Comparative effectiveness of alternative times to opioid agonist treatment taper initiation on taper completion and all‐cause mortality among people with opioid use disorder: A retrospective population‐based target trial emulation study in British Columbia, Canada, 2010–2020 Authors: Ruyu Yan, Jeong Eun Min, Megan Kurz, Shaun R. Seaman, Paxton Bach, Sander Greenland, Paul Gustafson, Mohammad Ehsanul Karim, Lawrence McCandless, Robert William Platt, Uwe Siebert, María Eugenia Socías, Hui Xie, Bohdan Nosyk Published: 2026 ABSTRACT AbstractBackground and aimsOpioid use disorder (OUD) treatment guidelines worldwide recommend opioid agonist treatment (OAT) as a long‐term, potentially indefinite treatment for managing OUD. However, many individuals express a strong interest in eventually tapering fully off treatment. Current clinical practice guidelines offer relatively limited guidance or evidence on the appropriate timing to initiate a taper. We aimed to determine the safety and comparative effectiveness of different times from completion of OAT induction at which tapering could be considered to maximize the likelihood of taper completion and minimize the risk of mortality.DesignPopulation‐based retrospective observational study and target trial emulation based on nine linked administrative health databases.SettingBritish Columbia, Canada, from 1 January 2010 to 17 March 2020.ParticipantsIndividuals (identified via linkage of nine provincial health administrative databases) completing OAT induction with methadone or buprenorphine/naloxone who were ≥18 years of age with no known pregnancy, no history of cancer or palliative care and not currently incarcerated. We executed both incident‐user (no OAT experiences) and prevalent‐new‐user (no OAT within the past month) analyses.Intervention and comparatorThe time between completed OAT induction and taper initiation: <3 months, 3–6 months, 6–12 months, compared with 12–48 months.MeasurementsThe primary outcomes were completed taper (reaching a final daily dose of ≤5 mg/day for methadone, or ≤2 mg/0.5 mg/day for buprenorphine/naloxone) and all‐cause mortality. A clone‐censor‐weight approach was used to adjust for informative censoring and balance baseline characteristics between the groups. Logistic regression and pooled logistic regression models were used to estimate odds ratios (ORs) for completed taper and hazard ratios (HRs) for all‐cause mortality, respectively, each with 95% compatibility (‘confidence’) intervals.FindingsWe included 17 726 incident users (buprenorphine/naloxone: 36.9%) and 49 515 treatment episodes (buprenorphine/naloxone: 31.2%) from 31 231 prevalent new users who completed induction in the analyses. Among prevalent new users, beginning tapering within 3 months, between 3 and 6 months and between 6 and 12 months of completing induction was associated with an increased likelihood of completed taper [methadone: <3 months: adjusted odds ratio (aOR) = 3.09, 95% compatibility interval (95% CI) = 2.58–3.68; buprenorphine/naloxone: <3 months: aOR = 6.90, 95% CI = 5.19–9.16] but a higher risk of mortality [methadone: <3 months: adjusted hazard ratio (aHR) = 1.18, 95% CI = 1.12–1.25; buprenorphine/naloxone: <3 months: aHR = 1.12, 95% CI = 1.05–1.19], compared with initiating a taper between 12 and 48 months. Similar results were found among incident users.ConclusionsAlthough initiating early tapering off opioid agonist treatment may be associated with a greater likelihood of taper completion, this practice also increases the risk of mortality. INTRODUCTION Opioid use disorder (OUD) is a chronic recurrent illness [1, 2] affecting an estimated 26.8 million individuals, with a 47.3% increase in prevalence over the past 20 years worldwide [3]. OUD treatment guidelines in British Columbia (BC) and worldwide recommend opioid agonist treatment (OAT) as a long‐term, potentially indefinite treatment for managing OUD [1, 2]. However, many individuals express a strong interest in eventually tapering fully off treatment [4]. OAT tapering involves a gradual reduction in daily dosage to achieve a state of full opioid abstinence [5]. Common reasons for tapering off OAT include increased risks of side effects such as drowsiness, cardiac arrhythmias and myocardial infarction [6, 7, 8]. Other motivations include expectations for recovery, familial or peer pressure, the often daily inconvenience of visiting the pharmacy, financial or travel restrictions and the stigma associated with ongoing substance use treatment [1, 5, 9, 10, 11]. A cross‐sectional study of 145 individuals receiving OAT at public clinics in Sydney, Australia, found that 62% of participants expressed a strong desire to gradually taper off OAT over the next 6 months, while an additional 15% showed moderate interest [12]. Evidence on the optimal time to initiate a taper for clients is limited and often outdated for many settings. Achieving sustained abstinence following completed OAT tapers is