Initiation of Opioid Prescription and Risk of Suicidal Behavior Among Youth and Young Adults

As in our previous study of opioid initiation,²⁹  we analyzed a nationwide birth cohort using data through December 31, 2013 from registers linked on personal identity numbers by Statistics Sweden.^30,31  The registers included: Swedish Prescribed Drug Register (all prescriptions dispensed outside hospitals since mid 2005),^32,33  National Patient Register (hospitalizations nationwide since 1987, visits to outpatient specialists since 2001),³⁴  Multi generation Register (family relationships for people residing in Sweden since 1961),³⁵  Longitudinal Integrated Database for Health Insurance and Labour Market Studies (LISA; socioeconomic information since 1990),³⁶  National Crime Register (criminal convictions since 1973),³⁷  Cause of Death Register (causes and dates of death since 1961),³⁸  Total Population Register (residence and migration information),³⁹  and Medical Birth Register (antenatal care and live births since 1973).⁴⁰ 

We began by identifying Sweden-born youth and young adults (operationalized as those aged 9–29 years by January 1, 2013). Eligibility for inclusion began on January 1, 2007, or ninth birthday, whichever occurred later, among those without prior recorded opioid prescriptions, thus ensuring ≥1.5 years of opioid-prescription-free washout. Supplemental Table 3 describes the inclusion criteria and cohort derivation. The Indiana University Institutional Review Board and the Regional Ethics Committee in Stockholm approved the study and determined that informed consent was not necessary for this analysis of deidentified register data.

We defined opioid initiation as the dispensing date of first recorded immediate-release or extended-release or long-acting opioid analgesic prescription, not counting cough and cold preparations (Supplemental Table 4).^41,42  We followed previous register-based research to exclude methadone and buprenorphine prescribed for treatment of opioid use disorder from our opioid analgesic definition.^43,44  As have other studies of opioid initiation, we selected prescription NSAIDs as a comparator medication.^29,45 

We defined suicidal behavior as first diagnosis of self-injurious behavior (including suicide attempt) or death by suicide, recorded using International Classification of Diseases, Tenth Revision codes from inpatient hospitalizations and outpatient specialist visits or from causes of death, respectively (Supplemental Table 5). Health records underestimate the prevalence of suicide (eg, when intent cannot be established).^46–48  Thus, consistent with previous register-based studies,^49–51  we included injuries of undetermined intent in our definition. We also examined 3 secondary outcomes of suicidal behavior by drug/medication overdose, excluding overdose or any other self-poisoning, and excluding injuries of undetermined intent. We excluded individuals with suicidal behavior prior to initiation to study incident outcomes.

To adjust for measured potential confounders, we selected covariates, assessed prior to initiation, that could plausibly influence initiation and suicidal behavior. Covariates included individual (demographics, inpatient/outpatient specialist psychiatric diagnoses, and psychoactive prescriptions), maternal pregnancy-related (smoking during pregnancy, birth year, and birth order), parental (demographics, inpatient psychiatric hospitalizations, opioid prescription, education levels, and criminal convictions), and socioeconomic (parental cohabitation at birth, within-year decile ranks⁵²  of neighborhood deprivation⁵³  and family income [defined as the average of available parental disposable incomes] at age 8) factors (Supplemental Table 6 lists all covariates).

We used a survival analysis framework, because follow-up time could be right censored by emigration, nonsuicide death, end of study, or change in exposure condition.⁵⁴  We report the Kaplan-Meier cumulative incidence of suicidal behavior within 5 years of initiation as an illustrative index of absolute risk, and we estimated the hazard ratio (HR) for the full follow-up using Cox proportional hazards regression. We analyzed the data in SAS 9.4 (SAS Institute, Inc, Cary, North Carolina).

Our first comparison characterized incident suicidal behavior among opioid initiators relative to noninitiators. We matched singleton (to avoid matching twins) initiators with noninitiators 1:1 on year and month of birth, sex, and county of residence at the start of follow-up. We assumed that suicidal behavior recorded on the initiation date was premorbid,⁵⁵  so follow-up time began the day after initiation or, for noninitiators, the corresponding date for their matched initiator. We compared matched initiators and noninitiators using stratified Cox regression.

