.9%)		54.9% (4.1%)	57.5% (2.2%)
Male	54.0% (1.0%)	54.4% (0.8%)		49.5% (1.3%)	50.5% (0.9%)		45.1% (4.1%)	42.5% (2.2%)
Health Insurance			<.001			<.001			<.001
Yes	76.9% (0.8%)	85.0% (0.5%)		75.7% (1.1%)	84.3% (0.7%)		75.4% (3.3%)	90.3% (1.3%)
No	23.1% (0.8%)	15.0% (0.5%)		24.3% (1.1%)	15.7% (0.7%)		24.6% (3.3%)	9.7% (1.3%)
Age			<.001			0.02			0.001
12–17 years old	10.7% (0.3%)	7.6% (0.2%)		7.1% (0.4%)	7.2% (0.3%)		13.0% (1.4%)	5.8% (0.6%)
18–25 years old	27.2% (0.6%)	22.5% (0.6%)		28.6% (0.8%)	30.6% (0.9%)		19.5% (1.8%)	16.4% (1.0%)
26–34 years old	22.9% (0.9%)	23.0% (0.6%)		26.6% (1.3%)	22.1% (0.8%)		23.1% (3.2%)	20.7% (1.7%)
35–49 years old	21.8% (0.8%)	23.9% (0.6%)		20.4% (1.0%)	19.5% (0.7%)		23.8% (2.9%)	20.8% (2.0%)
50+ years old	17.4% (1.0%)	23.0% (0.9%)		17.3% (1.3%)	20.6% (1.2%)		20.6% (3.5%)	36.4% (3.0%)
Race			0.51			0.43			0.001
White (Non-Hispanic)	65.3% (0.9%)	66.6% (0.9%)		74.7% (1.1%)	73.8% (1.0%)		67.7% (3.7%)	77.7% (1.7%)
Black/African American (Non-Hispanic)	12.1% (0.7%)	10.8% (0.5%)		7.4% (0.5%)	7.6% (0.5%)		8.0% (1.5%)	6.1% (1.1%)
Asian (Non-Hispanic)	2.4% (0.3%)	2.3% (0.3%)		1.7% (0.3%)	1.7% (0.4%)		0.7% (0.4%)	1.1% (0.4%)
Native American/Alaska Native (Non-Hispanic)	0.8% (0.1%)	0.7% (0.1%)		0.6% (0.1%)	0.4% (0.1%)		1.3% (0.5%)	0.2% (0.1%)
Native Hawaiian/Other Pacific Islander (Non-Hispanic)	0.4% (0.1%)	0.4% (0.1%)		0.1% (0.0%)	0.2% (0.1%)		0.5% (0.3%)	0.1% (0.1%)
More Than 1 Race (Non-Hispanic)	2.3% (0.2%)	2.8% (0.2%)		1.9% (0.3%)	2.8% (0.3%)		1.3% (0.4%)	3.6% (0.8%)
Hispanic	16.6% (0.7%)	16.5% (0.6%)		13.5% (1.0%)	13.5% (0.7%)		20.5% (3.7%)	11.1% (1.1%)
Gender by Age			<.001			0.22			<.001
Males, 12–17	5.1% (0.2%)	3.6% (0.2%)		3.1% (0.2%)	3.4% (0.2%)		5.2% (0.9%)	2.7% (0.5%)
Females, 12–17	5.6% (0.2%)	4.0% (0.2%)		4.0% (0.3%)	3.8% (0.3%)		7.8% (1.2%)	3.2% (0.4%)
Males, 18–25	15.3% (0.5%)	12.3% (0.5%)		15.7% (0.6%)	16.2% (0.7%)		8.0% (1.2%)	7.9% (0.8%)
Females, 18–25	11.8% (0.4%)	10.2% (0.3%)		12.9% (0.5%)	14.5% (0.4%)		11.5% (1.3%)	8.4% (0.6%)
Other	62.1% (0.7%)	69.9% (0.7%)		64.3% (1.0%)	62.2% (0.9%)		67.5% (2.4%)	77.8% (1.3%)
Gender by Race			0.51			0.69			0.05
Male, White (Non-Hispanic)	36.0% (0.8%)	36.4% (0.8%)		36.4% (1.2%)	36.8% (1.1%)		30.3% (3.2%)	32.1% (2.1%)
Female, White (Non-Hispanic)	29.4% (0.8%)	30.3% (0.7%)		38.4% (1.3%)	37.0% (1.0%)		37.4% (3.5%)	45.6% (2.3%)
Male, Black/African American (Non-Hispanic)	6.3% (0.5%)	5.8% (0.3%)		3.8% (0.4%)	4.4% (0.3%)		4.7% (1.5%)	3.5% (0.8%)
Female, Black/African American (Non-Hispanic)	5.8% (0.4%)	5.0% (0.3%)		3.6% (0.4%)	3.2% (0.3%)		3.2% (0.7%)	2.6% (0.8%)
Male, Hispanic	8.4% (0.6%)	9.1% (0.5%)		6.8% (0.6%)	7.1% (0.5%)		9.4% (3.3%)	4.5% (0.8%)
Female, Hispanic	8.2% (0.6%)	7.3% (0.4%)		6.7% (0.8%)	6.4% (0.4%)		11.2% (2.4%)	6.6% (1.0%)
Male or Female, Other Race	5.9% (0.4%)	6.1% (0.4%)		4.3% (0.4%)	5.1% (0.5%)		3.8% (0.8%)	5.1% (0.9%)