rare, with a particularly high risk of relapse among clients who initiate tapers within the first year of OAT [12]. Moreover, the optimal duration of stability before initiating a taper may vary depending on factors such as the client’s clinical progress, the severity of their condition, their adherence to treatment and their social circumstances, including family support, financial stability, housing and the neighbourhood environment [1, 5, 13, 14, 15, 16, 17]. Current clinical practice guidelines offer relatively limited guidance or evidence on the appropriate timing to initiate a taper [2, 15, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28]. The available evidence informing strategies for transitioning off OAT among individuals who wish to taper their treatment is extremely limited. There is a paucity of empirical evidence regarding the feasibility of discontinuing OAT for stable clients without causing relapse [29]. Using population‐based linked health administrative data, we aimed to determine the safety and comparative effectiveness of differing times between the initiation of OAT and the initiation of tapering, to ascertain the appropriate timescale for maximising the likelihood of taper completion and minimising the risk of mortality, as observed in clinical practice in BC. METHODS Study design We conducted a population‐based retrospective observational study emulating a target trial for alternative times from the initiation of OAT to the completion of tapering, assessing the impact of these timescales on taper completion and all‐cause mortality. We used the linkage of nine provincial health administrative databases in BC from 1 January 2010 to 17 March 2020 (Table S1). An individual’s personal health number served as the unique identification that linked these databases [30]. The study population was identified using Drug Identification Numbers specific to OAT medications in PharmaNet [31], which was used to differentiate prescriptions for pain management (Table S2). The research question and analysis plan, including information on the data sets used, case‐finding algorithms, exposure variables and outcomes of interest, were specified a priori in a published study protocol before the study began [32]. Consistent with the Transparent Reporting of Studies Emulating a Target Trial (TARGET) guideline emphasis on clear specification of the target trial protocol [33]. Specification of the target trial In the absence of randomisation, we emulated a target trial using observational data and followed the TARGET guideline [33]. The study included individuals aged 18 years or older with no diagnosis of cancer and no history of receiving palliative care at completed induction, who initiated OAT with either methadone or buprenorphine/naloxone, and completed induction between 1 January 2010 and 17 March 17 2020 (Table 1). Completed induction was defined as reaching a minimum of 2 weeks of continuous OAT with no dose changes. Individuals who were incarcerated at completed induction were excluded, as treatment within the criminal justice system may follow different management strategies [34]. Individuals with known pregnancy at completed induction were excluded from the analysis, as there were different clinical protocols for pregnant women compared with the general population with OUD. The prescribing and dispensing policies for these medications in BC were outlined elsewhere [35]. The first date of healthcare contact related to pregnancy was used to determine pregnancy status. TABLE 1Key components of the emulated target trial.ComponentTheoretical target trialEmulating using observational dataEligibility criteriaPeople aged ≥18 years with a diagnosis of opioid dependence and no prior attempt at OAT who were not incarcerated, pregnant or had terminal cancer Incident new user design: people aged ≥18 years with a diagnosis of opioid dependence and no prior attempt at OAT who were not incarcerated, pregnant or had history of cancer or palliative care at induction completion Prevalent new user design: selection criteria above, with no history of OAT within the past month Assignment proceduresRandomisation to a specific time to start tapering, done at the time of completing inductionClone individuals to each exposure strategy, and censor individuals who no longer follow the assigned strategies. Covariates were defined via systematic reviewTreatment strategiesParticipants were randomly assigned to initiate a taper at various time points (<3 months, 3–6 months, 6–12 months, 12–48 months)Treatment strategies are assigned according to the time from completed OAT induction at which a taper was initiatedTime zeroDate of induction completionDate of completed induction (the time at which a patient is randomised to a time for taper initiation)Outcomes Completed taper All‐cause mortality Completed taper is defined as reaching a dose of ≤5 mg/day for MET and ≤2 mg/day for BNX on the last date of dispensed medication (a