Our second comparison reduced confounding via an active comparator design,^27,56,57  which adjusted for all factors shared across individuals receiving prescription opioids and NSAIDs (eg, pain management indications). We required individuals to be new initiators of either opioids or NSAIDs (not both simultaneously) from January 1, 2007 (or their ninth birthday) onward, with follow-up beginning after first prescription of either medication. We adjusted for measured covariates using inverse probability of treatment weighting (IPTW). Specifically, we weighted observations by the inverse of the predicted probability of their medication exposure conditional on their covariates to ensure that exposure was independent of covariates.⁵⁸  We stabilized weights by taking their product with the predicted marginal probability of exposure status,⁵⁹  and we truncated at the first and 99th percentiles to reduce the impact of extreme values.⁶⁰  We then fit weighted Cox proportional hazards regressions using robust standard errors.⁶¹  Because our IPTW approach required complete data, we excluded individuals with missing covariates (Supplemental Table 3).

Our third set of analyses examined the robustness of the results. First, we examined whether the association varied by opioid strength (weak [codeine, dextropropoxyphene, or tramadol] vs strong) and formulation (extended-release or long-acting vs immediate-release).²⁹  Second, we excluded individuals who were dispensed methadone or buprenorphine as their initial prescription to ensure that opioid use disorder treatment misclassified as analgesia did not bias the results. Third, we examined whether associations differed between youth (initiation age <19 years) and young adults. Fourth, we examined whether associations varied across IPTW specifications.⁶¹  Fifth, we examined unadjusted associations when including individuals with missing covariates to examine the extent to which the complete-covariate-case approach biased the results. Sixth, because NSAIDs are recommended for acute dental pain,⁶²  we limited our analysis to dental initiations (prescriptions written by dentists working in dentistry, jaw surgery, or oral surgery clinics) to attempt to further reduce confounding by examining a specific indication.^63,64  Seventh, we excluded individuals who died of cancer or who received cancer-related analgesic prescriptions to reduce confounding by cancer-related pain specifically.⁵²  Eighth, we explored a more common mental health outcome of incident depressive disorder diagnosed from inpatient and outpatient specialist visits.⁵⁵ 

Finally, similar to our previous study,²⁹  we applied a sibling comparison as an alternative approach to reduce confounding.^28,65  This design adjusted for confounding from all genetic and environmental sources of sibling similarity by comparing opioid initiators with their noninitiator full siblings.⁶⁶  Noninitiators were assigned the same follow-up start date as their initiator sibling (censoring at later prescription to a noninitiator sibling). We compared siblings using stratified Cox regression, with covariates that could differ meaningfully within families.⁶⁷ 

The cohort comprised 1 895 984 youth and young adults (48.5% female) without prior recorded opioid prescriptions, representing 93.8% of all eligible births in Sweden. Of these individuals, 201 433 (10.6%) initiated opioid prescription during follow-up (median initiation age = 20.74 years; interquartile range [IQR], 17.89–23.48). Supplemental Table 7 presents specific opioids included.

As a preliminary characterization of the incidence of suicidal behavior among opioid initiators without prior suicidal behavior, we compared the 180 808 included singleton initiators with demographically matched noninitiators. The cumulative incidence within 5 years was 2.9% (95% CI, 2.7–3.0) among opioid initiators and 1.2% (95% CI, 1.1–1.2) among noninitiators (Fig 1). This difference corresponded to a more than doubled hazard of suicidal behavior among initiators (HR = 2.64; 95% CI, 2.47–2.81).

To help rule out confounding (eg, by indication), we compared prescription opioid and NSAID initiators among those new to both medications. After removing 41 757 individuals (10.9%) with missing covariates, this comparison included 86 635 opioid and 255 096 NSAID initiators (Table 1). As shown in Fig 2, the weighted cumulative incidence of suicidal behavior within 5 years was 2.2% (95% CI, 2.1–2.4) for opioid initiators and 1.9% (95% CI, 1.9–2.0) for NSAID initiators (HR = 1.19; 95% CI, 1.11–1.28; Table 2). See Supplemental Table 8 for weighted covariate descriptive statistics. As shown in Table 2, we found comparably small associations for secondary suicidal behavior outcomes: by overdose, excluding overdose and all other self-poisoning, and excluding injuries of undetermined intent.

