Initiation of opioid agonist treatment following hospital-treated opioid toxicities and the risk of repeat events in Ontario, Canada: a cohort study using inverse probability of treatment weights

Abstract

Introduction

The period following hospital discharge for opioid toxicity presents an increased risk for repeat events. While opioid agonist treatment (OAT) can reduce this risk, in-hospital initiation rates remain low, and the impact of early initiation on subsequent repeat toxicities is not well understood.

Methods

We conducted a population-based cohort study of Ontario residents discharged following a hospital-treated opioid toxicity between 1 January 2014 and 31 December 2021, followed to 31 July 2022. The exposure was OAT initiation within 30 days of discharge, with a subgroup analysis of OAT initiation within 1 day. The primary outcome was repeat opioid toxicity within 6 months. Stabilised inverse probability treatment weighting (sIPTW) was used to balance baseline covariates, and Cox proportional hazards models to assess the association between OAT initiation and hazard of repeat events.

Results

Our cohort included 20 523 opioid toxicities of which 68.5% were male, with a mean age of 36.5 years. Before sIPTW, important differences were observed, including more prior year opioid toxicities (mean 1.35 vs 1.05; standardised difference (Std Diff)= 0.18) and greater public drug benefit eligibility (70.1% vs 58.1%; Std Diff=0.25) among the exposed. Overall, 2544 people (12.4%) initiated OAT within 30 days of hospital discharge. In weighted models, people who initiated OAT in the month following discharge had a lower hazard of repeat toxicity (1.62 per 1000 person-days) compared with those who did not (2.04 per 1000 person-days; weighted HR=0.57; 95% CI 0.47 to 0.69). Findings were consistent in our subgroup analysis (weighted HR=0.44; 95% CI 0.57 to 0.91).

Conclusions

Among the small proportion who initiated OAT after a hospital-treated opioid toxicity, there was a significant reduction in repeat toxicities, emphasising the importance of early OAT initiation. To facilitate this, care pathways between hospital and community must be strengthened through enhanced investments in addiction medicine training, hospital resources and reducing stigma in hospitals.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• The period following a hospital-treated opioid toxicity presents an increased risk for repeat events and mortality.

• Connection to opioid agonist treatment (OAT) can help reduce this risk; however, in-hospital initiation rates remain low.

WHAT THIS STUDY ADDS

• In the 6 months following a hospital-treated opioid toxicity, almost one-quarter of individuals experienced a repeat event.

• Among the small proportion of individuals who received OAT, this risk was reduced by almost half, highlighting the importance of connection to OAT soon after an opioid toxicity.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• In order to improve treatment initiation among people with opioid use disorder following hospitalisation, pathways between hospital and community settings must be strengthened.

• This may be achieved through enhanced investments in addictions medicine-specific training, hospital-based resources and reducing stigmatising experiences in hospital.

Introduction

The unregulated drug toxicity crisis is a complex public health issue, driven by a volatile and unpredictable drug supply comprised primarily of fentanyl and its analogues.¹ This crisis continues to be associated with substantial mortality and morbidity across Canada, with 5514 hospitalisations and 7146 deaths attributed to opioid toxicities (commonly referred to as an overdose) in 2024.¹ People with opioid use disorder (OUD) are most frequently impacted by the harms associated with substance use,² with estimates suggesting that one million Canadians will be diagnosed with OUD in their lifetime.³ This population often has complex medical needs with elevated rates of coexisting substance use disorders (ie, alcohol, stimulant and benzodiazepine), mental-health-related conditions and chronic pain,^{4 5} highlighting that the importance of ensuring people with OUD is adequately connected to comprehensive care. However, people who use drugs often report significant barriers to accessing healthcare, including experiences of stigma, discrimination and structural marginalisation, resulting in a lack of access to timely care and outpatient services.⁶