binary outcome) All‐cause mortality is observed during the follow‐up period (a time‐to‐event outcome); ascertained via linkage to Vital Statistics records Follow‐upFrom the date of completed induction until the earliest occurrence of death, OAT dropout, pregnancy, palliative care, cancer diagnosis, medication switch, loss to follow‐up or end of study follow‐upFrom the date of completed induction until the earliest occurrence of death, OAT dropout, pregnancy, palliative care, cancer, medication switch or end of study follow‐upCausal contrastsITT, PP effectsPP effectAnalysis plan ITT: model estimating the relative risk of completed taper across strategies PP: ising clone censoring and a pooled logistic regression model with IPCW for completed taper and for all‐cause mortality Using clone censoring and a logistic regression model with IPCW for completed taper, and a pooled logistic regression model with IPCW for all‐cause mortalityAbbreviations: BC = British Columbia; BNX = buprenorphine/naloxone; IPCW = inverse probability of censoring weights; ITT = intention to treat; MET = methadone; OAT = opioid agonist treatment; PP = per protocol. Key components of the emulated target trial. Incident new user design: people aged ≥18 years with a diagnosis of opioid dependence and no prior attempt at OAT who were not incarcerated, pregnant or had history of cancer or palliative care at induction completion Prevalent new user design: selection criteria above, with no history of OAT within the past month Completed taper All‐cause mortality Completed taper is defined as reaching a dose of ≤5 mg/day for MET and ≤2 mg/day for BNX on the last date of dispensed medication (a binary outcome) All‐cause mortality is observed during the follow‐up period (a time‐to‐event outcome); ascertained via linkage to Vital Statistics records ITT: model estimating the relative risk of completed taper across strategies PP: ising clone censoring and a pooled logistic regression model with IPCW for completed taper and for all‐cause mortality Abbreviations: BC = British Columbia; BNX = buprenorphine/naloxone; IPCW = inverse probability of censoring weights; ITT = intention to treat; MET = methadone; OAT = opioid agonist treatment; PP = per protocol. Time zero was the date of completed induction. We executed both incident user (individuals initiating OAT for the first time during the study period, without prior OAT dispensation dating back to 1 January 1996) and prevalent new user (including both incident users and those with no prior treatment attempts within the past month) analyses to capture the clinical experience of both groups. All analyses were stratified by the medication received at completed induction: either methadone or buprenorphine/naloxone. Key measures The primary exposure was the time from completed OAT induction to the initiation of tapering within a single continuous treatment episode (with no breaks in dispensed doses ≥5 or 6 days for methadone or buprenorphine/naloxone, respectively; for OAT episode construction, see Appendix S1 and Figure S1). Individuals who began OAT prior to hospital admission were assumed to continue their treatment during hospitalisation. Taper initiation was defined as the date of the second dose reduction, with at least 5 mg and 2 mg dose reduction for methadone and buprenorphine/naloxone, respectively, after completed induction, without increases in the interim. Two consecutive daily dose decreases were used to distinguish taper starts from dose adjustments. The 5/6‐day interval for treatment discontinuation was chosen to align with BC guidelines, which recommend adjusting treatment to starting doses after periods of non‐compliance to ensure safety [36]. We compared different times from completed induction to taper initiation: <3 months, 3–6 months, 6–12 months, and 12–48 months. The primary outcomes of interest included taper completion (binary) and all‐cause mortality (time‐to‐event). A completed taper was defined as reaching a final daily dose of ≤5 mg/day for methadone, or ≤2 mg/0.5 mg/day for buprenorphine/naloxone, with no subsequent OAT dispensation for at least 5 days for methadone or 6 days for buprenorphine/naloxone. By contrast, OAT discontinuation before taper initiation, or OAT discontinuation after taper initiation with a last daily dose exceeding these thresholds, was considered treatment dropout. All‐cause mortality was observed during the follow‐up period through linkage to Vital Statistics records [37]. For the outcome of the completed taper, all eligible individuals were followed weekly from the date of completed OAT induction between 2010 and 2020 until the earliest occurrence of death, OAT discontinuation, pregnancy, receipt of palliative care, cancer diagnosis, incarceration, medication switch prior to taper initiation or the end of the study period (66 months or 17 March 2020). For the outcome of all‐cause mortality, the study follow‐up began on the date of completed