Abstract
Background:
Overdose due to concomitant use of opioids and benzodiazepines has been raised as a major public health concern, although little research has examined whether this risk extends to intentional overdose or other self-harm. This study examined whether prescription opioids and benzodiazepines interact to increase the rate of suicide attempt and intentional self-harm.
Methods:
The study analyzed 4,762,438 users of opioids, benzodiazepines, both drugs concomitantly, or neither drug from the MarketScan Commercial Claims and Encounters databases (2014–2016). The four groups were matched using inverse probability of treatment weighting and a difference in difference design was used to examine associations with risk of suicide attempt and intentional self-harm, including suicide death resulting in a medical claim.
Results:
There was a small association for opioids (HR=1.23; 95% CI 1.06 to 1.43) but a larger association for benzodiazepines (HR=2.55; 95% CI 2.12 to 3.05) with suicide attempt and intentional self-harm. The medication interaction was opposite to the expected direction (HR=0.70; 95% CI 0.55 to 0.89), indicating that risk associated with concomitant use was lower than would be expected on an additive basis. Sensitivity analyses found no evidence of increased risk due to interaction between the two drug classes.
Conclusions:
Increased risk of suicide attempt and intentional self-harm for concomitant use of opioids and benzodiazepines is in large part attributable to benzodiazepine use alone. In typically prescribed quantities, opioids and benzodiazepines may not represent a drug interaction in terms of yielding increased risk of suicide attempt and intentional self-harm resulting in medical care.
Keywords: Opioids, Benzodiazepines, Suicide Attempt, Self-harm, Difference in Difference Analysis
1. Introduction
Although the overall rate of opioid prescribing in US treatment of acute and chronic pain has decreased since 2012, it remains at three times the level that it was in 1999 (CDC, 2018; NAS, 2019) and greatly exceeds rates elsewhere worldwide (INCB, 2018). It is well-established that people with opioid use disorder are at greater risk of suicide (Olfson et al, 2019). In contrast, the effect of prescribed opioid analgesic use is less clear, with some—but not all (Samples et al, 2019)—emerging evidence supporting a possible contribution to suicide risk (Bohnert and Ilgen, 2019). For example, among Veterans Health Affairs patients receiving opioid prescription, higher dosage is associated with greater risk of suicide mortality and self-inflicted injuries (Ilgen et al, 2016; Hayes et al, 2020). Opioid use for over a year in parents is also associated with a doubling of suicide attempt risk in their children (Brent et al, 2019). Benzodiazepines are prescribed for anxiety and insomnia and include drugs like diazepam, clonazepam and alprazolam (Olfson et al, 2015). Use of benzodiazepines has increased dramatically from 2003 to 2015 (Agarwal and Landon, 2019), from 3.8% to 7.4% of outpatient ambulatory care visits resulting in a benzodiazepine prescription. Benzodiazepine use has been found to be associated with suicide (Cato et al, 2019) and suicide attempts (Gibbons et al, 2019).
Concomitant use of opioids and benzodiazepines is associated with overdose deaths, with 23% of opioid overdose deaths also involving benzodiazepines (NIDA, 2018). This risk likely reflects a pharmacologic interaction wherein combined use can potentiate respiratory depressant effects Jann et al, 2014; White and Irvine, 1999). Several pharmacoepidemiologic studies have found that, relative to patients receiving prescription opioid analgesics only, those receiving prescription benzodiazepines concomitantly have greater risk of overdoses (Sun et al, 2017), overdose deaths (Park et al, 2015; Dasgupta et al, 2016; Garg et al, 2017), and all-cause mortality (Hawkins et al, 2019). Similarly, studies have demonstrated greater risk of all-cause and overdose mortality among those receiving opioids and benzodiazepines than among those receiving benzodiazepines alone (Hawkins et al, 2019; Xu et al, 1999). Consequently, co-prescription of opioid analgesics and benzodiazepines has been strongly discouraged (Dowell et al, 2016). Critically, it is also possible that this potential for harm would extend to intentional overdose or other intentional self-harm. For example, the opioid-benzodiazepine interaction might worsen the medical severity of an intentional overdose. Whether this possibility would be clinically relevant, particularly for commonly prescribed medication quantities, is unknown (Ilgen, 2018). As evidence of the importance of this problem, the Department of Veterans Affairs (VA) recently commissioned the National Academy of Sciences to design a pharmacoepidemiologic study to determine the effects of concomitant prescribing of opioids and benzodiazepines in terms of all-cause mortality and suicide mortality (NAS, 2019).
We note that, to the extent that patients who are prescribed opioids with benzodiazepines concomitantly experience greater risk of suicide attempt and intentional self-harm, any causal drug effects must be differentiated from alternative (confounding) processes. For example, increased risk of suicide attempts, deaths, and other intentional self-injury may be due to comorbid pain (Racine, 2018) and anxiety disorders which may have existed before either drug exposure. Alternatively, suicide risk associated with prescription benzodiazepines may be higher than that for prescription opioids, giving the appearance of a drug interaction, when the increased risk is simply due to increased suicide risk associated with prescription benzodiazepine use. Moreover, even if individually these do represent true adverse drug reactions, the combined effects may simply represent the sum of their individual contributions to suicidal risk (i.e. the effects are additive [Park et al, 2015; Turner and Liang, 2015] rather than multiplicative). Alternatively, increased risk associated with concomitant use of prescription opioids and benzodiazepines may well represent a drug interaction which is associated with higher risk of suicide than is expected by the sum of the two individual prescription drug risks. The purpose of this paper is to explore these various alternatives concerning risk of suicide attempt and intentional self-harm using large scale medical claims data.