Note. PDM = prescription drug misuse

Otherwise, models indicated few sociodemographic differences between cohorts. The only drug for which model indicated differences in race was sedatives, for which there was an increase in the proportion of people who identified as non-Hispanic White from the Restrictive to the Broad Definition Cohort.

Clinical Severity Differences Between the Restrictive (2012–2014) and Broad (2015–2017) Definition Cohorts

Table 2 provides unadjusted population estimates of clinical variables of interest for each cohort. The findings reported below include results from adjusted analyses controlling for age, gender, and health insurance status and were conducted in the subgroup of people who reported past-year misuse for each drug type. Results of these analyses are available in Table 3.

Table 2.

Clinical Characteristics of People Who Report Past-Year Prescription Drug Misuse in the 2012–2014 (Restrictive Definition) and 2015–2017 (Broad Definition) Cohorts (weighted, unadjusted estimates; % [SE])

	Past-Year Opioid Misuse		Past-Year Tranquilizer Misuse		Past-Year Sedative Misuse
	2012–2014	2015–2017	2012–2014	2015–2017	2012–2014	2015–2017
Prescription Drug Misuse
DSM-IV abuse and/or dependence for prescription drug	17.3% (0.7%)	16.1% (0.5%)	8.9% (0.6%)	11.4% (0.6%)	18.5% (2.7%)	11.9% (1.3%)
Past-Year Misuse of Other Prescription Drugs
Prescription opioid misuse	n/a	n/a	56.4% (1.2%)	46.5% (1.0%)	48.5% (3.5%)	45.7% (1.9%)
Tranquilizer misuse	27.7% (0.8%)	24.4% (0.7%)	n/a	n/a	43.2% (3.0%)	39.5% (2.5%)
Sedative misuse	2.8% (0.3%)	5.9% (0.3%)	5.0% (0.5%)	9.7% (0.7%)	n/a	n/a
Past-Year Use of Other Drugs
Heroin	4.9% (0.3%)	5.2% (0.4%)	7.0% (0.5%)	5.6% (0.4%)	7.4% (1.6%)	5.7% (0.9%)
Cocaine and/or crack cocaine	3.0% (0.3%)	3.1% (0.2%)	4.0% (0.5%)	4.2% (0.4%)	3.6% (1.2%)	4.5% (1.0%)
Cannabis	50.5% (0.9%)	47.9% (0.8%)	61.7% (1.3%)	60.9% (1.0%)	41.9% (3.2%)	43.8% (2.2%)
Alcohol	84.5% (0.7%)	82% (0.7%)	91.0% (0.9%)	87.7% (0.8%)	83.2% (3.3%)	81.3% (2.1%)
Tobacco	63.2% (0.9%)	58.5% (0.8%)	67.8% (1.3%)	67.4% (0.9%)	56.6% (3.6%)	54.2% (2.6%)
