Pain as a Predictor of Opioid Use Disorder in a Nationally Representative Sample

The 2001–2002 NESARC (wave 1), and the 2004–2005 follow-up (wave 2), is a nationally representative survey of the noninstitutionalized adult U.S. population conducted by the U.S. Census Bureau under the direction of the National Institute on Alcoholism and Alcohol Abuse, as described elsewhere (12–14). The wave 1 response rate was 81%. Excluding ineligible respondents (e.g., deceased), the wave 2 response rate was 86.7%, resulting in a cumulative response rate of 70.2% (N=34,653). Wave 2 NESARC weights include a component that adjusts for nonresponse, demographic factors, and psychiatric diagnoses to ensure that the wave 2 sample approximated the target population, that is, the original sample minus attrition between the two waves (12).

All diagnoses for waves 1 and 2 were made using the Alcohol Use Disorder and Associated Disabilities Interview Schedule–DSM-IV version (AUDADIS-IV) (15). Consistent with previous reports, nonmedical use of a prescription opioid was defined as using a prescription drug “without a prescription, in greater amounts, more often, or longer than prescribed, or for a reason other than a doctor said you should use them” during the 12 months preceding the interview. More than 30 symptom items are used by the AUDADIS-IV to define 12-month prescription opioid use disorders (abuse or dependence) according to DSM-IV criteria. The NESARC also collected information for other substance use disorders (nicotine dependence, alcohol use disorders, and drug use disorders, including other prescription drug use disorders). The reliability of the AUDADIS-IV prescription opioid use questions (kappa=0.66) and that of associated substance use disorder diagnoses (kappa values, 0.53–0.84) are well documented in both clinical (16) and general population (17) samples. The prescription opioid use disorder diagnosis also has established concurrent and predictive validity related to impairment and treatment seeking (18).

Mood disorders included major depressive disorder, dysthymia, bipolar I disorder, and bipolar II disorder. Anxiety disorders included panic disorder, social anxiety disorder, specific phobia, and generalized anxiety disorder. AUDADIS-IV methods to diagnose these disorders have been described in detail elsewhere (19). Test-retest reliabilities for AUDADIS-IV mood, anxiety, and personality disorders in the general population and in clinical settings were fair to good. Convergent validity was good to excellent for all mood, anxiety, and personality disorder diagnoses, showing good agreement with psychiatrist reappraisals (kappa values, 0.64–0.68) (20). As in our previous work (21), family histories of alcohol use disorders, drug use disorders, antisocial personality disorder, and depression were sought for first-degree relatives and were also assessed with the AUDADIS-IV, using readily observable manifestations of the disorders. The test-retest reliability for AUDADIS family history variables is very good to excellent (14).

Consistent with previous national studies (4), pain was assessed using the pain item of the Medical Outcomes Study 12-item Short-Form Health Survey (SF-12) (22), a valid measure that is commonly used in population surveys and is sensitive to change (23, 24). The pain item measures the amount to which pain interfered with daily activities during the past month, on a 5-point scale (not at all, a little bit, moderately, quite a bit, and extremely) (25). The pain measure was collapsed into two levels to indicate whether pain was associated with no or little interference (“no pain”) or moderate to extreme interference (“pain”) (2). In preliminary analyses, pain interference was associated with lower scores on each of the other 11 items of the SF-12 (all p values, <0.001) and increased the likelihood of all medical conditions assessed in the NESARC (odds ratios, 2.8–7.6; all p values, <0.001). This confirmed previous analyses indicating that greater interference of the SF-12 pain item was associated with higher health care expenditures, more missed days of work, lower productivity (4), and greater probability and frequency of nonmedical use of prescription drugs, even after adjusting for multiple comorbidities (9).

Wave 1 descriptive demographic and clinical characteristics were compared between individuals with and without pain and between individuals with and without a prescription opioid use disorder. Group differences were evaluated with chi-square or t tests.

In separate analyses, odds ratios and adjusted odds ratios, controlling for age, sex, other substance use disorders, and mood or anxiety disorders at wave 1, were then used to assess the strength of associations at wave 1 between pain severity and prescription opioid use disorder.

We built our model for the prospective relationship between pain and prescription opioid use disorder in two steps. First, a series of logistic regression models were used to examine whether pain at wave 1 was associated with the prevalence and incidence of prescription opioid use disorders at wave 2 and whether prescription drug use disorder at wave 1 was associated with the prevalence and incidence of pain at wave 2. Incidence was defined as the number of new cases reported between wave 1 and wave 2. Exploratory analyses examined whether there were any interactions between time of pain assessment (wave 1 versus wave 2) and likelihood of prescription drug use disorders at wave 2. All logistic regression models were adjusted for age, sex, other substance use disorders, and mood or anxiety disorders at wave 1, which have been shown to be associated both with pain (2–4) and with prescription opioid use disorders (26, 27). In addition, the prevalence models adjusted for current prescription opioid use disorder or level of pain at wave 1, as appropriate.