In response to the ongoing drug toxicity crisis, it is critical that a range of evidence-based harm reduction and treatment options are available, including opioid agonist treatment (OAT), prescribed safer opioid supply, supervised consumption sites and access to high-quality primary care. The initiation of OAT significantly reduces the risk of opioid toxicity among people with OUD,7,9 with receipt of OAT associated with a 66% reduction (HR= 0.34; 95% CI 0.14 to 0.82) in repeat toxicities over the following year¹⁰; however, evidence from Canada suggests that only 5.6% of people are connected to OAT within 1 week of a toxicity.¹¹ This low uptake following toxicities is particularly concerning as this period presents a time of increased risk for adverse clinical outcomes and repeat events.^{12 13} On discharge, approximately 30% of individuals are readmitted to the hospital for any cause within 1 year, with as many as 20% experiencing repeat events over the same period.^{14 15} Given the high prevalence of repeat toxicities following an initial toxicity, it is critical to better understand how the initiation of OAT affects this risk. Therefore, we sought to evaluate the association between OAT initiation following a hospital-treated opioid toxicity and repeat hospital-treated toxicities within the following 6 months in Ontario, Canada.

Methods

Setting

We conducted a population-based cohort study among Ontario residents aged 15 years and older (for whom treatment with OAT is recommended first line) who experienced a hospital-treated non-fatal opioid toxicity between 1 January 2014 and 31 December 2021. Cohort members were followed to 31 July 2022. The study protocol was preregistered using the Open Science Framework (doi:osf.io/cb7ur) and is reported following the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.¹⁶

Study cohort

Our study cohort consisted of individuals who were discharged alive following an emergency department (ED) visit or in-patient hospitalisation for opioid toxicity. We defined hospital-treated opioid toxicity events using the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, diagnosis codes T40.0–T40.4 or T40.6, a definition used by the Canadian Institute of Health Information.¹⁷ We captured all non-fatal hospital-treated opioid toxicities over the study period, meaning a single individual could contribute to the cohort multiple times if they fulfilled the inclusion criteria. To account for within-person correlation due to repeated events, we used a robust sandwich estimator.¹⁸ The exposure of interest was post-toxicity initiation of OAT, defined as the first community-dispensed prescription (inclusive of methadone, buprenorphine-containing products and slow-release oral morphine) (online supplemental table S1) within 30 days following hospital discharge for the initial toxicity. Among exposed individuals, we defined the index date as the date of the first prescription. We could not follow unexposed individuals from the date of discharge, as this could introduce immortal time bias. Therefore, we randomly assigned an index date, drawing from the distribution of the number of days between hospital discharge and OAT initiation among exposed individuals,¹⁹ and ensured that all individuals were alive on their assigned index date. This approach ensured that both groups had a comparable distribution of index dates relative to discharge. We restricted our cohort to people with OUD, as they would be eligible to initiate OAT. We defined this as a healthcare encounter related to an OUD diagnosis (online supplemental table S2) or those who received treatment for OUD (receipt of an OAT prescription or ≥3 OAT-related outpatient physician visits) in a maximum lookback period of 5 years before each opioid toxicity to allow for a broad capture (see online supplemental table S3 for feasibility work using different lookback periods).

We excluded individuals who did not have a valid Ontario health card number or were non-Ontario residents at the index date to allow for data linkage. To restrict the cohort to people newly initiating OAT, we excluded those with a recent relevant prescription before each opioid toxicity (30 days for methadone or buprenorphine/naloxone, 90 days for buprenorphine injection and 270 days for buprenorphine implant).^{11 20 21} We also excluded people who received daily dispensed immediate-release hydromorphone (suggestive of safer opioid supply prescribing) in the 30 days before each hospital-treated opioid toxicity or before OAT initiation as these individuals may have a distinct risk of toxicity. We further excluded people with a non-toxicity-related hospitalisation during the time between hospital discharge and their index date as we could not determine if OAT was initiated during this hospitalisation. Finally, we excluded people with a cancer diagnosis or those receiving cancer treatment in the prior year^{22 23} (online supplemental table S4) because clinical decisions about opioid prescribing may differ in this population.