2. Material and Methods
2.1. The Sample
Data were obtained from the MarketScan Commercial Claims and Encounters databases (Hansen, 2016). The MarketScan data are distributed through IBM Watson and include inpatient, outpatient, and prescription claims from more than 100 insurers in the United States (164 million unique enrollee observations between 2005 and 2017). International Statistical Classification of Diseases and Related Health Problems codes (ICD-9-CM and ICD-10-CM) were used to identify suicide attempt and intentional self-harm, as well as other diagnoses of interest, from service claims and are presented in Supplemental Table 1. A list of specific drugs used in the analysis is presented in Supplemental Table 2.
The sample consisted of patients aged 18 and older with opioid and/or benzodiazepine prescriptions identified using the 2016 data. We used a new user cohort design to permit a difference in difference analysis. Patients who had at least one opioid or benzo prescription in 2016 and had no history of opioid and benzo prescriptions within 2 years prior to the index date were included. Patients were grouped into opioid only, benzodiazepine only, and concomitant opioid and benzodiazepine use groups. A no treatment group (no opioid or benzodiazepine prescription fills) was also identified using the 2016 data. These patients were randomly selected in number equal to the total number of patients in the opioid, benzodiazepine, and combined-use groups. Patients were followed from the index date, which was defined as the first opioid or benzodiazepine (or both for combined use) prescription date or first prescription of a non-opioid, non-benzodiazepine drug (for the no treatment comparison group) in 2016. Follow-up ended at a suicide attempt or intentional self-harm, disenrollment, or the end of 2017. The study was judged to be exempt by the University of Chicago IRB.
2.2. Statistical Methods
A multinomial propensity score model was developed including age, gender, clinically diagnosed psychiatric conditions (substance use disorder, depressive disorder, bipolar or other mood disorder, anxiety/stress/adjustment disorder, attentiondeficit/hyperactivity disorder, schizophrenia spectrum disorder, and sleep disorder) and filled psychoactive prescriptions (Z drugs (eszopiclone, zaleplon and zolpidem), ADHD medications, SSRIs, SNRIs, TCAs, duloxetine, other antidepressants, antipsychotics, mood stabilizers, gabapentin and pregabalin, alcohol use disorder drugs, opioid use disorder drugs, tobacco cessation drugs, non-benzodiazepine anxiolytics, muscle relaxants and antiepileptics) (see Supplemental Table 2). We included these variables as potential confounders in the propensity score adjusted analysis to balance the four groups in terms of drugs and indications for drugs (e.g. mental health, substance use and painful disorders) that are independently associated with suicide risk. We used inverse probability of treatment weighting (IPTW), based on the propensity score to create a synthetic sample in which the distribution of measured baseline covariates is independent of treatment assignment (Morgan and Todd, 2008; Rosenbaum, 1987). We estimated the average treatment effect (Lunceford and Davidian, 2004) and used stabilized weights (Robins et al, 2000), truncated at the 0.25th and 97.75th percentiles (to avoid extreme inverse weights - Cole and Hernan, 2008), in predicting suicide attempt and intentional self-harm diagnoses. Deaths by suicide would be included if they occurred in hospitalized patients and generated a medical claim.
The outcome model was a Cox regression model with and without adjustment for prior suicide attempt and intentional self-harm and excluding subjects with prior suicide attempt or intentional self-harm as sensitivity analyses. As a descriptive comparison, Cox models were first computed using a simple one-factor model in which risk of suicidal attempt and intentional self-harm following medication exposure in each drug group was compared to that in the no-use group.
To better account for unmeasured confounding, our primary analysis compared risk of suicide attempt and intentional self-harm in the two-year period before drug initiation with the risk after initiation and used a Cox model for a 2 × 2 × 2 design (opioid, benzodiazepine, pre- and post-exposure) with all main effects, two-way interactions and the three-way interaction. This model tests whether these exposures, either individually or in combination, are associated with increased risk of suicide attempt and intentional self-harm following drug initiation relative to the two years prior to drug initiation. This is a ―difference in difference‖ analysis (Angrist and Pischke, 2008), which is a within-subject design and thus eliminates between-subject factors that can confound the effects of opioids and benzodiazepines. Suicide attempt and intentional self-harm diagnoses on the index date were considered to be pre-exposure events (Man et al, 2017). Standard errors were adjusted for correlation using the robust sandwich estimator (Lin and Wei, 1989).
We conducted several sensitivity analyses to examine the specificity of the results to particular types of suicide attempt and intentional self-harm and patterns of medication treatment. First, we examined whether 2 × 2 × 2 difference in difference results would differ for drug/medication overdose suicide attempt and intentional self-harm relative to other events. Second, because claims data likely underestimate the incidence of suicide attempt and intentional self-harm, especially when the intent behind overdose and other self-injury is not known, we expanded our outcome definition to include injuries of undetermined intent (Bohnert et al, 2013). Third, we added analyses of accidental drug overdose as a comparison outcome given prior research on opioids and accidental overdose (e.g., Quinn et al., in press).
A final sensitivity analysis was performed to examine dosage effects. We examined thresholds of 1–90 days, 91–180 days, and 180 or more prescription day supply (of either drug or both depending on group), and repeated the 2 × 2 × 2 difference in difference analysis for each. The no Rx was based on their day supply on any prescription in 2016 (i.e. non-opioid and non-benzodiazepine drug).
3. Results
3.1. IPTW Adjustment
Using the 2016 data, we identified 1,745,199 opioid only patients, 396,894 benzodiazepine only patients, 239,126 combined-use patients, and 2,381,219 no-use patients. There were a total of 5,092 suicide attempts and intentional self-harm events. Table 1 displays summary statistics for the 25 potential baseline confounders (within 1 year of the index date) before and after IPTW based on the propensity score. Groups with benzodiazepine use had more female patients and had higher rates of antidepressants, antipsychotics, mood stabilizers and antiepileptic drugs prior to IPTW adjustment. Following IPTW adjustment the groups were quite similar with respect to all measured confounders.
