A major societal shift in the past two decades has been the substantial increase in the use of pharmaceutical opioids, Reference Kolodny, Courtwright, Hwang, Kreiner, Eadie and Clark1 especially in the USA, Reference Compton and Volkow2 Canada, the UK and Australia. Reference Blanch, Pearson and Haber3,Reference Gisev, Campbell, Lalic, Larney, Peacock and Nielsen4 There have been concerns that greater availability of pharmaceutical opioids may provide increased access to a means of self-inflicted (potentially lethal) harm. Reference Ilgen, Bohnert, Ganoczy, Bair, McCarthy and Blow5,Reference Miller, Swedler, Lawrence, Ali, Rockett and Carlson6 A recent ecological study found that increases in opioid use, higher opioid doses and long-term opioid use were associated with an increase in overall suicide and suicide by opioid poisoning. Reference Braden, Edlund and Sullivan7,Reference Olfson, Waidmann, King, Pancini and Schoenbaum8 This is supported by studies using individual-level data, which found an increased risk of suicide and/or self-harm hospitalisations associated with filling an opioid prescription, Reference Olfson, Waidmann, King, Pancini and Schoenbaum8 higher opioid doses Reference Ilgen, Bohnert, Ganoczy, Bair, McCarthy and Blow5,Reference Ilgen9,Reference Luo, Chen, Doshi, Rickles, Chen and Schwartz10 and longer duration. Reference Ilgen, Bohnert, Ganoczy, Bair, McCarthy and Blow5,Reference Olfson, Waidmann, King, Pancini and Schoenbaum8–Reference Luo, Chen, Doshi, Rickles, Chen and Schwartz10 Previous studies are limited, however, with either a focus on only one aspect of opioid prescribing, i.e. dose, limited controlling for known confounders, such as mental health and previous self-harm, or on specific sub-populations, such as veterans. Reference Ilgen, Bohnert, Ganoczy, Bair, McCarthy and Blow5 Additionally, previous studies have examined only the association between opioid use and self-harm presentations or suicide overall, or those related to opioid poisonings, with few studies examining the relationship between different self-harm and suicide methods. Reference Ilgen, Bohnert, Ganoczy, Bair, McCarthy and Blow5 It is important to consider other means of self-harm and suicide, beyond opioid involvement because, in the broader substance use literature, most suicides among people with substance use disorders do not involve overdose, and other methods such as the use of firearms and hanging are more commonly used, despite having access to potentially lethal methods. Reference Ilgen, Bohnert, Ganoczy, Bair, McCarthy and Blow5 It is important to understand whether, among people prescribed opioids, there is an increased risk for self-harm and suicide via methods beyond opioid poisoning.
The current study utilises linked individual-level data for a retrospective population-based cohort of all adult residents in New South Wales (NSW), Australia’s most populous state, and who initiated at least one prescription opioid treatment episode in the period 1 July 2003 to 31 December 2018. We examine the association between opioid use (on/off, cumulative duration and daily dose) and self-harm and suicide-related behaviours, and the method of self-harm hospitalisations and suicide, controlling for known risk factors.
Method
Study design and study population
This retrospective population-based cohort study comprised NSW adult residents who initiated a subsidised prescription opioid analgesic between 1 July 2003 and 31 December 2018. Prescription opioid analgesics were identified through Pharmaceutical Benefits Scheme (PBS) data and included buprenorphine, codeine, dextropropoxyphene, fentanyl, hydromorphone, methadone, morphine, oxycodone (±naloxone), pethidine, tapentadol and tramadol. Dispensing of methadone or buprenorphine formulations for the treatment of opioid dependence is not recorded in the PBS data-set and was therefore excluded. Cohort entry was defined as the date of the first dispensing of a subsidised prescription opioid, operationalised as no opioid dispensings in the previous 365 days. Each person was followed up until the end of the study (31 December 2018), date of death or date of the first palliative care indication, whichever occurred first. People who were receiving palliative care at cohort entry were excluded due to having a different opioid risk profile to other indications (see Appendix A available at https://doi.org/10.1192/bjp.2025.10436).