induction between 2010 and 2020 and ended at death, violation of eligibility criteria prior to taper initiation (i.e. medication switch, pregnancy, palliative care or cancer diagnosis, incarceration) or the end of the study follow‐up (66 months or 17 March 17 2020), whichever occurred first. We only censored prior to taper initiation, which was our primary exposure. Follow‐up was otherwise truncated at 66 months after completing induction, allowing for at least an 18‐month follow‐up period among strategies to assess outcomes. Covariate selection We identified fixed covariates (measured at completed induction) and time‐varying covariates (measured throughout follow‐up) associated with the primary outcomes from a systematic literature review [32]. We included socio‐demographic variables [sex, age, region (urban vs rural, and classified as urban if missing), receipt of income assistance, unstable housing, incarceration], measures of disease severity and comorbidity [the Charlson Comorbidity Index (CCI), drug‐related acute care visits, asthma or chronic obstructive pulmonary disease (COPD), dispensations of psychiatric or sedative medication (Table S3), dispensations of opioids other than OAT, other substance use disorder, alcohol use disorder, serious mental disorder, hepatitis C, chronic pain, treated tobacco use disorder (Table S4)] and treatment‐related factors [attachment to the OAT prescriber, OAT dose at completed induction, indicator of above‐minimum effective OAT dose (80 mg/day for methadone, 16 mg/4 mg/day for buprenorphine/naloxone), receipt of take‐home dose and urine drug test receipt]. Analyses on prevalent new users also controlled for treatment history (cumulative years of prior OAT exposure). Detailed definitions of all study variables are provided in Table S5. Causal assumption The primary assumption in this analysis is conditional exchangeability, meaning that individuals receiving different taper initiation strategies are exchangeable (i.e. as if randomised) conditional on measured covariates. We adjusted for a comprehensive set of measured covariates to reduce confounding; however, this assumption is not directly testable, and uncontrolled confounding cannot be ruled out. Additionally, the positivity assumption requires that individuals have a non‐zero probability of receiving different treatment strategies and remaining uncensored under each strategy, which we assessed by examining the distribution of stabilised inverse probability weights and truncating extreme values. Consistency was supported by defining taper initiation using explicit dose reduction criteria based on dispensing records, which enable a clear identification of the time of taper initiation. Statistical analysis We conducted a causal per‐protocol analysis to compare alternative times of initiating a taper on both primary outcomes. As the time of taper initiation was unknown at time zero (time of completing induction), defining exposure groups using future information from the follow‐up period would have the potential to introduce immortal time bias for treatment effect on all‐cause mortality and selection bias on the outcome of the completed taper. We therefore applied the clone–censor–weight approach (Appendix S2) [38, 39]. Four replicates (clones) were created for each individual at time zero, one for each tapering time strategy (Figure S2). These replicates were then merged into an expanded data set. Thereafter, at weekly intervals, we assessed whether replicates adhered to their assigned tapering time strategy. Replicates were artificially censored when they no longer followed their assigned strategy, that is, began tapering before the beginning of their assigned tapering time interval (e.g. 3–6 months) or had still not begun tapering by the end of their assigned tapering time interval. Individuals who did not initiate a taper were censored according to the strategy definition or at the end of follow‐up; taper initiation was not assumed to occur for all individuals. To address potential confounding by indication induced by artificial censoring, we estimated the inverse probability of censoring weights (IPCW) using a pooled logistic regression model for the client’s hazard of being artificially censored during each weekly time interval (Table S6) [40]. The stabilised weights for each replicate were defined as: SWt=∏k=0tPCk=0∣Ck−1=0,Dk=0,Gk=0,VPCk=0∣Ck−1=0,Dk=0,Gk=0,L¯k, where Ct denotes the indicator of artificial censoring owing to deviation from the assigned taper initiation strategy at week t; Dt is the binary indicator of outcome at week t; Gt is the indicator of other censoring at week t [i.e. pregnancy, incarceration, cancer diagnoses, palliative care, medication switch, death (for outcome of completed taper), OAT discontinuation (for outcome of completed taper) or end of study follow‐up]. The term L¯k represents a vector of covariate history by week t, and V denotes a vector of covariates measured at completed