Table 1:
Baseline covariates (within 1 year of index date)
| Unweighted | Weighted | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Opioid | Benzo | Op + Bz | None | Opioid | Benzo | Op + Bz | None | |||||||||
| (n=1,745,199) | (n=396,894) | (n=239,126) | (n=2,381,219) | (n=1,755,988) | (n=369,761) | (n=236,880) | (n=2,398,567) | |||||||||
|
| ||||||||||||||||
| Covariate | N (mean) | % (sd) | N (mean) | % (sd) | N (mean) | % (sd) | N (mean) | % (sd) | N (mean) | % (sd) | N (mean) | % (sd) | N (mean) | % (sd) | N (mean) | % (sd) |
| Age | 41.9 | 13.6 | 42.9 | 12.8 | 44.1 | 12.6 | 41.4 | 13.7 | 41.8 | 13.7 | 42.8 | 12.3 | 42 | 12.8 | 41.8 | 13.7 |
| Male | 7,76.876 | 44.52 | 1,33,141 | 33.55 | 83,138 | 34.76 | 10,65,030 | 44.73 | 7,55,816 | 43.04 | 1,54,246 | 41.72 | 1,03,761 | 43.80 | 10,39,655 | 43.34 |
| Prior self-harms/suicide | 563 | 0.03 | 468 | 0.12 | 318 | 0.13 | 313 | 0.01 | 747 | 0.04 | 173 | 0.05 | 234 | 0.10 | 580 | 0.02 |
| Medication Use | ||||||||||||||||
| Z Drugs | 44,152 | 2.53 | 25,132 | 6.33 | 13,883 | 5.80 | 31,643 | 1.33 | 45,359 | 2.58 | 11,137 | 3.01 | 5,968 | 2.52 | 63,630 | 2 |
| ADHD | 46,789 | 2.68 | 23,265 | 5.86 | 10,824 | 4.53 | 66,534 | 2.79 | 57,056 | 3.25 | 14,120 | 3.82 | 7,245 | 3.06 | 76,305 | 3.18 |
| SSRI | 1,47,647 | 8.46 | 1,11,114 | 28.00 | 37,370 | 15.62 | 1,65,061 | 6.93 | 1,77,774 | 10.12 | 48,070 | 13.00 | 24,135 | 10.19 | 2,46,451 | 10.27 |
| SNRI | 22,753 | 1.30 | 15,618 | 3.94 | 6,441 | 2.69 | 21,063 | 0.88 | 26,330 | 1.50 | 7,113 | 1.92 | 3,497 | 1.48 | 36,283 | 1.51 |
| TCA | 19,307 | 1.11 | 7,023 | 1.77 | 4,398 | 1.84 | 13,284 | 0.56 | 16,968 | 0.97 | 4,383 | 1.19 | 2,343 | 0.99 | 24,165 | 1.01 |
| Other Antidepressant | 68,250 | 3.91 | 43,832 | 11.04 | 18,920 | 7.91 | 61,398 | 2.58 | 76,873 | 4.38 | 21,125 | 5.71 | 10,180 | 4.30 | 1,06,926 | 4.46 |
| Duloxetine | 19,329 | 1.11 | 9,913 | 2.50 | 5,940 | 2.48 | 11,269 | 0.47 | 18,460 | 1.05 | 4,905 | 1.33 | 2,508 | 1.06 | 26,391 | 1.10 |
| Antipsychotics | 12,179 | 0.70 | 13,175 | 3.32 | 4,905 | 2.05 | 13,179 | 0.57 | 19,373 | 1.10 | 5,151 | 1.39 | 2,347 | 0.99 | 26,827 | 1.12 |
| Mood stablizer | 9,407 | 0.54 | 9,797 | 2.47 | 3,514 | 1.47 | 10,588 | 0.44 | 14,949 | 0.85 | 3,719 | 1.01 | 1,754 | 0.74 | 19,820 | 0.83 |
| Gabapentin/Pregabalin | 55,283 | 3.17 | 11,910 | 3.00 | 11,902 | 4.98 | 20,600 | 0.87 | 38,041 | 2.17 | 9,785 | 2.65 | 5,318 | 2.25 | 55,859 | 2.33 |
| Alcohol use disorder | 449 | 0.03 | 436 | 0.11 | 122 | 0.05 | 422 | 0.02 | 729 | 0.04 | 213 | 0.06 | 92 | 0.04 | 956 | 0.04 |
| Opioid use disorder | 3,027 | 0.17 | 2,858 | 0.72 | 827 | 0.35 | 5,252 | 0.22 | 5,415 | 0.31 | 1,456 | 0.39 | 686 | 0.29 | 7,235 | 0.30 |
| Tobacco cessation | 10,780 | 0.62 | 2,615 | 0.66 | 1,757 | 0.73 | 8,348 | 0.35 | 8,856 | 0.50 | 2,056 | 0.56 | 1,224 | 0.52 | 12,002 | 0.50 |
| Non-benzo anxiety | 35,926 | 2.06 | 18,194 | 4.58 | 7,691 | 3.22 | 27,733 | 1.16 | 34,892 | 1.99 | 9,288 | 2.51 | 4,910 | 2.07 | 50,141 | 2.09 |
| Muscle Relaxant | 2,17,122 | 12.44 | 27,486 | 6.93 | 29,865 | 12.49 | 80,384 | 3.38 | 1,31,770 | 7.50 | 30,092 | 8.14 | 18.114 | 7.65 | 1,82,377 | 7.60 |
| Antiepileptics | 33,273 | 1.91 | 13,889 | 3.50 | 7,546 | 3.15 | 29,826 | 1.25 | 33,311 | 1.90 | 8,738 | 2.36 | 4,352 | 1.84 | 44,743 | 1.87 |
| Disease Condition | ||||||||||||||||
| Subtance Disorder | 24,843 | 1.42 | 11,178 | 2.82 | 4,954 | 2.07 | 19,128 | 0.80 | 23,599 | 1.34 | 6,670 | 1.80 | 3,199 | 1.35 | 35,605 | 1.48 |
| Depression | 1,00,617 | 5.77 | 61,993 | 15.92 | 22,948 | 9.59 | 94,934 | 3.99 | 1,08,899 | 6.20 | 30,533 | 8.26 | 15,291 | 6.46 | 1,53,643 | 6,41 |
| Anxiety | 1,32,205 | 7.58 | 1,56,640 | 39.47 | 44,016 | 18.40 | 1,32,566 | 5.57 | 1,80,503 | 10.28 | 42,098 | 11.39 | 23,910 | 10.09 | 2,52,863 | 10.54 |
| ADHD | 36,926 | 2.12 | 16,684 | 4.20 | 7,205 | 3.01 | 48,044 | 2.02 | 41,990 | 2.39 | 11,035 | 2.98 | 5,734 | 2.24 | 56,856 | 2.37 |
| Bipolar Disorder | 14,627 | 0.84 | 12,059 | 3.04 | 4,506 | 1.88 | 14,516 | 0.61 | 19,364 | 1.10 | 5,627 | 1.52 | 2,625 | 1.11 | 27,181 | 1.13 |