Due to variation in the completeness of PBS data capture during the study period, for the period between 1 July 2003 and 30 June 2013, only people who were concessional beneficiaries (i.e. those earning a low income and eligible for additional medicine subsidies) were included. For the period 1 July 2013 to 31 December 2018, both concessional and general beneficiaries were included. Our previous work has found minimal variation in cohort characteristics when defining the cohort based on beneficiary status. For further details please see the POPPY II cohort profile. Reference Gisev, Pearson, Dobbins, Buizen, Murphy and Wilson11
Data sources
We used data from the POPPY II study. Reference Gisev, Pearson, Dobbins, Currow, Blyth and Larney12 Full details of the study, setting, data sources, linkage and overall characteristics of the cohort are reported previously. Reference Gisev, Pearson, Dobbins, Buizen, Murphy and Wilson11 Briefly, the POPPY II cohort is comprised of 3.57 million adults who initiated a prescription opioid between 2003 and 2018 in NSW. Just over half were female and 1 in 4 were aged ≥65 years at the time of cohort entry. Reference Gisev, Pearson, Dobbins, Buizen, Murphy and Wilson11 The cohort was defined from PBS data and linked to ten national and state data-sets. In the current study, the linked data-sets included: (a) the NSW Admitted Patient Data Collection (APDC), which records in-patient separations from public hospitals (including psychiatric), multipurpose health services, private hospitals and private day procedure centres in NSW; (b) the NSW Controlled Drugs Data Collection (CoDDaC), which records all NSW recipients of opioid agonist therapy (methadone or buprenorphine treatment for opioid dependence); (c) the NSW Ambulatory Mental Health Dataset (MH-AMB), which records non-admitted mental healthcare services including mental health day programmes, outreach services, community health service contacts and out-patient psychiatric contacts; and (d) the National Death Index (NDI), which includes death registrations along with underlying and contributing causes of death. Diagnoses recorded in APDC and MH-AMB are coded according to ICD-10-AM. Medicines in the PBS data were coded according to Anatomical Therapeutic Chemical (ATC) codes 13,14 (see Appendix A).
Exposures
We define opioid exposure using the following three time-varying measures.
‘On/off opioid’
Each person’s follow-up period was divided into opioid exposure intervals – namely current, recent and former use, using the Individualised Dispensing Patterns (IDP) method. Reference Bharat, Degenhardt, Pearson, Buizen, Wilson and Dobbins15 This method uses information from an individual’s pattern of prior dispensings to estimate the expected duration for a dispensed opioid supply to be exhausted. The ‘recent’ exposure period represents a 7-day period following the expected duration of supply or until the next dispensing (whichever occurred earlier), to mitigate the impact of brief interruptions in supply. The ‘former’ exposure interval was the period without new opioid dispensings. For the main analysis of this study, we combine the current and recent exposure intervals to define a period during which a person was ‘on opioids’, while the former exposure interval is categorised as periods ‘off opioids’.
Cumulative duration of opioid exposure
We calculated the cumulative number of days that a person was on opioids in the past 180 days as a time-varying measure. To account for potential variability in the amount of opioid exposure over these 180 days (i.e. frequency, strength and quantity of dispensings), we divided each on/off interval into 7-day periods and, at the beginning of each of these periods, we calculated cumulative duration of exposure in the 180-day look-back. For periods shorter than 7 days, we adjusted the number of days of the cumulative duration interval accordingly. The past 180 days were selected based on previous research, Reference Luo, Chen, Doshi, Rickles, Chen and Schwartz10,Reference Karmali, Bush, Raman, Campbell, Skinner and Roberts16 where long-term opioid therapy is often defined as 3–6 months of continuous exposure. Cumulative duration of exposure was categorised as no exposure, 1–30 days, 31–60 days, 61–90 days, 91–120 days, 121–150 days and 151–180 days.
Daily opioid dose
For ‘on opioid’ periods, an estimated daily opioid dose was calculated and expressed as total oral morphine equivalents (OME) milligrams per day, multiplying the relative strength of the opioids dispensed using published conversion factors. Reference Nielsen, Degenhardt, Hoban and Gisev17 The resulting estimated OME values were grouped into the following predefined categories: Reference Campbell, Noghrehchi, Nielsen, Clare, Bruno and Lintzeris18,Reference Dowell, Haegerich and Chou19 1–49 OME, 50–89 OME, 90–199 OME and ≥200 OME mg.
Outcomes
Because non-fatal self-harm is one of the most well-established risk factors for suicide, Reference Bostwick, Pabbati, Geske and McKean20 the current study considered both non-fatal self-harm and suicide as outcomes.
