Associations of Mental Health and Family Background With Opioid Analgesic Therapy

A Nationwide Swedish Register-Based Study

Patrick D. Quinn; Martin E. Rickert; Johan Franck; Amir Sariaslan; Katja Boersma; Paul Lichtenstein; Henrik Larsson; Brian M. D'Onofrio


Pain. 2019;160(11):2464-2472. 

In This Article


Registers and Cohort

We identified a cohort of adolescents and adults in Sweden who were naive to prescription opioid analgesics, drawing data from population registers linked by Statistics Sweden.[53,70] As has been described previously, Swedish and other Nordic register data represent unique resources for epidemiology because of their whole-population coverage and potential for linkage through personal identity numbers that are assigned at birth or immigration.[2,40,48] The linked registers included: Total Population Register (including demographics, residence, and migration),[40] Swedish Prescribed Drug Register (all prescription medications dispensed outside of hospitals since July 2005),[79,82] Multi-generation Register (familial relationships for individuals born since 1932 or living in Sweden since 1961),[16] National Patient Register (nationwide inpatient hospitalizations since 1987 and outpatient specialist visits since 2001),[41] Longitudinal integration database for health insurance and labour market studies (LISA; diverse socioeconomic information since 1990),[42] and Cause of Death Register (deaths since 1961).[49] The Indiana University Institutional Review Board and the Regional Ethics Committee in Stockholm approved this study.

We included Sweden-born residents aged 13 to 73 years on January 1, 2007. We excluded individuals with any recorded opioid analgesics dispensed in 2005 to 2006, as well as those with terminal cancer or with dispensed cancer-related opioid prescriptions. We followed all individuals from January 1, 2007, through the first of emigration, death, or the end of the study (December 31, 2014), requiring at least 6 months of follow-up. The final cohort comprised 5,071,193 individuals (48.4% female), of whom 158,496 (3.1%) had less than 8 years of follow-up. The cohort represented 78.8% of the original birth cohort and 89.0% of those without registered deaths or emigrations before the start of follow-up.

In addition, SES data were not available from LISA before 1990.[42] We therefore examined associations between family background and offspring opioid receipt among a subcohort of individuals with age-12 family-of-origin data by January 1, 2007 (ie, those aged 13–29 years in 2007; N = 1,482,462). See eTable 1, supplemental digital content, for complete inclusion and exclusion criteria (available at

Opioid Prescription

The Swedish Prescribed Drug Register includes dispensing dates and information on products and prescribers for dispensed prescription medications.[82] Using anatomical therapeutic chemical codes, we identified opioid analgesic prescriptions dispensed from 2007 to 2014, including weak opioids (ie, codeine, dextropropoxyphene, and tramadol) but not medications listed as cough and cold preparations (eTable 2, supplemental digital content, lists all included anatomical therapeutic chemical codes, available at To differentiate buprenorphine and methadone dispensed as medication-assisted treatment (MAT) of opioid use disorder (OUD) from prescriptions for analgesia, we followed prior research.[1,37] Specifically, prescriptions of transdermal buprenorphine, prescriptions dispensed to adolescents (younger than 18 years), or prescriptions dispensed through the "ApoDos" multidose packaging system were considered analgesics. Other buprenorphine and methadone prescriptions were counted as analgesics only if they were prescribed by nonpsychiatry providers in nonpsychiatry practices.

Long-term Opioid Therapy. Prior research has used multiple operationalizations of LTOT, and there is no single agreed-upon definition.[15,81] In Sweden, opioid prescriptions are dispensed for a maximum of 3 months' supply.[66] We therefore defined LTOT as strong (ie, not weak) opioids dispensed on 2 or more dates within a 6-month window, which is consistent with previous Swedish research and matches the 6-month windows used in some US studies.[24,51,74] Sensitivity analyses, described below, evaluated this definition.

Concurrent Benzodiazepine Therapy. We examined concurrent benzodiazepine-opioid therapy as an index of potentially harmful deviation from opioid prescription guidelines. We defined concurrence as prescriptions dispensed within 30 days of each other, with the concurrence beginning on the later dispensing date.