Data sources

We leveraged routinely collected administrative health data, which were linked by unique encoded identifiers and analysed at ICES (formerly known as the Institute for Clinical Evaluative Sciences). ICES is an independent, non-profit research institute whose legal status under Ontario’s Personal Health Information Protection Act privacy law allows it to collect and analyse healthcare and demographic data, without consent of Ontario residents registered with the Ontario Health Insurance Plan for health system evaluation and improvement. We used the Canadian Institute for Health Information’s National Ambulatory Care Reporting System, Discharge Abstract Database and Ontario Mental-Health Reporting System to identify all hospital encounters for opioid toxicities and to extract diagnoses for comorbid conditions. We used the narcotics monitoring system (NMS), which contains information on all controlled substances dispensed from Ontario community pharmacies, regardless of payer, to identify prior prescriptions for OAT and other controlled substances. We identified prescriptions for OAT prior to the establishment of the NMS (July 2012)²⁴ using the Ontario Drug Benefit Claims database, thereby providing an extended observation period when identifying people with OUD. We identified relevant outpatient visits with a primary care provider using the Ontario Health Insurance Plan Claims and the Community Health Centre Databases. Finally, we used the registered persons database to determine demographic information and annual estimates of the overall population. Data used in this study are authorised under section 45 of Ontario’s Personal Health Information Protection Act and do not require review by a research ethics board.

Primary outcome—repeat opioid toxicity

Our primary outcome was the occurrence of a repeat hospital-treated non-fatal opioid toxicity (online supplemental table S5) or a fatal opioid toxicity (ie, those with a coroner’s record indicating an accidental toxicity death where an opioid directly contributed) in the 182 days following each index date (online supplemental figure S1). We followed people until the occurrence of the outcome, censoring on changes in OAT exposure status (discontinuation among the exposed and initiation of OAT among the unexposed), non-opioid related mortality and the end of follow-up (182 days), whichever occurred first. If an individual experienced both a non-fatal and a fatal opioid toxicity during follow-up, we used the first event as the outcome. We allowed the exposed to switch between OAT formulations (ie, methadone, a buprenorphine-containing product or slow-release oral morphine), with discontinuation defined as missing a minimum of 14 consecutive days of OAT.^{25 26}

Statistical analyses

We used descriptive statistics to summarise index date demographic characteristics, measures of comorbidity and related hospital and geographic variables. We summarised data by exposure group and calculated the standardised difference (Std Diff) between groups, with an imbalance defined as Std Diff>0.100.²⁷ To account for differences between exposure groups, we used inverse probability treatment weighting (IPTW).²⁸ These weights were estimated using propensity scores (PSs) by regressing exposure on a number of person, hospital and geographic-related covariates (all 52 variables included in our PS models and the corresponding definitions can be found in online supplemental tables S6 and S7). We used a non-parsimonious model-building approach,²⁹ where all covariates were selected a priori through consultation with the study team. To avoid overfitting our models, we checked for collinearity using the variance inflation factor (VIF), with VIF>5 used to identify potential collinearity between variables.³⁰

We added two random effects to our PS model to account for clustering among hospitals and geographic regions. We included restricted cubic splines based on five knots placed at the 5th, 27.5th, 50th, 72.5th and 95th percentiles³¹ (three knots for number of prior opioid toxicities and hospitalisations) to model the relationship between each continuous variable and the log odds of treatment.²⁸ For income quintile and region of residence, we imputed missing values (2.6% of events) using the median and most frequent values, respectively. Given the presence of extreme IPTW weights, we applied trimming methods to exclude individuals with a PS of either<0.1 or >0.9³² (0.5% of events) and applied the stabilised IPTW (sIPTW).²⁸

We calculated event rates per 1000 person-days of follow-up and estimated the average effect of OAT using a univariate Cox proportional hazards model. We used sIPTW-weighted Cox proportional hazards models to evaluate the association between the initiation of OAT and the hazard of repeat hospital-treated opioid toxicity. Our models included a robust sandwich covariance matrix to account for correlation between repeated events among the same individual.¹⁸ We tested for violations of the proportional hazards assumption by examining the log–log plot and by adding a time-dependent covariate representing the interaction of time into our model. The proportional hazards assumption was not met, and therefore, we reported the HR at four different time points.