Mental Health
Past-year suicidal thinking	12.5% (0.7%)	14.9% (0.5%)	14.9% (1.2%)	17.9% (1.0%)	16.5% (2.3%)	17.5% (1.5%)
Suicide plan	4.8% (0.4%)	5.6% (0.3%)	5.8% (0.6%)	7.1% (0.6%)	5.2% (1.1%)	7.4% (1.0%)
Attempted suicide	2.0% (0.2%)	2.9% (0.2%)	2.4% (0.3%)	3.8% (0.4%)	3.8% (1.1%)	4.3% (0.8%)
Kessler 6, worst value in past year	M=8.69 (SE=0.152)	M=9.00 (SE=0.106)	M=10.02 (SE=0.169)	M=10.23 (SE=0.164)	M=10.38 (SE=0.649)	M=9.82 (SE=0.268)
Other Health and Functioning
Excellent	15.2% (0.6%)	15.2% (0.6%)	15.2% (0.8%)	15.9% (0.9%)	12.4% (1.9%)	18.1% (2.0%)
Very Good	37.6% (0.8%)	35.1% (1.0%)	39.3% (1.2%)	36.3% (1.0%)	37.9% (3.1%)	37.2% (2.3%)
Good	31.7% (0.9%)	32.1% (0.8%)	30.0% (1.1%)	32.0% (0.9%)	26.4% (3.2%)	26.7% (2.3%)
Fair	15.5% (0.7%)	17.6% (0.8%)	15.4% (1.2%)	15.8% (0.8%)	23.2% (3.4%)	17.9% (1.8%)
Functional Impairment
WHO Disability Assessment Schedule	M=6.48 (SE=0.153)	M=7.02 (SE=0.121)	M=7.93 (SE=0.176)	M=8.17 (SE=0.190)	M=8.15 (SE=0.599)	M=8.64 (SE=0.310)
Source of Misused Drug a
One doctor	21.7% (0.8%)	34.8% (0.9%)	12.9% (1.1%)	20.2% (0.9%)	19.4% (3.0%)	32.2% (2.2%)
More than one doctor	2.8% (0.4%)	1.5% (0.3%)	0.8% (0.2%)	0.8% (0.2%)	N/A*	N/A*
Stole from doctor, hospital or clinic	0.5% (0.2%)	0.7% (0.1%)	0.3% (0.1%)	0.4% (0.2%)	N/A*	N/A*
Got from a friend or relative for free	51.4% (1.0%)	40% (0.9%)	60.7% (1.3%)	50.2% (1.0%)	53.9% (3.9%)	50.4% (2.1%)
Bought from a friend or relative	10.5% (0.5%)	9.7% (0.5%)	12.7% (0.7%)	12.9% (0.6%)	9.5% (2.1%)	5.7% (1.1%)
Took from a friend or relative	4.5% (0.4%)	3.5% (0.3%)	3.9% (0.5%)	3.3% (0.4%)	3.7% (1.2%)	4.6% (0.8%)
Bought from a drug dealer or stranger	4.6% (0.3%)	5.6% (0.3%)	5.2% (0.4%)	8.5% (0.6%)	3.3% (1.0%)	2.2% (0.6%)
Some other source	3.9% (0.3%)	4.3% (0.3%)	3.5% (0.6%)	3.8% (0.4%)	10.2% (3.1%)	4.9% (1.0%)

Note: PDM = prescription drug misuse. Bolded values are significantly different at adjusted alpha of .003;

^a.model effect of cohort significant at adjusted alpha of .003 for opioids and tranquilizers, see Table 3 for ORs for each source;

^*these motives were not assessed for this drug type.