Self-harm hospitalisation, herein referred to as self-harm, was identified from principal and secondary diagnoses in records of hospital admissions occurring during the follow-up period. Suicide mortality, herein referred to as suicide, was identified from underlying and contributing causes of death fields in mortality data. We adapted previously defined categories to characterise the method of self-harm and suicide, Reference Colledge-Frisby, Jones, Degenhardt, Hickman, Padmanathan and Santo21,Reference Santo, Bharat, Colledge-Frisby, Chrzanowska, Man and Moran22 which included intentional poisoning involving opioids, intentional poisoning involving non-opioid substances and other means (which could include hanging, exposure to gas, firearms and drowning; see Appendices B and C for ICD-10-AM diagnostic codes). We assessed self-harm hospitalisations and suicide deaths separately.
Covariates
A full description of how each covariate was defined is presented in Appendix A.
Demographic
Age at cohort entry was categorised as 18–34, 35–44, 45–54, 55–64, 65–74, 75 years and older. Geographical remoteness was determined based on each person’s postcode of residence at cohort entry. 23 Postcode-based socioeconomic disadvantage quintiles at cohort entry for each person were categorised using the Index of Relative Socio-Economic Disadvantage (IRSD). Reference Pink24
Clinical variables
We measured the following medical conditions, creating time-varying covariates updated daily during follow-up to indicate where a related medicine dispensation, hospital presentation or contact with mental health services had occurred within the previous 12 months. These included anxiety or depression, severe mental health disorders (schizophrenia, manic episodes, bipolar affective disorder, psychosis), non-opioid substance use disorders, cancer and self-harm. Opioid use disorders were also defined based on evidence of any opioid agonist treatment.
Statistical analysis
All analyses were performed using SAS 9.4. Descriptive statistics include frequency (%) and mean (s.d.) to summarise the characteristics of the cohort at baseline. We calculated crude rates of self-harm and suicide deaths, overall and method-specific, with 95% confidence intervals per 100 000 person-years for each measure of opioid exposure (on/off, cumulative duration and dose). Duration of follow-up is calculated as the number of days from cohort entry to study end-point, then converted to years of follow-up (days divided by 365.25).
For self-harm outcomes, we used generalised estimating equations (GEE, Poisson distribution, offset term as the log of duration of time-varying interval) to estimate crude and adjusted incidence rate ratios (IRRs) and 95% confidence intervals. GEE models adjust for multiple observations per individual using a working correlation structure, providing robust parameter estimates.
For suicide outcomes, we used time-varying Cox proportional hazard models to calculate crude and adjusted hazard ratios and 95% confidence intervals. Schoenfeld residuals were used to test the proportional hazards assumption.
All models adjusted for the above-described covariates.
To assess the robustness of findings to the simplified on/off definition of exposure, we conducted sensitivity analyses in which we examined the association between opioid exposure as current, recent and former intervals and outcomes of self-harm and suicide (Appendices D and E).
Ethical standards
The overall POPPY II study received full ethical approval from the Australian Institute of Health and Welfare (AIHW) Ethics Committee (no. EO2016/4/314), NSW Population and Health Services Research Committee (no. 2017/HRE0208), the ACT Health Human Research Ethics Committee (no. ETHLR.18.094) and the ACT Calvary Public Hospital Bruce Ethics Committee (no. 5-2019). A waiver of the requirement to obtain consent was granted by the reviewing ethics committees due to the nature of the data (i.e. population-based data linkage). The current study obtained ethics approval from the University of New South Wales Human Research Ethics Committee (no. HC200700).
Results
Participant characteristics
There were 3 268 282 individuals included in the current study (Table 1). Over half were female (53.1%, n = 1 735 126), with a mean age at cohort entry of 51.25 years (s.d. 19.70). One in 5 had evidence of anxiety or depression (20.7%, n = 675 607), 3.3% cancer (n = 108 977) and 2.9% a severe mental disorder (n = 94 754) in the 12 months prior to cohort entry.
Cohort characteristics at cohort entry (N = 3 268 282)
a. Classified using the Index of Relative Socioeconomic Disadvantage, Census 2011.
b. Classified using the Remoteness Area indices, Census 2016.
c. Derived from in-patient hospital data and community mental health services using ICD-10-AM codes; pharmaceutical dispensing data using Anatomical Therapeutic Chemical (ATC) codes; and authorised prescription of opioid agonist therapies in opioid treatment services. See Appendix C for detailed descriptions of covariate definition, operationalisation, ICD-10-AM and ATC codes.