| Schizophrenia | 2,305 | 0.13 | 2,458 | 0.62 | 616 | 0.26 | 3,662 | 0.15 | 3,795 | 0.22 | 1,084 | 0.29 | 510 | 0.22 | 5,375 | 0.22 |
| Sleep Disorder | 1,24,832 | 7.15 | 47,190 | 11.89 | 21,495 | 8.99 | 93,249 | 3.92 | 109227 | 6.22 | 28528 | 7.72 | 15150 | 6.40 | 155210 | 6.47 |
3.2. Post-Exposure Analysis
Table 2 presents results for the comparisons between the different drug groups, with and without statistical adjustment for prior suicide attempt and intentional self-harm, and with and without the exclusion of patients with prior suicide attempt and intentional self-harm from the analysis. In the unadjusted data, rates for benzodiazepines alone and the combined-use group were similar at 32 and 35 per 10,000 patient years (PY), respectively; however, in the adjusted analysis rates were somewhat lower for the benzodiazepines only group (21 per 10,000 patient-years) versus the combined-use group (29 per 10,000 patient-years). Relative to no use (adjusted for prior suicide attempt and intentional self-harm), rates were significantly higher for opioid use only (HR=1.36; 95% CI, 1.27, 1.45), benzodiazepine use only (HR=3.39; 95% CI, 3.13, 3.67) and especially combined use (HR=4.51; 95% CI, 4.14, 4.91). Excluding or not adjusting for prior suicide attempt and intentional self-harm produced quite similar results.
Table 2:
HR and 95% CI from Cox models with weighting
| Unweighted |
Weighted |
Weighted HR (95% CI) |
|||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Outcome | Group | # of Subjects | # of Events | PY | Rate | # of Events | PY | Rate | No Adj. for Pre | Adj. for Pre | Excluding Pre |
| Self-Harms | Opioid | 17,45,199 | 801 | 19,76,599 | 4.05 | 872 | 19,88,414 | 4.38 | 1.28 (1.17–1.40) | 1.27 (1.16–1.39) | 1.36 (1.24–1.50) |
| Benzodiazepine | 3,96,894 | 708 | 4,41,836 | 16,02 | 431 | 4,17,981 | 10.31 | 3.01 (2.69–3.37) | 3.01 (2.69–3.38) | 3.22 (2.87–3.62) | |
| Op + Benzo | 2,39,126 | 415 | 2,53,393 | 16.38 | 347 | 2,53,110 | 13.69 | 3.95 (3.50–4.47) | 3.84 (3.40–4.34) | 4.18 (3.69–4.74) | |
| None | 23,81,219 | 701 | 30,09,182 | 2.33 | 1,018 | 30,42,743 | 3.34 | ||||
| Drug Overdose | Opioid | 17,45,199 | 1,083 | 19,76,411 | 5.48 | 1,205 | 19,88,197 | 6.06 | 1.41 (1.31–1.53) | 1.42 (1.31–1.53) | 1.41 (1.30–1.53) |
| Benzodiazepine | 3,96,894 | 1,081 | 4,41,562 | 24.48 | 663 | 4,17,795 | 15.87 | 3.71 (3.37–4.07) | 3.72 (3.38–4.08) | 3.79 (3.45–4.17) | |
| Op + Benzo | 2,39,126 | 712 | 2,53,221 | 28.12 | 577 | 2,52,978 | 22.80 | 5.28 (4.78–5.83) | 5.20 (4.71–5.74) | 5.36 (4.85–5.92) | |
| None | 23,81,219 | 863 | 30,09,068 | 2.87 | 1,283 | 30,42,603 | 4.22 | ||||
| Self-Harms + Rx Overdose | Opioid | 17,45,199 | 1,152 | 19,76,151 | 7.70 | 1,665 | 19,87,915 | 8.38 | 1.36 (1.27–1.45) | 1.36 (1.27–1.45) | 1.41 (1.31–1.51) |
| Benzodiazepine | 3,96,894 | 1,416 | 4,41,333 | 32.08 | 868 | 4,17,666 | 20.79 | 3.37 (3.11–3.66) | 3.39 (3.13–3.67) | 3.55 (3.27–3.86) | |
| Op + Benzo | 2,39,126 | 878 | 2,53,114 | 34.69 | 727 | 2,52,882 | 28.74 | 4.61 (4.23–5.03) | 4.51 (4.14–4.91) | 4.82 (4.41–5.27) | |
| None | 23,81,219 | 1,276 | 30,08,775 | 4.24 | 1,836 | 30,42,153 | 6.04 | ||||
| Accidental Drug Overdose | Opioid | 17,45,199 | 3,601 | 19,74,529 | 18.24 | 3,661 | 19,86,350 | 18.43 | 1.63 (1.56–1.71) | 1.63 (1.56–1.71) | 1.63 (1.55–1.71) |
| Benzodiazepine | 3,96,894 | 1,611 | 4,41,223 | 36.51 | 1,199 | 4,17,446 | 28.73 | 2.55 (2.38–2.72) | 2.55 (2.39–2.72) | 2.56 (2.39–2.74) | |
| Op + Benzo | 2,39,126 | 1,485 | 2,52,676 | 58.77 | 1,300 | 2,52,470 | 51.49 | 4.52 (4.24–4.82) | 4.39 (4.11–4.67) | 4.54 (4.25–4.85) | |
| None | 23,81,219 | 2,751 | 30,07,581 | 9.15 | 3,357 | 30,41,006 | 11.04 | ||||
3.3. Pre- and Post-Exposure Difference in Difference Analysis
Of the 4,762,438 patients included in the post-exposure analysis, 4,380,817 (92.0%) patients had enrollment within 2 years before the index drug exposure date and were included in the pre- and post-exposure analysis.