Table 3.

Results of Adjusted Regression Models Examining Association between Cohort and Clinical Variables

	Past-Year Opioid Misuse	Past-Year Tranquilizer Misuse	Past-Year Sedative Misuse
	OR (95% CI) or Est (SE)	OR (95%) or Est (SE)	OR (95%) or Est (SE)
Prescription Drug Misuse
DSM-IV Diagnosis of abuse and/or dependence for prescription drug	0.92 (0.81, 1.06)	1.35 (1.14, 1.61)	.64 (.41, .98)
Presence of past-year misuse of other prescription drugs
Prescription Opioid Misuse	N/A	.69 (.61, .79)	1.10 (.79, 1.53)
Tranquilizer Misuse	.90 (.81, 1.00)	N/A	1.02 (.74, 1.41)
Sedative Misuse	2.20 (1.70, 2.84)	2.03 (1.58, 2.61)	N/A
Presence of past-year misuse of other drugs
Heroin	1.19 (.97, 1.46)	.85 (.68, 1.06)	.96 (.51, 1.81)
Cocaine and/or crack cocaine	1.05 (.80, 1.37)	1.13 (.80, 1.60)	1.86 (.80, 4.16)
Cannabis	1.03 (.93, 1.14)	1.03 (.89, 1.20)	1.60 (1.11, 2.29)
Alcohol	.88 (.76, 1.02)	.73 (.55, .96)	1.12 (.64, 1.95)
Tobacco	.90 (.81, 1.01)	1.07 (.91, 1.25)	1.26 (.90, 1.78)
Mental Health
Past-year suicidal thinking	1.31 (1.12, 1.54)	1.30 (1.02, 1.66)	1.36 (.91, 2.03)
Suicidal Plans	1.29 (1.04, 1.61)	1.33 (1.03, 1.73)	1.81 (1.01, 3.23)
Attempted Suicide	1.66 (1.25, 2.2)	1.63 (1.18, 2.27)	1.44 (.69, 3.02)
Kessler 6, worst value in past year	0.53 (0.18)	.38 (.24)	.09 (.62)
Other Health and Functioning
Health	0.04 (.05)	.06 (.07)	−0.34 (.17)
WHO Disability Assessment Schedule	0.63 (.20)	.36 (.26)	.80 (.65)
Source of Misused Drug (Ref: friend/relative for free)
One doctor	1.93 (1.69, 2.20)	1.81 (1.43, 2.29)	1.42 (0.89, 2.26)
More than one doctor	0.64 (0.42, 0.96)	1.11 (0.53, 2.32)	n/a
Stole from doctor, hospital or clinic	1.72 (0.79, 3.76)	2.08 (0.58, 7.43)	n/a
Bought from a friend or relative	1.26 (1.07, 1.48)	1.24 (1.03, 1.48)	0.66 (0.35, 1.26)
Took from a friend or relative	1.01 (0.76, 1.35)	0.99 (0.70, 1.38)	1.58 (0.78, 3.18)
Bought from a drug dealer or stranger	1.75 (1.44, 2.14)	2.05 (1.61, 2.61)	0.87 (0.37, 2.06)
Some other source	1.46 (1.14, 1.87)	1.32 (0.83, 2.09)	0.46 (.18, 1.16)

There were no significant differences between cohorts in conditional substance use disorder for people with past-year opioid analgesic or sedative misuse. In contrast, people who misused tranquilizers in the past year were more likely to meet criteria for tranquilizer use disorder in the Broad Definition Cohort (aOR = 1.35, 95% CI = 1.14, 1.61, p = .001).