To assess the impact of rapid access to OAT immediately following each hospital-treated opioid toxicity, we conducted a subgroup analysis restricted to people with an index date within 1 day of hospital discharge (ie, those dispensed OAT from a community pharmacy on the date of hospital discharge or the day following). We also conducted four prespecified stratified analyses using an interaction term with the exposure variable (OAT: yes/no) in our models, including sex, number of hospital-treated opioid toxicities in the past year (grouped as 0, 1, 2 and ≥3), the number of days (grouped as 0–7, 8–21 and 22–30 days) between initial hospital discharge and index date and whether someone had a primary care visit during this period. Finally, in a post hoc analysis, we conducted a stratified analysis by the type of OAT formulation initiated by each exposed individual (ie, methadone, buprenorphine-containing products, SROM or combination therapy (defined as initiation of more than one OAT formulation)). All analyses were conducted using SAS V.9.4 (SAS Institute, Cary, North Carolina, USA).

Sensitivity analyses

We conducted four sensitivity analyses to assess the robustness of our results. First, we restricted our cohort to each person’s first hospital-treated opioid toxicity, as connections to treatment in subsequent toxicities may be influenced by previous events. Second, we recalculated our PS model without the addition of random effects to see if this differed from a fixed-effects model. Third, we removed our censoring criteria on changes in OAT status to conduct an intention-to-treat analysis to assess how findings differed irrespective of whether individuals remained adherent to their initial exposure group. We also conducted this analysis among individuals included in our subgroup analysis.

Finally, given the period between each toxicity and each person’s index date, our primary analysis includes only individuals who survived until their index date, which may bias our findings.¹⁹ To address this potential bias, we conducted a time-varying exposure analysis where all individuals were followed from the date of hospital discharge (assigned as the index date) and allowed to contribute person-time to both the unexposed and exposed groups. In this analysis, OAT treatment status was updated over time, allowing us to avoid misclassifying person-time prior to OAT initiation as exposed, which is a source of immortal time bias in the conventional approaches.¹⁹ All covariates previously included in our PS model were redefined using this updated index date and added as covariates to our Cox proportional hazards model. Given that all individuals were followed from the date of hospital discharge, we added an additional censoring criterion of non-toxicity-related hospitalisations.

Involvement of people with lived experience

The Ontario Drug Policy Research Network’s³³ opioid-related research is informed by a Lived Experience Advisory Group, comprised of nine individuals with lived and living experience using opioids; this group was consulted during the study design process. In addition, two members with previous experience being hospitalised served as coauthors (JB and CM) on this study. In this role, these individuals were full members of the study team, providing feedback on the study design and interpretation and contextualisation of findings. Notably, they advised on the selection of person, hospital and geographic variables that should be included in our PS models. All people with lived and living experience were compensated for their time spent on this study³⁴ and were offered coauthorship in this article.

Results

Between January 2014 and December 2021, there were 62 685 hospital-treated opioid toxicities where people left hospital alive, of which 23 582 (37.6%) had no OUD diagnosis, 15 884 (25.3%) had recent OAT or daily dispensed hydromorphone and 2585 (4.1%) met one of the additional exclusion criteria outlined above and were, thus, excluded (figure 1). Following the application of PS trimming, our cohort included 20 523 hospital-treated opioid toxicities (12 842 unique individuals), of which the average age was 36.5 years and 68.5% were male. Among the unique individuals in our cohort, 8614 had one hospital-treated toxicity, 3574 had two events and 654 had three or more.

Figure 1. Cohort creation flowchart. OAT, opioid agonist treatment.

Initially, we observed several meaningful differences between exposure groups (table 1). For example, compared with unexposed individuals, those exposed to OAT were younger (34.7 vs 36.8 years old; Std Diff=0.19); experienced more hospital-treated opioid toxicities in the prior year (1.35 vs 1.05; Std Diff=0.18) and were more often eligible for public drug benefits (70.1% vs 58.1%; Std Diff=0.25). Following the application of sIPTW, the exposure groups were balanced (Std Diff<0.10 for all covariates; see online supplemental figures S2–S6). Among the hospital-treated 20 523 opioid toxicities, 2544 people (12.4%) initiated OAT within 30 days of hospital discharge. This included 1359 (53.4%) initiations of methadone, 1033 (40.6%) buprenorphine-containing products, 81 (3.2%) slow-release oral morphine and 71 (2.8%) combinations of the aforementioned. The median number of days retained on OAT was 29 days (IQR = 5.0–163.5 days).