This difference in difference analysis, which formally compared pre- and post-exposure rates using the same weights as above, revealed a small but statistically significant main effect of opioid use (HR=1.23; 95% CI, 1.06, 1.43), a significant main effect of benzodiazepine use (HR=2.55; 95% CI, 2.12, 3.05), and a significant but inverse interaction for combined use (HR=0.70; 95% CI, 0.55, 0.89; Table 3). That is, increases in risk of suicide attempts from pre- to post-initiation for combined use were actually lower than would be expected based (additively) upon the main effects for opioid and benzodiazepine use alone.
Table 3:
Hazard Ratios and 95% CIs from DiD Analysis
| Self-Harms (SH) | Drug Overdoses (DO) | SH +DO | Unintended Injuries (UI) | SH + DO + UI | Accidental Drug Overdose | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
|
||||||||||||
| Rx | HR | 95% CI | HR | 95% CI | HR | 95% CI | HR | 95% CI | HR | 95% CI | HR | 95% CI |
| Op | 0.99 | 0.80–1.21 | 1.29 | 1.07–1.56 | 1.23 | 1.06 | 0.90 | 0.82–0.98 | 0.98 | 0.90–1.06 | 1.47 | 1.34–1.61 |
| Bz | 2.23 | 1.73–2.88 | 2.76 | 2.20–3.45 | 2.55 | 2.12–3.05 | 1.41 | 1.21–1.63 | 1.71 | 1.52–1.92 | 1.91 | 1.68–2.17 |
| Op + Bz | 0.76 | 0.54–1.08 | 0.69 | 0.51–0.94 | 0.70 | 0.55–0.89 | 0.99 | 0.82–1.20 | 0.90 | 0.77–1.05 | 0.82 | 0.69–0.98 |
3.4. Secondary Outcomes
Table 3 also presents results of the difference in difference analysis for subsets of suicide attempt and intentional self-harm. For suicide attempt and intentional self-harm by overdose, associations were quite comparable to those for any suicidal event. In contrast, whereas the benzodiazepine use main effect persisted for non-overdose suicide attempt and intentional self-harm (HR=2.23; 95% CI, 1.73, 2.88), the opioid use main effect was null and not statistically significant (HR=0.99; 95% CI, 0.80, 1.21), see Table 3. The interaction for combined use was similar in magnitude to the primary analysis, although it was no longer statistically significant (HR=0.76; 95% CI, 0.54, 1.08). When we added injuries of undetermined intent to the outcome definition to capture additional possible suicide attempts and intentional self-harm events, we found a smaller benzodiazepine main effect (HR=1.71; 95% CI, 1.52, 1.92) and not statistically significant opioid (HR=0.98; 95% CI, 0.90, 1.06) and interaction (HR=0.90; 95% CI, 0.77, 1.05) effects (see Table 3). For accidental overdose, associations were quite comparable to those for suicide attempt and intentional self-harm by overdose (see Table 3). The main effect of opioids (HR=1.47; 95% CI, 1.34, 1.61), and the benzodiazepine main effect (HR=1.91; 95% CI, 0.68, 2.17) were both significant. The interaction for combined use was similar in magnitude to the primary analysis (HR=0.82; 95% CI, 0.69, 0.98), and significantly less than the simple additive effects of the two drugs (see Table 3).
3.5. Dosage Analysis
Table 4 presents results of the difference in difference analysis based on subsamples that had 1–90 days, 91–180 and > 180 days of exposure. The total number of patients included in the analyses for 1–90 days of exposure was 4,308,399 (opioid = 1,639,029, benzodiazepine = 315,548, combined use = 215,011, no use = 2,138,811 and 4,024 suicide attempt and intentional self-harm); 91–180 days of exposure 2,182,275 (opioid = 15,666, benzodiazepine = 25,888, combined use = 1,910, no use = 2,138,811 and 1,320 suicide attempt and intentional self-harm); >180 days of exposure 2,167,765 (opioid = 10,337, benzodiazepine = 14,708, combined use = 3,909, no use = 2,138,811 and 1,274 suicide attempt and intentional self-harm). Table 4 reveals no statistically significant increase in risk of suicide attempt and intentional self-harm pre- versus post-opioid exposure except in the 1–90 days analysis, although the number of events and population at risk were much smaller for the higher doses. For benzodiazepines, the pre- versus post-exposure difference was significant for all three dosages (e.g. > 180 days of exposure, HR=2.21; 95% CI, 1.25, 3.91). Although only statistically significant for the 1–90 days and 91–180 days analyses, the interaction effect was seen in similar (inverse) direction for all durations (e.g. 1–90 days of exposure, HR=0.55; 95% CI, 0.59, 0.99).