Sedative misuse was higher in the Broad Definition Cohort among people with past-year opioid analgesic misuse (aOR = 2.20, 95% CI = 1.70, 2.84, p < .001) and those with past-year tranquilizer misuse (aOR = 2.03, 95% CI = 1.58, 2.61, p < .001), compared to the Restrictive Definition Cohort. Past-year opioid misuse was lower in the Broad Definition Cohort among people misusing tranquilizers (aOR = 0.69, 95% CI = 0.61, 0.79, p < .001).

Other markers of substance use did not indicate differences in polysubstance use; specifically, there were no significant differences across cohorts in past-year use of alcohol, tobacco, cocaine, cannabis, or heroin.

The proportion of respondents whose source of misused prescription drug was a prescription from a doctor increased from the Restrictive to the Broad Definition Cohort for both opioid analgesics and tranquilizers. Although the estimated percentage of people obtaining sedatives from one doctor was 19.4% in the Restrictive Definition Cohort and 32.2% in the Broad Definition Cohort, this change was not statistically significant.

Among people who reported past-year opioid analgesic misuse, mental health indicators were worse in the Broad Definition Cohort, including higher odds of reporting past-year serious thoughts of suicide (aOR = 1.31, 95% CI = 1.12, 1.54, p = .001) and suicide attempt (aOR = 1.66, 95% CI = 1.25, 2.19, p = .001). Similarly, this Broad Definition Cohort reported significantly worse psychiatric severity overall on the Kessler-6 (Est. = 0.53, SEest = .18, t = 3.05, p = .003).

Similarly, among people with past-year tranquilizer misuse, mental health indicators trended more severe in the Broad Definition Cohort than the Restrictive Definition Cohort; however, only the suicide attempt outcome met the pre-specified corrected significance level (aOR = 1.63, 95% CI = 1.18, 2.27, p = .004).

There were no significant differences in mental health indicators in the sedative group between cohorts.

There were no significant differences in self-reported overall health status between cohorts for any PDM group. Likewise, there was no significant difference in disability between cohorts for either the tranquilizer or sedative misuse groups. In contrast, for opioid analgesics, the Broad Definition Cohort reported significantly higher disability than the Restrictive Definition Cohort (Est. = 0.63, SEest = .20, t = 3.17, p = .002).

Discussion

We found that a change in the definition of PDM in the National Survey on Drug Use and Health from one that incorporated source and motive to one that that captured any use outside of that prescribed was associated with a significant increase in the estimated prevalence of misuse. This increase was significant across all three prescription drug classes examined, despite different overall trends in prevalence of these drugs over time (e.g., steep decreases in opioid analgesic misuse, relatively stable prevalence estimates of sedative misuse).

Consistent with the expansion of PDM to include use of a prescribed medication in any way not directed by a doctor, the Broad Definition Cohort was more likely to report one doctor as their most recent source of misused medication. Specifically, 20.2% to 34.8% of those reporting PDM in the Broad Definition Cohort reported one doctor as their most recent source of misused prescription medication, compared with 12.9% to 21.7% in the Restrictive Definition Cohort. In general, the Broad Definition Cohort was also older and more likely to have health insurance, both of which are associated with greater exposure to these medications via prescription^15–19. Notably, misuse of one’s own prescription is not necessarily indicative of a low-risk profile. Prior reports of NSDUH data have indicated that individuals who misuse their own prescription exhibited either comparable or higher odds of conditional dependence on that medication compared to those whose source is a friend/relative for free and comparable odds compared to those whose source is purchasing the medication illicitly^20,21.

Greater levels of psychiatric distress among those in the Broad Definition Cohort with opioid and tranquilizer misuse, compared to the Restrictive Definition Cohort, may indicate the broader definition of misuse is more likely to capture individuals misusing prescription medications to cope with inadequately managed mental health symptoms. There were few differences between the cohorts before and after the misuse definition change on general health and disability indicators. The one exception was the opioid analgesic group, for which disability was significantly worse in the Broad Definition Cohort. Of note, the WHODAS may be capturing pain-related disability, which may similarly suggest that the broader definition captured people seeking to cope with inadequately managed pain. The NSDUH only added questions on PDM motives in 2015, precluding the ability to test whether reasons for misuse shifted with the definition change. Accordingly, these interpretations are speculative and will require empirical testing.