Table 1. Characteristics of exposed and unexposed individuals, before sIPTW weighting.

Characteristic	Exposed N=2544	Unexposed N=17 979	Std Diff
Person-related
Continuous variables—mean (SD)
Age	34.69 (10.45)	36.77 (11.81)	0.19
Number of physician visits (prior 2 years)	34.75 (30.02)	25.08 (25.76)	0.35
Number of ED visits (prior 2 years)	11.33 (14.49)	11.47 (17.76)	0.01
Number of inpatient hospitalisations (prior 2 years)	0.98 (2.40)	0.97 (2.07)	0.00
Number of physician visits (between hospital discharge and the index date)	0.31 (0.75)	0.23 (0.74)	0.10
Number of ED visits (between hospital discharge and the index date)	0.48 (1.13)	0.22 (0.83)	0.27
Opioid toxicity count (prior year)	1.35 (1.85)	1.05 (1.47)	0.18
Categorical variables
Rostered to a primary care physician	2383 (93.7%)	16 410 (91.3%)	0.09
Enrolled to a CHC	371 (14.6%)	2454 (13.6%)	0.03
Sex—male	1753 (68.9%)	12 311 (68.5%)	0.01
Income quintile
1	1004 (39.5%)	7674 (42.7%)	0.07
2	563 (22.1%)	3790 (21.1%)	0.03
3	450 (17.7%)	3057 (17.0%)	0.02
4	264 (10.4%)	1791 (10.0%)	0.01
5	263 (10.3%)	1667 (9.3%)	0.04
Region of residence—urban	2348 (92.3%)	16 464 (91.6%)	0.03
Aggregated diagnosis group
0–5	538 (21.1%)	4461 (24.8%)	0.09
6–9	1013 (39.8%)	6216 (34.6%)	0.11
0–5	538 (21.1%)	4461 (24.8%)	0.09
Prescriptions (prior year)
Benzodiazepines	711 (27.9%)	5048 (28.1%)	0.00
Stimulants	250 (9.8%)	1519 (8.4%)	0.05
Non-OAT opioids	796 (31.3%)	5168 (28.7%)	0.06
Likely injection-related infection	227 (8.9%)	1332 (7.4%)	0.06
Anxiety disorder	331 (13.0%)	2443 (13.6%)	0.02
Deliberate self-harm	1050 (41.3%)	6830 (38.0%)	0.07
Mood disorder	138 (5.4%)	1475 (8.2%)	0.11
Schizophrenia	148 (5.8%)	1378 (7.7%)	0.07
Other mental-health-related disorders	108 (4.2%)	873 (4.9%)	0.03
Eligible for public drug benefits	1783 (70.1%)	10 447 (58.1%)	0.25
HIV	31 (1.2%)	270 (1.5%)	0.02
Hepatitis C	458 (18.0%)	2972 (16.5%)	0.04
Stimulant use disorder	1048 (41.2%)	7210 (40.1%)	0.02
Benzodiazepine use disorder	201 (7.9%)	1112 (6.2%)	0.07
Alcohol use disorder	795 (31.3%)	6129 (34.1%)	0.06
Stimulant toxicity	506 (19.9%)	3547 (19.7%)	0.00
Benzodiazepine toxicity	228 (9.0%)	1435 (8.0%)	0.04
Alcohol toxicity	82 (3.2%)	744 (4.1%)	0.05
Opioid toxicity hospital-encounter variables
Treated in the ED	2116 (83.2%)	14 990 (83.4%)	0.01
Length of stay in hospital (days)—mean (SD)	2.09 (6.24)	1.56 (3.71)	0.10
Left hospital via a self-directed discharge	430 (16.9%)	2749 (15.3%)	0.04
Hospital admission time
8:00–23:59	320 (12.6%)	2247 (12.5%)	0.00
12:00–16:59	618 (24.3%)	4283 (23.8%)	0.01
17:00–23:59	1011 (39.7%)	7004 (39.0%)	0.02
24:00–7:59	595 (23.4%)	4445 (24.7%)	0.03
Discharged on the weekend	689 (27.1%)	5046 (28.1%)	0.02
Hospital-related variables
Hospital type
Community	1603 (63.0%)	11 392 (63.4%)	0.01
Teaching	884 (34.7%)	6150 (34.2%)	0.01
Other	57 (2.2%)	437 (2.4%)	0.01
Hospital with an affiliated rapid access addictions medicine clinic	114 (4.5%)	611 (3.4%)	0.06
Hospital-treated opioid toxicity rate (per 100 discharges)	1.47 (1.11)	1.38 (0.99)	0.08
Geographic-related variables
OAT prescriber rate (per 100 people on OAT)	6.78 (3.77)	6.36 (3.54)	0.11
Opioid-related death rate (per 1000 population)	0.16 (0.10)	0.15 (0.10)	0.07