Table 4:
Hazard Ratios and 95% CIs for Dosage Analyses
(Outcome: Self-Harms + Drug Overdoses Combined)
| Pre Period |
Post Period |
|||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Exposure | Tx | # of Subjects | # of Events | PY | Rate/10K | # of Events | PY | Rate/10K | HR | 95% CI |
| >= 1 Day | Op | 16,65,032 | 1,476 | 24,67,139 | 5.98 | 1,556 | 18,92,702 | 8.22 | 1.23 | 1.06–1.43 |
| Bz | 3,56,144 | 253 | 4,18,188 | 6.05 | 618 | 3,61,150 | 17.12 | 2.55 | 2.12–3.05 | |
| Op+Bz | 2,20,830 | 280 | 2,65,502 | 10.54 | 596 | 2,31,902 | 25.71 | 0.70 | 0.55–0.89 | |
| No Rx | 21,38,811 | 1,641 | 30,92,396 | 5.31 | 1,657 | 27,68,126 | 5.99 | |||
| 1–90 Days | Op | 16,39,029 | 1,463 | 24,47,943 | 5.98 | 1,522 | 18,62,381 | 8.17 | 1.22 | 1.06–1.42 |
| Bz | 3,15,548 | 219 | 3,93,766 | 5.57 | 460 | 3,17,840 | 14.47 | 2.33 | 1.92–2.83 | |
| Op+Bz | 2,15,011 | 273 | 2,63,654 | 10.34 | 570 | 2,25,555 | 25.26 | 0.55 | 0.59–0.99 | |
| No Rx | 21,38,811 | 1,641 | 30,92,396 | 5.31 | 1,657 | 27,68,126 | 5.99 | |||
| 91–180 Days | Op | 15,666 | 4 | 12,865 | 3.49 | 19 | 17,327 | 11.05 | 2.88 | 0.73–11.36 |
| Bz | 25,888 | 21 | 16,531 | 12.46 | 89 | 25,582 | 34.68 | 2.56 | 1.63–4.02 | |
| Op+Bz | 1,910 | 3 | 732 | 46.81 | 9 | 1,696 | 55.15 | 0.15 | 0.02–0.90 | |
| No Rx | 21,38,811 | 1,641 | 30,92,396 | 5.31 | 1,657 | 27,68,126 | 5.99 | |||
| > 180 Days | Op | 10,337 | 8 | 6,330 | 13.03 | 15 | 12,993 | 11.23 | 0.79 | 0.24–2.56 |
| Bz | 14,708 | 13 | 7,892 | 16.53 | 70 | 17,728 | 39.43 | 2.21 | 1.25–3.91 | |
| Op+Bz | 3,909 | 4 | 1,115 | 32.58 | 17 | 4,651 | 36.92 | 0.62 | 0.11–3.45 | |
| No Rx | 21,38,811 | 1,641 | 30,92,396 | 5.31 | 1,657 | 27,68,126 | 5.99 | |||
4. Discussion
Results of this study revealed that, after initiation, concomitant treatment with both opioids and benzodiazepines was associated with greater suicide attempt and intentional self-harm risk relative to either drug alone or neither drug; however, the difference between benzodiazepines alone or in combination with opioids was not large (21 versus 29 per 10,000 adjusted). Critically, we applied a within-individual, difference in difference approach to examine change from before to after treatment initiation, which ruled out confounding from time-stable factors. In that design, increased risk among benzodiazepine initiators and, to a lesser extent, opioid initiators persisted. However, the statistical interaction term was opposite in direction to what would be predicted by a synergistic drug interaction: The increased rate of suicide attempt and intentional self-harm for the concomitant treatment group was 30% (relatively) lower than would be expected based on the sum of the two monotherapy drug group associations (i.e., opioids and benzodiazepines alone). However, Table 2 reveals that rates for all event categories are nominally additive in that the rate for the combined group are similar to the sum of the two monotherapies. As such the significant interaction, showing decreased risk relative to the expectation based on the sum of the individual risks, should not be viewed as beneficial in terms of reducing suicide risk.
The overall pattern of results was largely consistent across the different outcomes. Although associations for opioids appeared to be specific to suicide attempt and intentional self-harm by overdose, the opioid-benzodiazepine interaction term remained below 1 for all outcomes. The largest opioid main effect was for accidental overdose. This result is consistent with our prior within-individual analysis of prescribed opioid dose and (accidental) overdose during long-term opioid therapy (Quinn et al., in press). Interestingly, we again found no evidence of synergistic drug interaction for accidental overdose. Although we cannot rule out that such an interaction would be observed at higher doses than those typically used by patients in our cohort, our results are consistent with some prior research suggesting the risk of accidental overdose may reflect additive, rather than multiplicative, effects (Park et al, 2015).
These data and analyses reveal that although the rate of suicide attempt and intentional self-harm is substantially greater among opioid users who also take a benzodiazepine relative to opioid users alone, that difference may represent increased risk associated with benzodiazepines themselves rather than a drug interaction. That is, it is the benzodiazepine that appears to be associated with the greater risk of suicide attempt and intentional self-harm, whether or not it is taken with a prescription opioid, at least in commonly prescribed opioid quantities. Moreover, when we defined the dosage of exposure based on prescription treatment duration (1–90 days, 91–180 days, and > 180 days), we found no clear evidence of a dose-response relationship for either drug alone or in combination. Increases in treatment duration did not appear to explain the absence of a drug interaction.
Our results resonate with recent findings (Gibbons et al, 2019) showing increased risk of suicide attempt in patients taking alprazolam and diazepam following exposure relative to pre-exposure rates, in an analysis that simultaneously screened over 900 drugs for suicide attempt risk. The authors also found evidence for increased risk for the opioid analgesic combination products acetaminophen/hydrocodone bitartrate and codeine phosphate/promethazine hydrochloride; however, they did not adjust for concomitant benzodiazepine use. Taken together, this evidence supports recent calls for careful screening and regular reevaluation of patients prescribed benzodiazepines (Hirschtritt et al, 2021).