The only difference in other substance use between cohorts was an increased proportion of sedative misuse with the broader definition among those with opioid and tranquilizer misuse (which is consistent with the large increase in sedative misuse following the definition change). The prevalence of other substance use did not shift with the broadened definition of misuse. Nonetheless, it is of note that other substance use was highly prevalent in both cohorts, reflecting the high prevalence of substance use among people with PDM. Our results did not support a reduction in substance use or other health severity, suggesting that a broader definition continues to capture a comparable proportion of people who are currently experiencing, or at risk for, poor outcomes. Nonetheless, it is also possible that the methods used in the present analysis are not sensitive to subtle shifts in participant characteristics, as it is not a direct test of individuals who are captured by varying definitions.

Taking all findings together, the broadened definition appeared to result in a larger group of respondents who reported misuse, with some shifts in demographic characteristics and few clinical differences. Nonetheless, there is variability within this broad group based on misuse characteristics (e.g., source, motives, route of administration), mental and physical health symptoms, polysubstance use, and risk behaviors for overdose and infectious disease (e.g., injection drug use). Further characterization of this heterogeneity is critical, as different subgroups likely necessitate targeted interventions to reduce harms associated with PDM⁵. For example, subgroups characterized by frequent misuse of their own prescriptions to cope with negative affective and somatic states might require adjunctive psychotherapy to manage these symptoms. In contrast, those characterized by frequent misuse without a prescription, polysubstance use, and high-risk behaviors for overdose and infectious disease (e.g., co-ingestion, injection drug use) might require harm reduction interventions (e.g., drug checking devices, receipt of naloxone) and general substance use treatment; and, subgroups characterized by infrequent misuse and low-risk behaviors may require limited or less intensive (e.g., education and monitoring) intervention focused on mitigating risk for worsening of symptoms and escalation of use. Person-centered approaches, such as latent variable mixture models²² could be used in future studies to characterize this heterogeneity among individuals who fall under the broad umbrella of those with PDM.

There are several limitations to this analysis. First, this was a secondary analysis of an existing data source. This study should not be considered a direct test of individuals captured by varying definitions of misuse. Rather, this naturalistic study leveraged a definition change to understand the implications of different definitions of PDM. Due to the naturalistic design, we cannot rule out the possibility that observed differences reflected changes in PDM unrelated to the definition change (e.g., changes in prevalence overall or in certain sociodemographic subgroups, changes in opioid prescribing following Centers for Disease Control guideline updates in 2016). Nonetheless, this exploratory study provides data that can be used to inform future direct tests of this question.

In addition, the NSDUH underwent several changes in 2015 that limited the variables that could be included in our analyses. Medications included in the prescription stimulant class changed substantially; therefore, we could not examine the impact of the misuse definition change on those with stimulant misuse. Several factors that might provide more insight into the characteristics of those with misuse and how they vary across definitions, including the frequency of misuse and motives for misuse, were also modified or added between cohorts. The NSDUH did not explicitly ask about some forms of misuse used in other studies, such as using a prescription drug with another substance to produce a psychoactive effect or using via a non-prescribed route of administration⁵. Finally, the sampling strategy of the NSDUH has several limitations, including the absence of active duty military, incarcerated people and people with housing instability. Accordingly, there may be underrepresentation of certain behaviors (e.g., heroin use)²³, and the generalizability to subgroups not captured with this sampling strategy cannot be assumed.

Conclusions

The broadened definition of PDM in the 2015 NSDUH survey was associated with increased prevalence of misuse for all three prescription drug classes, including opioid analgesics, tranquilizers, and sedatives. Although our observational study cannot establish causality, it is of note that this effect was observed controlling for yearly trends and was consistently observed across all drug types. This definition change appears to capture a broader group of people, particularly increasing representation of people who may be exposed to these medications via prescription.