CHC, community health centre; ED, emergency department; OAT, opioid agonist treatment; sIPTW, stabilised inverse probability treatment weighting; Std Diff, standardised difference.

During the 6-month follow-up period, we observed a total of 4973 repeat opioid toxicities, among which 255 occurred among the exposed (1.62 per 1000 person-days) and 4718 among the unexposed (2.04 per 1000 person-days) (table 2). People who initiated OAT in the month following discharge had a significantly lower hazard of repeat toxicity compared with those who did not (weighted HR=0.57; 95% CI 0.47 to 0.69); however, this relationship differed over the duration of follow-up. Specifically, although the association persisted throughout the follow-up period, the strength of the association significantly (p<0.001) increased over time (7 days: weighted HR=0.85; 95% CI 0.83 to 0.87 vs 90 days: weighted HR=0.12; 95% CI 0.08 to 0.17) (table 3).

Table 2. Number of events, follow-up time and rate of repeat opioid toxicities by exposure group—overall and subgroup analyses.

	Number of initial opioid toxicity events (cohort entry)	Person-days of follow-up	Number of repeat opioid toxicities	Rate of repeat toxicities per 1000 person-days
Full cohort (OAT initiation within 30 days of discharge)	20 523	2 466 423	4973	2.02
Unexposed	17 979	2 309 290	4718	2.04
Exposed	2544	157 133	255	1.62
Subgroup analysis cohort (OAT initiation within 1 day of discharge)	3899	459 264	1083	2.36
Unexposed	3457	430 607	1045	2.43
Exposed	442	28 657	39	1.36

OAT, opioid agonist treatment.

Table 3. Association between community-dispensed OAT following a hospital-treated opioid toxicity and repeat events.

Overall Cox proportional hazards model	HR	95% CIs	P value
Crude	0.71	0.61 to 0.83	<0.001
sIPTW* adjusted†	0.57	0.47 to 0.69	<0.001
sIPTW adjusted interaction (time)			<0.001
Day 7	0.85	0.83 to 0.87
Day 30	0.49	0.44 to 0.55
Day 60	0.24	0.19 to 0.30
Day 90	0.12	0.08 to 0.17
Stratified analyses—OAT formulation
Methadone	0.73	0.58 to 0.91	0.01
Buprenorphine-containing products	0.36	0.24 to 0.53	<0.001
SROM	0.53	0.25 to 1.14	0.10
Combination therapy	1.25	0.56 to 2.76	0.59
Sensitivity analyses
Time-varying exposure analysis	0.59	0.51 to 0.68	<0.001
No random effect added to the sIPTW model (to account for clustering)	0.57	0.48 to 0.69	<0.001
First opioid toxicity per person only	0.58	0.46 to 0.74	<0.001
Intention-to-treat analysis (no censoring on OAT status switch)	0.87	0.79 to 0.96	0.01
Subgroup analyses (OAT initiation within 1 day of discharge)
Crude	0.50	0.36 to 0.69	<0.001
sIPTW adjusted	0.44	0.28 to 0.67	<0.001
Sensitivity analysis—intention-to-treat (no censoring on OAT status switch)	0.72	0.57 to 0.91	0.01

^*sIPTW.