Opioid exposure
In this cohort of 3 268 282 opioid-exposed individuals, there was a total of 18 758 134 person-years of follow-up. The total number of person-years on opioids was 2 924 705, compared with a total of 15 833 429 off opioids. For cumulative exposure, there was a total number of 14 081 208 person-years for no exposure, with 1–30 days the next most common category (1 042 378 person-years), followed by 151–180 days (943 801 person-years). For cumulative exposure categories between 31 and 150 days, the number of person-years declined as the number of cumulative exposure days increased. For daily opioid dose during on opioid periods, the lowest dose, 1–49 OME mg/day, had the greatest number of person-years (2 424 995). The number of person-years declined as the dose increased. For doses ≥200 OME mg/day there were 82 243 person-years (see Appendix F).
Self-harm hospitalisations
During the study period, there were 49 215 self-harm events at a crude rate of 262.4 per 100 000 person-years (95% CI: 260.1–264.7). The most common method of self-harm involved non-opioid poisoning, accounting for 58% of all self-harm events (n = 28,736, crude rate 153.2 per 100 000 person-years, 95% CI: 151.4–255.0); 25% involved methods other than poisoning (n = 12,506, crude rate 66.7 per 100 000 person-years, 95% CI: 65.5–67.8). Opioid poisonings were less common, accounting for 16% of all self-harm events (n = 7973, crude rate 42.5 per 100 000 person-years, 95% CI: 41.6–43.4; see Appendix F).
Periods on/off opioids
The crude rate for self-harm by any method was 229.5 (95% CI: 227.2–231.9) per 100 000 person-years for periods off opioids, and 440.2 (95% CI: 432.7–447.9) per 100 000 person-years for periods on opioids. Following multivariable adjustment, periods of opioid use (on opioids) were associated with an increased risk of self-harm, overall and by all methods (Table 2). The association was strongest for self-harm involving opioid poisoning, with a crude rate of 120.5 per 100 000 person-years (95% CI: 116.5–124.5) for periods on opioids, compared with a crude rate of 28.1 per 100 000 person-years (95% CI: 27.3–28.9) for periods off opioids (adjusted IRR 3.71, 95% CI: 3.52–3.91).
Self-harm associated with opioid exposure: crude rates per 100,000 person-years, crude unadjusted and adjusted incidence rate ratios (IRRs) and 95% confidence intervals
Cumulative opioid exposure
In adjusted models, the rate for all methods of self-harm was higher for longer cumulative exposure periods compared with no exposure (Table 2). The only exception was for self-harm involving non-opioid substances, where no association was observed with the cumulative exposure of 151–180 days (Table 2).
The strongest associations between cumulative exposure and the different methods of self-harm related to self-harm involving opioids (Table 2). For example, for self-harm involving opioids, a cumulative exposure of 151–180 days was associated with a significantly increased risk (crude rate 154.3 per 1000 000 person-years, 95% CI: 146.4–162.4), compared with no exposure (crude rate 26.5 per 100 000 person-years, 95% CI: 25.7–27.4; adjusted IRR 4.42, 95% CI: 4.09–4.79). There was a clear dose–response correlation following 61–90 days of exposure, with risk for self-harm involving opioids increasing as cumulative duration increased.
Opioid dose
In adjusted models, higher opioid doses were positively associated with self-harm involving opioid poisoning, but not with any other method-specific self-harm. Risk of self-harm was found to be higher at the highest dose (≥200 mg/day, crude rate 554.5 per 100 000 person-years, 95% CI: 504.7–607.8) than at the lowest dose category (1–49 OME mg/day, crude rate 87.4, 95% CI: 83.7–91.2; IRR 2.23, 95% CI: 1.97–2.53]; Table 2).
Other covariates
In adjusted models, previous 12-month self-harm hospitalisation was most strongly associated with all methods of self-harm. Other significant predictors included younger age, being female (except for self-harm by other methods), past 12-month hospitalisation for anxiety, depression or severe mental disorders and past 12-month evidence of opioid use disorder or substance use disorder (see Appendix G).
Suicide
There were 3086 suicide deaths during the study period, at a crude mortality rate of 16.5 per 100 000 person-years (95% CI: 15.9–17.0; see Appendix H).