Preexisting Mental Health

The National Patient Register includes discharge dates and diagnoses (recorded as International Classification of Diseases [ICD] codes) from inpatient hospitalizations and outpatient specialist visits.[41] We identified inpatient and outpatient clinical diagnoses before 2007 for OUD and other SUD, suicide attempt or other self-injury, and attention-deficit/hyperactivity, depressive, anxiety, schizophrenia spectrum, and bipolar disorders (eTable 3, supplemental digital content, lists all ICD codes, available at We adapted ICD codes from prior opioid-related and register-based research.[52,53] As an alternative index of mental health, we also examined 2005 to 2006 receipt of psychoactive medications (ie, antidepressants, antipsychotics, benzodiazepines, mood stabilizers, nonbenzodiazepine z-hypnotics, and medications used in the treatment of attention-deficit/hyperactivity disorder, OUD, and other SUD). We counted as MAT of OUD only those nonanalgesic buprenorphine and methadone prescriptions (defined above) dispensed from psychiatry specialists in psychiatry practices.

Family Background

To assess family mental health history, we identified maternal and paternal diagnoses of anxiety or depressive disorder, OUD or other SUD, and suicide attempt or other self-injury from inpatient hospitalizations before 2007. We also included several indices of family-of-origin SES. First, we identified the highest level of maternal and paternal education through 2006.[53] Second, we estimated family income at offspring-age 12 years (ie, before the start of follow-up). We calculated family income as the mean of available maternal and paternal disposable family incomes.[36] To adjust for inflation or transient economic events, we created within-year decile-rank scores. Third, we included a deprivation score for the neighborhood of residence at age 12 years, which we also analyzed using within-year decile-rank scores. As described elsewhere, the deprivation score was generated for each year from a principal components analysis of all 25- to 64-year-old neighborhood members' socioeconomic characteristics (ie, welfare receipt, unemployment, immigration, divorce, secondary education, disposable income, and crime), where neighborhood was defined using Statistics Sweden's homogenous geographic Small Areas for Market Statistics.[61] In addition to the registers described above, the neighborhood deprivation score additionally drew on the National Crime Register, which includes convictions since 1973.[18] Fourth, we examined parental immigrant status, defined as having at least one parent born outside of the Nordic countries. Finally, we included an indicator of metropolitan residence (in Stockholm, Göteborg, or Malmö) at age 12 years.

Statistical Analysis

Our objective was to examine the extent to which opioid therapy rates differed among those with and without preexisting mental health conditions and family background factors. We used a survival analysis approach in SAS 9.4 (SAS Institute, Inc) to examine these associations because follow-up time was right censored. We calculated Kaplan–Meier estimates of the cumulative incidence of opioid receipt and estimated hazard ratios (HRs) using Cox proportional hazards regressions.[3] We set censoring dates to 6 months before the end of available follow-up (ie, the last date at which an individual was eligible to initiate a 6-month LTOT window). Consistent with our previous studies, we report the cumulative incidence of opioid therapy receipt within 3 years as an index of absolute risk.[51,52]

The first set of models examined associations with rates of opioid initiation, defined as first dispensed opioid. The timescale was years since January 1, 2007. The second and third sets of models examined associations with the transition to LTOT and concurrent benzodiazepine therapy, respectively. These models were limited to those who had initiated opioids, and the timescale was years since initiation. We examined each mental health condition or family factor in a separate regression because our goal was largely descriptive (ie, characterizing opioid prescription patterns among those with and without mental health and family background factors). We included demographic covariates, which are detailed in eTable 4, supplemental digital content, available at (county of residence in 2007 [Stockholm, Västra Götaland, Skåne, or other counties], age group, sex, and—for LTOT and concurrent benzodiazepines—calendar year of opioid initiation). In the family background analyses, we accounted for the clustering of individuals born to the same mother with robust standard errors.

Sensitivity analyses tested the robustness of the associations with opioid initiation and LTOT. First, to ensure that differences in predictor assessment periods did not bias the results, we restricted the preexisting mental health condition diagnoses to the period when outpatient and inpatient records were available (2001-2006). Second, we excluded the 34,374 individuals who were ever dispensed buprenorphine or methadone to ensure that our results were not biased by misclassification of MAT vs analgesia. Finally, we examined the sensitivity of the results to the operationalization of LTOT. Our definition, based on previous Swedish and US studies,[24,51] required only 2 dispensed strong opioid prescriptions. If some recipients merely switched drugs because of negative reactions to an acute therapy prescription, the definition could have overestimated LTOT incidence. We therefore additionally examined a stricter definition of LTOT (ie, strong opioids in 3 of 4 consecutive 90-day quarters) drawn from other previous Nordic research.[21]