^†All person, hospital and geographic-related variables included in the PS models can be found in online supplemental table S5.

OAT, opioid agonist treatment; PS, propensity score; sIPTW, stabilised inverse probability treatment weighting; SROM, slow-release oral morphine.

In our subgroup analysis among individuals who initiated OAT on the day of, or immediately following hospital discharge, findings were consistent (weighted HR=0.44; 95% CI 0.57 to 0.91). Among our four prespecified stratified analyses, only one had a significant interaction with exposure (p<0.001) when added to the sIPTW-weighted model—number of prior hospital-treated opioid toxicities (online supplemental table S8) where we observed heterogeneous effects related to this variable with a similar protective effect of OAT observed regardless of the number of prior hospital-treated toxicities experienced. Among our analysis stratified by type of OAT formulation, we observed important differences in the hazard of toxicity among those who initiated methadone (weighted HR=0.73; 95% CI 0.58 to 0.91) compared with buprenorphine-containing products (weighted HR=0.36; 95% CI 0.24 to 0.53). Initiating SROM or combination therapy was not significantly associated with the outcome, although our sample size was small.

Sensitivity analyses

Among our sensitivity analyses (table 3), we observed no meaningful differences when using a time-varying exposure analysis (adjusted HR=0.59; 95% CI 0.51 to 0.68), when we removed the random effects from our PS model (weighted HR=0.57; 95% CI 0.48 to 0.69) or when we restricted to the first hospital-treated opioid toxicity per person in our cohort (weighted HR=0.58; 95% CI 0.46 to 0.74). Finally, in our intention-to-treat analysis (removing censoring on OAT exposure switches), the protective effect of OAT remained statistically significant, despite a decreasing effect size in both our overall cohort (weighted HR=0.87; 95% CI 0.79 to 0.96) and in the subgroup initiating OAT soon after hospital discharge (weighted HR=0.72; 95% CI 0.57 to 0.91).

Discussion

In this study of 20 523 hospital-treated opioid toxicities, we found that almost one-quarter of events (24.2%) were followed by repeat opioid toxicity within the following 6 months. Although connection to OAT within 1 month of discharge helped to reduce this risk by almost half, treatment was only initiated among 12% of opioid toxicities. The low uptake of OAT represents critical missed opportunities for healthcare providers to engage with and support people with OUD when in hospital, given that the postdischarge period presents an elevated risk for rehospitalisation, subsequent toxicities and mortality.^{12 35 36} Further, when we assessed how the risk of repeat events differed across follow-up time and by OAT formulation, we observed an approximate dose–response effect where the longer each person was retained on OAT, the greater the protective effect, and a stronger protective effect among people initiated and retained on buprenorphine-containing products as compared with methadone. These findings align with the large evidence base supporting the effectiveness of OAT in reducing toxicity risk and the favourable safety profile of buprenorphine.37,40 Overall, this work highlights the importance of increasing the uptake of OAT following toxicity events, which may require increased knowledge among hospital-based physicians and enhanced access to specialty addictions medicine providers, particularly in the ED, where over 80% of people in our cohort were treated.

Our finding that the initiation of OAT following a hospital-treated opioid toxicity reduced the risk of repeat events, with increased benefits among those connected either during hospitalisation or immediately following discharge, is important as little is known regarding if this type of initiation reduces the risk of future harms. However, even among those who initiated OAT in our cohort, the median number of days retained in treatment was approximately 1 month. In our intention-to-treat analysis, we observed that once people abruptly discontinue OAT, their risk of toxicity drastically increases, likely due to a return to the unpredictable unregulated drug supply.^{41 42} For example, a population-based study conducted in British Columbia, Canada found that the rate of opioid-related death was highest in the first week after stopping OAT (31.9 per 1000 person-years), before declining to 8.9 per 1000 throughout weeks 5–12 since OAT discontinuation.³⁷ This highlights the importance of not only supporting connection to OAT but also addressing systematic barriers to treatment retention for people at a higher risk of discontinuing OAT, such as those with co-occurring substance use and mental-health-related disorders.^{43 44} Barriers to OAT adherence, including stigmatising experiences, requirements for daily pharmacy visits, inadequate dosing and transportation challenges,^{45 46} require a multifaceted approach, including strengthening wrap-around supports, increased flexibility (such as offering take-home doses) and improving accessibility, particularly for those in rural regions. Further, toxicity risk factors, including unstable housing, using alone and lack of access to harm reduction services,^{47 48} may further limit the effectiveness of OAT if not addressed in parallel. Integrated models of care that combine OAT with social supports, housing and harm reduction services may help address these risks and improve both retention and outcomes.