The majority of suicides were via methods other than substance poisoning (i.e. involving hanging, gas, drowning or firearms; 81%, n = 2,500, crude rate 13.3 per 100 00 person-years, 95% CI: 12.8–13.9). Of all suicides, 10% involved non-opioid substances (n = 330, crude rate 1.8 per 100 000 person-years, 95% CI: 1.6–2.0) and 8% involved opioids (n = 256, crude rate 1.4 per 100 000 person-years, 95% CI: 1.2–1.5; see Appendix H).
Periods on/off opioids
The crude rate for suicide by any method was 25.0 (95% CI: 23.2–26.9) during periods of opioid use (on opioids), compared with 14.9 per 100 000 person-years (95% CI: 14.3–15.5) when off opioids (adjusted hazard ratio 1.52, 95% CI: 1.39–1.65; Table 3). When adjusted for covariates, associations were strongest between opioid exposure and suicide involving opioids.
Suicide associated with opioid exposure: crude rates per 100,000 person-years, unadjusted and adjusted hazard ratios and 95% confidence intervals
Cumulative duration
In adjusted models, a longer cumulative duration of opioid use was associated with a greater risk of suicide death by opioid poisoning (Table 3). For example, a cumulative exposure of 151–180 days, compared with no exposure, was associated with a 7-fold increased risk of suicide by opioid poisoning (crude rate 8.3, 95% CI: 6.5–10.3, versus crude rate 0.7, 95% CI: 0.6–0.9; adjusted hazard ratio 6.64, 95% CI: 4.90–9.00).
There was no association between cumulative duration and suicide involving other substances in adjusted models. There was a slightly increased risk for cumulative duration and suicide involving other methods, although there was no clear dose response (Table 3).
Opioid dose
Higher opioid doses were associated with an increased risk of any suicide and suicide involving opioid poisonings (Table 3). At the highest dose, ≥200 OME mg/day, there was a near 8-fold increased risk of suicide deaths involving opioid poisoning (crude rate 35.3 per 100 000 person-years, 95% CI: 23.6–50.6, versus crude rate 2.0 per 100 000 person-years, 95% CI: 1.5–2.7; adjusted hazard ratio 7.93, 95% CI: 4.74–13.26). There was no association between opioid dose and suicide deaths involving other methods in adjusted models.
Other covariates
In all adjusted models, other risk factors associated with increased suicide risk included younger age, male gender, past 12-month evidence of anxiety, depression, severe mental disorders, substance use disorder and self-harm hospitalisation (see Appendix I).
Sensitivity analysis
Our analyses, using opioid exposure defined as current, recent and former (as opposed to our main analysis combining current and recent into an ‘on-opioid’ interval), yielded adjusted IRRs for self-harm and adjusted hazard ratios for suicide that were similar to those from the main analyses (see Appendices D and E).
Discussion
The current study aimed to describe opioid prescribing characteristics with respect to opioid exposure, including on/off periods of use, cumulative duration and dose, and their relationships to self-harm and suicide. We considered both self-harm and suicide as outcomes because non-fatal self-harm is one of the most well-established risk factors for suicide. Reference Bostwick, Pabbati, Geske and McKean20 The cohort spent greater periods not exposed (off) to opioids (15 833 429 person-years) compared with periods exposed (on) to opioids (2 924 705 person-years). Additionally, consistent with our previous work, Reference Gisev, Buizen, Hopkins, Schaffer, Daniels and Bharat25 during periods on opioids the cohort was most likely to be exposed to opioids for less than 30 days and at the lowest dose (1–49 OME mg/day), suggesting that most people prescribed opioids have low, time-limited exposure. Among the cohort, during periods of opioid exposure overall self-harm events (crude rate 440 per 100 000 person-years, 95% CI: 432.7–447.9) and suicide (crude mortality rate of 25.0 per 100 00 person-years, 95% CI: 23.2–26.9) were more common than in the general population (crude rate 116.3 and 12.3 per 100 000, respectively). 26
There has been concern that the increased rates of prescribing opioids have been associated with an increase in self-harm and suicide due to increased access to highly lethal means. Reference Olfson, Waidmann, King, Pancini and Schoenbaum8 A recent ecological study found that, in US regions where there had been declines in opioid prescribing, there were associated decreases in overall suicide deaths, suggesting a link between prescription opioids and suicide, potentially through access to means. Reference Olfson, Waidmann, King, Pancini and Schoenbaum8 In the current study we found that the most common method of self-harm hospitalisations involved non-opioid substances (58%), and most suicides (81%) were via other means (i.e. involving hanging, gas, drowning or firearms). Opioid-related self-harm and suicide were the least common methods (16% for self-harm hospitalisations and 8% for suicide). These findings suggest that the relationship between opioid prescribing, self-harm and suicide may be more complex than simple access to means. If it were a simple relationship, we would expect opioid-related self-harm and suicide to be more common in the current cohort. Despite this, for relatively small proportions of self-harm and suicide involving opioids, we found a clear association between exposure to opioids, longer cumulative duration and higher daily opioid dose, even after adjusting for known risk factors. These findings are consistent with previous research Reference Ilgen, Bohnert, Ganoczy, Bair, McCarthy and Blow5,Reference Olfson, Waidmann, King, Pancini and Schoenbaum8-Reference Luo, Chen, Doshi, Rickles, Chen and Schwartz10,Reference Oliva, Bowe, Manhapra, Kertesz, Hah and Henderson27 and indicate that, for some people, exposure to greater doses and duration are associated with an increased risk of opioid-related self-harm and suicide.