There are several limitations of this study. First, we do not have information on non-prescription users of opioids (e.g., diverted and misused opioid analgesics, heroin or other illicit opioids), and those individuals may have higher risk for suicide attempt and intentional self-harm than those opioid users who filled prescriptions in our study. Second, our data are based on individuals with private health insurance, different suicidal event rates and opioid use rates would be expected in patients with public health insurance (e.g., Medicaid). Third, we defined medication exposure on the basis of filled prescriptions, rather than on prescriptions written by providers or taken by patients. To the extent that some prescriptions were filled but not taken (or, alternatively, taken intermittently on an as-needed basis), exposure misclassification may have biased our results toward the null. Fourth, our definition of suicide attempt and intentional self-harm is based on medical claims for treatment for suicide attempt and intentional self-harm and is likely to miss many events that were not clinically recognized. Similarly, suicide deaths that did not result in a medical claim are not included in our analysis, again leading to underestimates of suicide attempt and intentional self-harm. In the claims data, we have no way of differentiating fatal from non-fatal suicide attempts or intentional self-harm events. Fifth, analyses span the ICD-9 and ICD-10 coding periods, during which there was a substantial change in all codes, including codes for suicide attempts and intentional self-harm events. This can lead to increased variability in the data and increased uncertainty in our estimated treatment-related effects. Sixth, our data are from 2014–2017, and these results should be confirmed using newer data as they become available. Finally, further research is needed to determine whether interactions may vary for specific opioid and benzodiazepine drug combinations.
There are several strengths of our study. First, we were able to identify a cohort of over 4 million people based on filled prescriptions, allowing us to have a sufficient number of suicide attempt and intentional self-harm for a meaningful analysis. Second, we studied a new user cohort in which we were able to exclude patients with recorded prior opioid and benzodiazepine use for a period of two years. Third, we were able to provide propensity score weighting to balance all four groups simultaneously on a large number of potential confounders. Of course, bias can still remain due to unmeasured confounders that were not included in our propensity score. However, our difference in difference analysis is within-subject and therefore is robust to residual between-subject differences. For example, by directly comparing post-exposure risk to pre-exposure risk, it implicitly adjusted for patient characteristics that are associated with increased pre-exposure suicidal risk and not the actual effect of these medications. We note, though, that our approach does not rule out time-varying confounding.
In summary, our study reveals that the observed elevated risk of suicide attempts and intentional self-harm for the concomitant prescription of opioids and benzodiazepines is driven in large part by benzodiazepines and to a lesser extent by opioids. Although our results should not be interpreted as necessarily relevant to accidental overdose or other adverse opioid-related outcomes, they do not support the hypothesis of a clinically recognizable synergistic opioid/benzodiazepine drug interaction at typically prescribed dosages and treatment durations with regard to suicide attempts and intentional self-harm.
Supplementary Material
Highlights.
Concomitant opioid and benzodiazepine use is associated with increased risk of suicide attempt and self-harm, but significantly less than is expected by the sum of the individual drug risks.
Examined suicide risk and self-harm by overdose and did not find evidence of a drug interaction.
Increased risk of suicide attempts and self-harm is driven predominantly by benzodiazepines.
Acknowledgments:
All data used in this study were obtained from Truven health as a part of their MarketScan database under license to the University of Chicago. Robert Gibbons of the University of Chicago had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Robert Gibbons and Kwan Hur of the University of Chicago performed the statistical analyses of these data.
Role of Funding Source: Research reported in this publication was supported by the National Institute of Mental Health of the National Institutes of Health [grant number R01MH100155] and by the National Institute On Drug Abuse of the National Institutes of Health [grant number R00DA040727]. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
IRB Approval: This study was deemed exempt from review by the University of Chicago institutional review board.
Dissemination Declaration: Dissemination to study participants is not applicable.
Patient and Public Involvement: It was not appropriate or possible to involve patients or the public in the design, or conduct, or reporting, or dissemination plans of our research.
Conflict of Interests: Dr. Gibbons has been an expert witness for the US Department of Justice, Merck, Glaxo-Smith-Kline, Pfizer and Wyeth and is a founder of Adaptive Testing Technologies, which distributes the CAT-MH™ battery of adaptive tests. The terms of this arrangement have been reviewed and approved by the University of Chicago in accordance with its conflict of interest policies. No support from any organization for the submitted work.
References
- 1.CDC (Centers for Disease Control and Prevention), 2018. Opioid data analysis and resources https://www.cdc.gov/drugoverdose/data/analysis.html (accessed January 17, 2020).
- 2.NAS (National Academies of Sciences, Engineering, and Medicine), 2019. An Approach to Evaluate the Effects of Concomitant Prescribing of Opioids and Benzodiazepines on Veteran Deaths and Suicides Washington, DC: The National Academies Press. [PubMed] [Google Scholar]
- 3.INCB (International Narcotics Control Board). Narcotic drugs: Estimated world requirements for 2018; Statistics for 2016 In. New York: United Nations; 2018. [Google Scholar]
- 4.Olfson M, Crystal S, Wall M, et al. , 2018. Causes of death after nonfatal opioid overdose. JAMA Psychiatry 75(8):820–827. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Samples H, Stuart EA, Olfson M, 2019. Opioid use and misuse and suicidal behaviors in a nationally representative sample of US adults. Am J Epidemiol 188(7):1245–1253. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Bohnert ASB, Ilgen MA, 2019. Understanding links among opioid use, overdose, and suicide. N Engl J Med 380(1):71–79. [DOI] [PubMed] [Google Scholar]
- 7.Ilgen MA, Bohnert ASB, Ganoczy D, et al. , 2016. Opioid dose and risk of suicide. Pain 157(5):1079–1084. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Hayes CJ, Krebs EE, Hudson T, et al. , 2020. Impact of opioid dose escalation on the development of substance use disorders, accidents, self-inflicted injuries, opioid overdoses and alcohol and non-opioid drug-related overdoses: a retrospective cohort study. Addiction 115(6):1098–1112. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Brent DA, Hur K, Gibbons RD, 2019. Association between parental medical claims for opioid prescriptions and risk of suicide attempt by their children. JAMA Psychiatry 76(9):1–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Olfson M, King M, Schoenbaum M, 2015. Benzodiazepine use in the United States. JAMA Psychiatry 72(2):136–142. [DOI] [PubMed] [Google Scholar]
- 11.Agarwal SD, Landon BE, 2019. Patterns in outpatient benzodiazepine prescribing in the United States. JAMA Network Open 2(1):e187399. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Cato V, Holländare F, Nordenskjöld, et al. , 2019. Association between benzodiazepines and suicide risk: a matched case-control study. BMC Psychiatry 19:317. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Gibbons RD, Hur K, Lavigne J, et al. , 2019. Medications and suicide: High dimensional empirical Bayes screening (iDEAS). Harvard Data Science Review 1(2):epub. [Google Scholar]
- 14.NIDA (National Institute on Drug Abuse), 2018. Benzodiazepines and opioids https://www.drugabuse.gov/drugs-abuse/opioids/benzodiazepines-opioids (accessed January 17, 2020).