Current Canadian clinical guidelines recommend the initiation of OAT for all people with OUD when in hospital.⁴⁹ Although the decision to initiate OAT should be shared between the patient and clinical team, in our subgroup analysis, we found that only 1 in 40 people received community-dispensed OAT within 1 day of leaving hospital. Although data on in-hospital provision of medications are largely unavailable nationally, we found that the initiation of treatment while in hospital remains relatively uncommon. A recent US study found that higher linkage rates to community-based OAT were observed among people who initiated treatment in a hospital setting.⁵⁰ The administration of OAT in hospital may have additional benefits, including improved patient–provider interactions and facilitating individuals remaining in hospital for longer to complete recommended treatment.⁵¹ Despite this, many ED physicians report that they feel unprepared to initiate OAT in hospital, with barriers, including a lack of training and experience in the treatment of OUD, concerns regarding linkage to outpatient care and competing priorities for time and resources.⁵² Evidence suggests that the integration of specialised teams in hospital, such as Addictions Consult Services, which have expanded in parts of Canada and the USA,⁵³ can help support integration of OAT into hospital settings.⁵⁴ Given these findings, there is a clear need to support the expansion of these services (including hours of operation) throughout the hospital system to improve OAT initiation in hospital and community-based referrals on discharge to help prevent repeat events.

Our study has some limitations. First, there is no validated definition of OUD, and so we relied on prior healthcare interactions and OAT to classify people with a diagnosis. While this definition may miss people with OUD who are disconnected from the healthcare system, we anticipate this potential misclassification to be minimal. Second, our data do not capture medications provided while in hospital. Therefore, our definition captures new receipt of OAT in a community-based setting and may include people who were initiated on treatment during their hospitalisation, which was continued in the community on discharge. Third, to ensure that we only captured new courses of treatment, we excluded all people who had accessed treatment in the 30 days prior to hospitalisation, thereby potentially excluding those recently engaged in treatment who may have been more likely to reinitiate treatment. Fourth, there is a potential for the issue of multiple testing, given the number of analyses conducted; however, given the strength of the statistical significance observed in our findings, it is unlikely that this would meaningfully impact the interpretation of our results. Fifth, it is possible that some hospital-treated events were missed, particularly in cases where an opioid was not identified as contributing to the toxicity in the patient’s medical record. Similarly, we do not have data on toxicities treated in the community. Finally, we lacked data on key variables, such as ethnicity, race and housing status, which may influence access to treatment and healthcare services. To assess the potential impact of unmeasured confounding, we calculated the E value,^{55 56} which quantifies the minimum strength of association that an unmeasured confounder would need with both the exposure and outcome to fully explain away the observed effect. The E value was 2.31, indicating that only a relatively strong confounder could fully explain the observed association. While it is unlikely that a single unmeasured confounder of this magnitude exists, we cannot rule out the combined influence of multiple unmeasured confounders.

Conclusions

In the 6 months following a hospital-treated opioid toxicity, almost one-quarter of individuals experienced a repeat toxicity. Among the small proportion of individuals who received OAT, this risk was reduced by almost half, highlighting the importance of connection to OAT soon after opioid toxicity and strengthening care pathways between hospital and community settings. In order to improve treatment initiation among people with OUD after they are hospitalised, enhanced investments and expansions in addictions medicine-specific training should be considered.

Data availability statement

Data may be obtained from a third party and are not publicly available.