We also considered the association between opioid prescribing characteristics and non-opioid-related self-harm and suicide means, namely, non-opioid poisonings and other means (i.e. involving hanging, gas, drowning or firearms). The broader substance use literature indicates that most suicides among people with substance use disorders do not involve overdose, Reference Ilgen, Conner, Valenstein, Austin and Blow28 i.e. that other methods such as firearms or hanging are used, despite having access to potentially lethal means. Additionally, a previous study found opioid dose was associated with an increased risk for suicide by means other than opioid poisoning. Reference Ilgen, Bohnert, Ganoczy, Bair, McCarthy and Blow5 With few exceptions, in unadjusted models we found that opioid exposure, higher doses and longer duration were associated with an increased risk of non-opioid poisonings and ‘other’ mechanisms of self-harm and suicide. However, after controlling for known suicide risk factors, we found that, although associations for opioid exposure remained, the findings relating to cumulative exposure and opioid dose were less consistent than those found for opioid poisonings, with no clear dose–response. We were able to control for a wide range of covariates, including mental health (identified through medicine dispensing and hospitalisations) and previous self-harm hospitalisations. These were associated with the greatest increased risk for self-harm and suicide in adjusted models. The inclusion of these comprehensive covariates may explain why the current study did not find a strong link between opioid characteristics and non-opioid self-harm and suicide, as has been found in previous studies. Reference Ilgen, Bohnert, Ganoczy, Bair, McCarthy and Blow5,Reference Olfson, Waidmann, King, Pancini and Schoenbaum8
In interpreting our findings, particularly for opioid-related self-harm and suicide, it is important to note that we do not suggest that the use of opioids is causally related to self-harm and suicide. People prescribed opioids for a long duration and at higher daily OME Reference Campbell, Nielsen, Bruno, Lintzeris, Cohen and Hall29–Reference Deyo, Hallvik, Hildebran, Marino, Dexter and Irvine30 typically have chronic pain and complex comorbidities, such as co-occurring mental health disorders, Reference Gisev, Buizen, Hopkins, Schaffer, Daniels and Bharat25 and a poorer quality of life. Reference Campbell, Nielsen, Bruno, Lintzeris, Cohen and Hall29 Additionally, chronic pain is independently associated with suicide-related behaviours, over and above mental health conditions. Reference Campbell, Bruno, Darke, Shand, Hall and Farrell31,Reference Racine32 It is generally uncommon for people to be prescribed opioids for a long duration and at high doses. Reference Deyo, Hallvik, Hildebran, Marino, Dexter and Irvine30 For the minority of people exposed to long-term, high-dose prescription opioids, continual monitoring and screening for suicide-related behaviours, including known risk factors such as a previous self-harm attempt and co-occurring mental health, should be considered.