Sensitivity analyses supported these results (Supplemental Table 9). First, associations were similar for immediate-release weak opioids, strong opioids, and extended-release or long-acting strong opioids, suggesting that opioid strength and formulation did not solely drive the association. Second, the association was identical when excluding buprenorphine and methadone initiators. Third, although the association was relatively small in magnitude for all ages, it was greater among young adults than youth/adolescents. Fourth, results were comparable across varying IPTW specifications. Fifth, the unadjusted association among all eligible individuals was similar to the unadjusted association among the analytic cohort, suggesting little bias due to the complete-covariate-case approach. Sixth, the association was somewhat greater when limited to dental indications, albeit with reduced precision, suggesting that residual confounding by indication may not have substantively inflated our results. Seventh, the association was similar when excluding individuals with cancer, suggesting little confounding by cancer-related pain. Eighth, whereas opioid initiators had doubled hazard of depressive disorder relative to demographically matched noninitiators (HR = 1.98; 95% CI, 1.89–2.07; N = 170 009 initiators), the active comparator association for depressive disorder was virtually null (HR = 0.99; 95% CI, 0.94–1.04; N = 83 984 opioid and 248 622 NSAID initiators).

We used a sibling comparison as an alternative means of adjusting for unmeasured confounding shared within families. The sibling comparison included 229 952 individuals in 97 366 discordantly exposed families (Supplemental Table 10) after removing 50 825 individuals (18.1%) with covariate missingness (Supplemental Table 3). Opioid initiators had an 83% relatively greater hazard of suicidal behavior compared with their noninitiator siblings (adjusted HR = 1.83; 95% CI, 1.67–2.01).

This study used nationwide Swedish data on suicide and other clinically diagnosed self-injurious behavior to examine a critical gap in understandings of the adverse effects of opioid analgesic prescription to youth and young adults. We found that young people who initiated opioid prescription had more than doubled risk of suicidal behavior relative to demographically similar noninitiators: 2.9% of initiators experienced at least some suicidal behavior within 5 years, compared with 1.2% of opioid nonrecipients. However, after adjustment for confounding, the association was greatly attenuated. Specifically, opioid initiators had 19% relatively greater adjusted risk of suicidal behavior than did prescription NSAID initiators. Although this association was statistically significant, it was small, corresponding to 3 additional opioid initiators per 1000 experiencing suicidal behavior within 5 years. We found similarly modest associations for overdose-specific suicidal behavior and across most sensitivity analyses. These results are consistent with the hypothesis that confounding from preexisting factors associated with pain indications, rather than opioid initiation itself, drives much of the observed risk of suicidal behavior among opioid initiators.

We note that when, similar to our previous study,²⁹  we used a sibling comparison as an alternative means of adjustment for unmeasured confounders, we found a somewhat stronger association. Relative to their noninitiators siblings, opioid initiators had an 83% relatively greater risk. This pattern may suggest that the sibling comparison had less capability than the active comparator design to adjust for critical confounders (eg, pain indications). Sibling comparisons can be biased by unmeasured factors that vary among siblings,⁶⁸  and pain is likely to be an important confounder given its strong relationship with suicide.^21,23  On the other hand, it is also possible that the different results across designs could indicate that the active comparator was biased toward the null. For example, it might be that NSAIDs could also be used in intentional self-poisoning or that unmeasured factors led to greater preexisting risk of suicidal behavior among NSAID initiators. We believe these explanations are unlikely, however, because findings persisted when excluding self-poisonings and preexisting mental health conditions were more prevalent among the opioid initiators. Thus, our results suggest that the active comparator design may be more appropriate than sibling comparisons for future research on opioid prescription and suicidal behavior, although incorporating additional designs (eg, Mendelian randomization or other instrumental variables) may help strengthen inferences.^17,19 

To the extent that at least some association with suicidal behavior did persist in our active comparator results, it may be explained by multiple possible processes. Opioid initiation itself may modestly increase risk. Interestingly, the persistence of associations for nonself-poisoning suicidal behavior indicates that access to lethal overdose means may not completely explain any increased risk, which is consistent with prior studies.^13,69  We also found no support for other mental health concerns (ie, depression) following initiation, although our assessment of specialist diagnoses (rather than those made in primary care) likely captured only severe depression.⁷⁰  It is important to note that the association may also reflect residual confounding from processes other than opioid initiation. For example, to the extent that they were not addressed by the NSAID comparison or measured covariates, specific painful conditions,²⁵  greater pain severity or distress,^71,72  psychopathology,¹³  or broader socioeconomic factors⁷³  could explain the present results.