- 15.Jann M, Kennedy WK, Lopez G, 2014. Benzodiazepines: A major component in unintentional prescription drug overdoses with opioid analgesics. J Pharm Pract 27(1):5–16. [DOI] [PubMed] [Google Scholar]
- 16.White JM, Irvine RJ, 1999. Mechanisms of fatal opioid overdose. Addiction 94(7):961–972. [PubMed] [Google Scholar]
- 17.Sun EC, Dixit A, Humphreys K, et al. , 2017. Association between concurrent use of prescription opioids and benzodiazepines and overdose: Retrospective analysis. BMJ 356:j760. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Park TW, Saitz R, Ganoczy D, et al. , 2015. Benzodiazepine prescribing patterns and deaths from drug overdose among US veterans receiving opioid analgesics: Case-cohort study. BMJ 350:h2698. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Dasgupta N, Funk MJ, Proescholdbell S, et al. , 2016. Cohort study of the impact of high-dose opioid analgesics on overdose mortality. Pain Med 17(1):85–98. [DOI] [PubMed] [Google Scholar]
- 20.Garg RK, Fulton-Kehoe D, Franklin GM, 2017. Patterns of opioid use and risk of opioid overdose death among Medicaid patients. Med Care 55(7):661–668. [DOI] [PubMed] [Google Scholar]
- 21.Hawkins EJ, Goldberg SB, Malte CA, et al. , 2019. New coprescription of opioids and benzodiazepines and mortality among Veterans Affairs patients with posttraumatic stress disorder. J Clin Psychiatry 80(4):18m12689. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Xu KY, Hartz SM, Borodovsky JT, et al. , 2020. Association between benzodiazepine use with or without opioid use and all-cause mortality in the United States, 1999–2015. JAMA Network Open 3(12):e2028557. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Dowell D, Haegerich TM, Chou R, 2016. CDC guideline for prescribing opioids for chronic pain — United States, 2016. MMWR Recomm Rep 65(1):1–49. [DOI] [PubMed] [Google Scholar]
- 24.Ilgen M, 2018. Pain, opioids, and suicide mortality in the United States. Ann Intern Med 169(7):498–499. [DOI] [PubMed] [Google Scholar]
- 25.Racine M, 2018. Chronic pain and suicide risk: A comprehensive review. Prog Neuropsychopharmacol Biol Psychiatry 87:269–280. [DOI] [PubMed] [Google Scholar]
- 26.Turner BJ, Liang Y, 2015. Drug overdose in a retrospective cohort with non-cancer pain treated with opioids, antidepressants, and/or sedative-hypnotics: Interactions with mental health disorders. J Gen Intern Med 30(8):1081–1096. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Hansen L, 2016. MarketScan white paper: The MarketScan databases for life sciences researchers Ann Arbor, MI: Truven Health Analytics. [Google Scholar]
- 28.Morgan SL, Todd JL, 2008. A diagnostic routine for the detection of consequential heterogeneity of causal effects. Sociological Methodology 38:231–281. [Google Scholar]
- 29.Rosenbaum PR, 1987. Model-based direct adjustment. The Journal of the American Statistician 82:387–394. [Google Scholar]
- 30.Lunceford JK, Davidian M, 2004. Stratification and weighting via the propensity score in estimation of causal treatment effects: A comparative study. Statistics in Medicine 23:2937–2960. [DOI] [PubMed] [Google Scholar]
- 31.Robins JM, Hernan MA, Brumback B, 2000. Marginal structural models and causal inference in Epidemiology. Epidemiology 11:550–560. [DOI] [PubMed] [Google Scholar]
- 32.Cole SR, Hernan MA, 2008. Constructing inverse probability weights for marginal structural models. Am J Epidemiol 168: 656–664. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Angrist JD, Pischke JS, 2008. Mostly Harmless Econometrics: An Empiricist’s Companion Princeton University Press. 227–243. [Google Scholar]
- 34.Man KKC, Coghill D, Chan EW, et al. , 2017. Association of risk of suicide attempts with methylphenidate treatment. JAMA Psychiatry 74(10):1048–1055. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Lin DY, Wei LJ, 1989. The Robust Inference for the Cox Proportional Hazards Model. Journal of the American Statistical Association 84(408):1074–078. [Google Scholar]
- 36.Bohnert ASB, McCarthy JF, Ignacio RV, et al. , 2013. Misclassification of suicide deaths: Examining the psychiatric history of overdose decedents. Inj Prev 19(5):326–330. [DOI] [PubMed] [Google Scholar]
- 37.Quinn PD, Chang Z, Bair MJ, et al. Associations of Opioid Prescription Dose and Discontinuation with Risk of Substance-Related Morbidity in Long-Term Opioid Therapy. Pain in press. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Hirschtritt ME, Olfson M, Kroenke K, 2021. Balancing the risks and benefits of benzodiazepines. JAMA 325(4):347–348. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.