In situations where a person is on long-term, high-dose opioids with several suicide-related risk factors, a clinician may consider tapering or withdrawing them from opioids to reduce access to means. This approach has been cautioned, due to the potential impact that such an approach may have on short-term suicide risk, Reference Larochelle, Lodi, Yan, Clothier, Goldsmith and Bohnert33 and the potential to increase suicide-related behaviours due to the experience of opioid withdrawal, increased pain and decreased physical functioning. Reference Ilgen9,Reference Larochelle, Lodi, Yan, Clothier, Goldsmith and Bohnert33 An alternative option is to consider rotation to opioid agonist treatment (OAT), such as buprenorphine. A recent review found that rotation to buprenorphine reduced chronic pain intensity without precipitating withdrawal in people with chronic pain on long-term, high-dose opioids. Reference Powell, Rosenberg, Yaganti, Garpestad, Lagisetty and Shannon34 OAT has also been found to reduce the risk of self-harm and suicide in people with opioid use disorder, Reference Colledge-Frisby, Jones, Degenhardt, Hickman, Padmanathan and Santo21,Reference Santo, Clark, Hickman, Grebely, Campbell and Sordo35 although careful monitoring should occur during its initiation and cessation. Reference Colledge-Frisby, Jones, Degenhardt, Hickman, Padmanathan and Santo21
Strengths and limitations
A significant strength of the current study is the use of several different data-sets to operationalise known suicide risk factors. These include hospitalisations, mental health ambulatory care and related medicines dispensed to ascertain mental health and substance use-related variables. Several limitations need to be acknowledged. Because PBS data were utilised to define the original POPPY II cohort, data on dispensings of private opioid prescriptions were not included. However, dispensings for private prescriptions contribute to a small proportion of overall opioid use in Australia Reference Busingye, Daniels, Brett, Pollack, Belcher and Chidwick36,Reference Gisev, Pearson, Karanges, Larance, Buckley and Larney37 (approximately 6%), and it is unlikely that the inclusion of private dispensings would impact the findings and generalisability of the results. Additionally, although we utilised multiple data-sets to identify confounders, we may have still under-identified these risk factors. For example, we were unable to identify people with mental health or substance use problems who had not been dispensed a relevant medicine, received public out-patient treatment or been hospitalised where the mental health condition influenced their care. Another important caveat is that we were unable to identify key physical health conditions known to be associated with self-harm and suicide, such as chronic non-cancer pain, due to difficulties involving accurate identification of such patients in the administrative data-sets used in this study. Most chronic pain is managed within the primary care setting and, because indications for medicine use are not recorded in the PBS data-set, we are unable to differentiate individuals prescribed opioids for a chronic pain condition. The inclusion of pain-related covariates may attenuate the relationship between opioids and self-harm and suicide. Reference Singhal, Ross, Seminog, Hawton and Goldacre38
Data availability
The data supporting this study’s findings are available from the senior author, G.C., upon reasonable request.
Acknowledgements
We thank the POPPY II Investigator team for their input into the design of the larger study from which data were accessed. We also thank the NSW Ministry of Health, the Centre for Health Record Linkage and AIHW for providing the data.
Author contributions
The study idea was conceived by G.C., B.D., L.D. and N.G. G.C. and B.D. secured funding for the current study through the American Foundation for Suicide Prevention Young Investigator Grant. N.G., L.D. and S.-A.P. conceived and obtained funding for the POPPY II cohort. G.C., D.T.T., C.I.B., L.D., B.D., S.-A.P., N.G. and A.H. contributed to the study design, research questions and statistical analysis plan. D.T.T. conducted the data analysis under the supervision of A.H., N.G. and C.I.B. G.C. drafted the initial manuscript, with all authors providing critical review and approving the final version. G.C. accepts full responsibility for the work and serves as the guarantor for the study.
Funding
The current study is supported by the American Foundation for Suicide Prevention via a Young Investigator Grant (no. YIG-1-024-19). The POPPY II project is supported by the National Health and Medical Research Council Health (NHMRC) project grant (no. 1138442). L.D. is supported by an NHMRC Senior Principal Research Fellowship (no. 1135991). D.T. is supported by the ASCEND programme (Advancing the health of people who use drugs: hepatitis C and drug dependence, NHMRC no. 1150078). The National Centre for Youth Substance Use Research (NCYSUR) and the National Drug and Alcohol Research Centre (NDARC) are supported by funding from the Australian Government Department of Health under the Drug and Alcohol Program. The funders of the study had no role in study design, data analysis, data interpretation or writing of the report.
Ethical standards
The current study received ethical approval from the University of New South Wales, Human Research Ethics Committee (no. HC200700). The overall POPPY II study protocol has received full ethical approval from the AIHW Ethics Committee (no. EO2016/4/314) and the NSW Population and Health Services Research Committee (no. 2017/HRE0208).
Declaration of interest
L.D. has received untied educational grants from Reckitt Benckiser, Indivior, Mundipharma Pty Ltd and Seqirus. These untied grants are all unrelated to the current study. All other authors declare no competing interests.
Open access
eLetters
No eLetters have been published for this article.