Increasing rates of suicide by opioid overdose⁵  have generated concern regarding opioid prescription in general,¹²  although less attention has been focused on youth.¹⁸  Our findings of a relatively small association of initiation with suicidal behavior in young people may help inform cautious harm/benefit evaluations of pain management with opioids at individual patient and broader policy levels.⁷⁴  Although our results may provide some reassurance regarding a minor potential increase in risk of suicidal behavior, the full decision-making context would include other adverse outcomes as well, including potential risk of persistent use,⁶³  diversion,⁷⁴  substance use disorder,²⁹  and unintentional overdose.⁷⁵  Moreover, our results should not be interpreted as relevant to the potential harms of other opioid prescription patterns. Indeed, previous research has demonstrated increased risk of adult suicidal behavior associated with dose escalation¹⁹  and discontinuation.⁷⁶  Further research is needed to explore other patterns, such as duration and concurrent pharmacotherapies (eg, benzodiazepines). Regardless of whether initiation truly impacts risk of suicidal behavior, however, it is clear that opioid recipients would benefit from assessment and, when relevant, interventions to prevent suicide and address related mental health concerns.¹²  Ongoing screening among patients receiving or being considered for pharmacologic pain management, including those without recorded mental health concerns, would be consistent with recommendations for systematic assessment of higher-risk populations.⁷⁷ 

Our analysis should be interpreted considering its limitations. First, the active comparator design cannot establish causality: As stated above, it does not account for unmeasured factors that differentiate opioid from NSAID initiators. Further adjustment for confounding by indication or other factors may have resulted in even greater attenuation of the association. The design also assumes no change in risk of suicidal behavior due to NSAID initiation.^57,78  Second, although the register linkage comprised nationwide data, it did not include records from primary care or diagnoses of some painful conditions (eg, fibromyalgia), precluding their use as statistical covariates. Future research examining specific painful conditions is warranted. Third, we attempted to address underestimation of suicidal behavior by assessing diagnoses and deaths of undetermined intent,⁴⁶  but not all suicidal behavior is clinically recognized or results in inpatient or specialist outpatient care.⁷⁹  Thus, our outcomes likely only included more severe suicidal behavior and may have also failed to capture some intentional self-injury (eg, overdoses) clinically classified as accidental. Fourth, like other health care-record research that defines initiation based on dispensed prescriptions, our results were analogous to an intention-to-treat approach (for those who were dispensed prescriptions).⁸⁰  Although we cannot ensure that participants never received opioids before the available data, our ≥1.5-year opioid-prescription-free washout compares favorably with other youth opioid research.⁴²  Finally, our study examined opioid prescription in Sweden through 2013. Although pharmacologic mechanisms should not vary across countries, we do not know how well our results would generalize beyond the Swedish pain management context. Moreover, as opioid prescription has continued to decline,^81,82  those initiating therapy more recently may experience greater pain severity or less response to other treatments.⁸³  How these trends may affect risk of suicidal behavior should be examined in future research.

Swedish youth and young adults initiating opioid prescription had more than doubled risk of suicidal behavior relative to noninitiators. However, compared with NSAID initiators, this association attenuated to a 19% relative difference, or an additional 3 of 1000 experiencing suicidal behavior within 5 years, providing evidence of considerable confounding by indication. Whereas the results cannot definitively establish the magnitude of a potential influence of opioid initiation, they suggest that any increase in risk of suicidal behavior is unlikely to be large, and decisions to initiate opioid prescription should consider the full range of potential benefits as well as harms.

CI

confidence interval

HR

hazard ratio

IPTW

inverse probability of treatment weighting

IQR

interquartile range

NSAID

nonsteroidal anti-inflammatory drug