In the second step, because of the bidirectional relationship between pain and prescription opioid use disorders, we used structural equation models to model the interdependency between pain and prescription opioid use disorders at waves 1 and 2. Jointly modeling pain and prescription use disorders avoids simultaneous equation bias (28), which can result when they are modeled in separate, unlinked regressions. Because the prevalence of pain and prescription drug use disorders, along with several covariates in the model, differs by sex (3, 29), we used multiple group structural equation model analyses to examine whether the model was invariant by sex, comparing a model with all parameters free to differ across sex to one in which they were all the same. Invariance was determined by finding this chi-square difference test nonsignificant (p>0.05). Additional structural equation models were examined substituting specific criteria (tolerance and withdrawal) or an indicator of using prescription opioids longer or at greater doses than prescribed for prescription opioid use disorder at wave 2. Logistic regression models were fitted and adjusted odds ratios estimated with 95% confidence intervals using SUDAAN. The structural equation models were fitted with standardized probit regression coefficients and p values in Mplus, using weighted least squares, an estimator that is robust for nonnormality. Because the structural equation models being estimated were fully saturated (i.e., all cross-sectional and longitudinal associations between pain and prescription opioid use disorder at both waves were freely modeled such that the degrees of freedom was zero), standard structural equation model fit statistics provide no additional information. Both SUDAAN and Mplus take into account the complex design features of the NESARC.

At wave 1, the prevalence of pain was 18.76% (N=6,943) and the prevalence of prescription opioid use disorders was 0.33% (N=104). Individuals with prescription opioid use disorders were younger than those without, were more likely to be male, and were more likely to have another current substance use disorder, a mood or anxiety disorder, and a family history of alcohol use disorders, drug use disorders, depression, and antisocial personality disorder. Individuals with pain were more likely to be older than those without and were more likely to be female, but they were also more likely than those without pain to have a current substance use disorder, a mood or anxiety disorder, and a family history of alcohol use disorders, drug use disorders, depression, and antisocial personality disorder (Table 1).

Pain was reported by 38.27% of individuals with prescription opioid use disorders and 18.69% of those without (odds ratio=2.70, 95% CI=1.66–4.39). After adjustment for the background demographic and clinical characteristics, a strong association persisted between prescription opioid use disorder and pain (adjusted odds ratio=2.38, 95% CI=1.51–3.76).

Logistic regression models indicated that pain at baseline was associated with both prevalence (odds ratio=2.05, 95% CI=1.33–3.18) and incidence (odds ratio=2.15, 95% CI=1.37–3.37) of prescription opioid use disorders, and the odds did not significantly change after adjustment for demographic and clinical covariates (adjusted odds ratio=2.17, 95% CI=1.35–3.48, and adjusted odds ratio=2.26, 95% CI=1.40–3.65, respectively). By contrast, while a prescription opioid use disorder at baseline was associated with prevalence of pain at wave 2 (odds ratio=1.84, 95% CI=1.04–3.25), it was not significantly associated with incidence (odds ratio=1.78, 95% CI=0.75–4.24), and neither association was significant after adjustment for demographic and clinical covariates (adjusted odds ratio=1.72, 95% CI=0.95–3.13, and adjusted odds ratio=1.90, 95% CI=0.76–4.70, respectively) (Table 2). There were no significant interactions between time of pain assessment and likelihood of prescription drug use disorder at wave 2 either in unadjusted models (odds ratio=0.96, 95% CI=0.39–2.38) or in models that adjusted for age, sex, other substance use disorders, or any mood or anxiety disorder at wave 1 (odds ratio=0.99, 95% CI=0.40–2.50).

In the model, which adjusted for demographic and clinical characteristics, prescription opioid use disorder and pain at wave 1 each were associated with themselves at wave 2 (for wave 1 pain with wave 2 pain, β=0.53, p<0.001; for wave 1 opioid use disorder with wave 2 opioid use disorder, β=0.25, p=0.002). Moreover, the adjusted correlation between pain and opioid use disorder was substantial at both wave 1 and wave 2. However, whereas prescription opioid use disorder at wave 1 did not lead to pain at wave 2, pain at wave 1 led to prescription opioid use disorder at wave 2 (Figure 1, Table 3). The predicted risk of developing wave 2 prescription opioid use disorder was 0.44% for persons without and 0.62% for those with wave 1 pain interference, an increase of 41% (data not shown). The model was invariant by sex (χ²=46.64, df=34, p=0.07). Post hoc analyses indicated that pain at wave 1 was significantly associated with opioid tolerance (β=0.17, p=0.002) and withdrawal (β=0.24, p<0.001) at wave 2, and the association approached significance for taking opioids longer or at greater doses than prescribed (β=0.12, p=0.07).

Having a mood or anxiety disorder or family history of alcohol use disorder or antisocial personality disorder at wave 1 was associated with pain and prescription opioid use disorder at wave 1. A family history of drug use disorders was further associated with pain at wave 1. In the structural model, age and sex were associated with both prescription opioid use disorder and pain at wave 1, but in opposite directions. Whereas male sex was directly associated and age inversely associated with prescription opioid use disorder, age was directly associated and male sex inversely associated with pain at wave 1. A family history of antisocial personality disorder at wave 1 was also associated with prescription opioid use disorder at wave 2. Having a mood or anxiety disorder and having a family history of alcohol use disorder or a depression disorder at wave 1 was directly associated and male sex inversely associated with pain at wave 2. Age continued to be directly associated with pain and inversely associated with prescription opioid use disorder at wave 2 (Table 3).