induction (time zero), which is a subset of L0. The number of weeks since completing induction (linear and quadratic terms for this), baseline variables and time‐varying variables (Table 2) were used as covariates in this logistic regression model. For the outcome of the completed taper, we applied an additional censoring model for cases where OAT was not discontinued, as the outcome (completed taper vs treatment dropout) could not be determined without OAT discontinuation (Table S7). For all‐cause mortality, a separate censoring model was fitted to account for violations of eligibility criteria prior to taper initiation. The estimated stabilised weights were truncated at the 99th percentile to prevent the influence of extreme values [41]. As this approach relies on the assumption that the censoring models are correctly specified, we examined the distribution of censoring weights to ensure the validity of our approach. Truncated IPCWs that are extreme in value or do not have a mean close to 1.00, indicated incorrectly specified models (Figures S3 and S4) [42, 43, 44]. TABLE 2Characteristics of incident users who completed opioid agonist treatment induction: 2010–2020.VariableTotal n (%)BNX n (%)MET n (%)(N = 17 726)(n = 6543)(n = 11 183)Age, years a 34 (27–45)36 (29–46)34 (27–44)Sex: female6032 (34.0)2292 (35.0)3740 (33.4)Residence: rural2396 (13.5)1016 (15.5)1380 (12.3)Receipt of income assistance in past year b 7869 (44.4)2348 (35.9)5521 (49.4)Unstable housing c Never or more than 5 years ago15 913 (89.8)5875 (89.8)10 038 (89.8)In the past 1–5 years596 (3.4)221 (3.4)375 (3.4)In the past year1217 (6.9)447 (6.8)770 (6.9)Incarcerated in the past year b 1195 (6.7)492 (7.5)703 (6.3)Concurrent conditions d Non‐opioid substance use disorder11 733 (66.2)4442 (67.9)7291 (65.2)Alcohol use disorder (ever)4073 (23.0)1589 (24.3)2484 (22.2)Non‐cancer chronic pain6200 (35.0)2370 (36.2)3830 (34.2)Hepatitis C virus (ever)654 (3.7)194 (3.0)460 (4.1)Any mental illness in the past year9965 (56.2)3835 (58.6)6130 (54.8)Serious mental disorders (ever) e 2869 (16.2)1202 (18.4)1667 (14.9)Tobacco use disorder in past year2187 (12.3)1001 (15.3)1186 (10.6)Asthma in the past year718 (4.1)277 (4.2)441 (3.9)COPD in the past year287 (1.6)131 (2.0)156 (1.4)Years since OUD diagnosis<1 year15 215 (85.8)5597 (85.5)9618 (86.0)1–5 years1551 (8.7)599 (9.2)952 (8.5)5–9 years554 (3.1)203 (3.1)351 (3.1)>10 years406 (2.3)144 (2.2)262 (2.3)Charlson Comorbidity Index > 0699 (3.9)302 (4.6)397 (3.6)Medications received in the past monthSedative4569 (25.8)2031 (31.0)2538 (22.7)Prescription opioids2447 (13.8)1062 (16.2)1385 (12.4)Psychiatric medication f 8703 (49.1)3995 (61.1)4708 (42.1)Received urine drug test in past week3959 (22.3)1434 (21.9)2525 (22.6)Received a carry g 4634 (26.1)3063 (46.8)1571 (14.0)Drug‐related acute care visits (past month)592 (3.3)440 (6.7)152 (1.4)Psychiatric hospitalisation (past month)327 (1.8)234 (3.6)93 (0.8)Time from completed induction to taper initiationNo taper12 016 (67.8)5079 (77.6)6937 (62.0)<3 months1619 (9.1)608 (9.3)1011 (9.0)3–6 months1145 (6.5)307 (4.7)838 (7.5)6–12 months1351 (7.6)281 (4.3)1070 (9.6)12–48 months1482 (8.4)259 (4.0)1223 (10.9)>48 months113 (0.6)9 (0.1)104 (0.9)Provider‐individual attachment in past 6 months h 1066 (6.0)381 (5.8)685 (6.1)<25%8115 (45.8)3397 (51.9)4718 (42.2)25–50%8545 (48.2)2765 (42.3)5780 (51.7)>50%34 (27–45)36 (29–46)34 (27–44)Abbreviations: BNX = buprenorphine/naloxone; COPD = chronic obstructive pulmonary disease; MET = methadone; OAT = opioid agonist treatment; OUD = opioid use disorder. a Median (Q1–Q3). b Plan C from PharmaNet and the income assistance record from the Ministry of Social Development and Poverty Reduction (SDPR). c ICD9: V60.0, V60.1; ICD10: Z59.0, Z59.1 from Medical Service Plan (MSP) and Discharge Abstract Databases (DAD) or no fixed address from Social Development and Poverty Reduction (SDPR). d For ICD‐9/ICD‐10 codes used to identify concurrent chronic conditions, see Table S5. e Major depressive disorder, bipolar disorder and schizophrenia. f Antidepressants, antimanic/anti‐epileptic, antipsychotics and sedatives were excluded. g Carries over 3 days (carry initiated prior to the current week and ended in the current week). h The percentage of an individual’s billing records attributed to the OAT prescribing provider among all outpatient billing records in the past 6 months prior to OAT initiation. Cell sizes of <10 are suppressed owing to the data sharing agreement. Characteristics of incident users who completed opioid agonist treatment induction: 2010–2020. Abbreviations: BNX = buprenorphine/naloxone; COPD = chronic obstructive pulmonary disease; MET = methadone; OAT = opioid agonist treatment; OUD = opioid use disorder. Median (Q1–Q3). Plan C from PharmaNet and the income assistance record from the Ministry of Social Development and Poverty Reduction (SDPR). ICD9: V60.0, V60.1; ICD10: Z59.0, Z59.1 from Medical Service Plan (MSP) and Discharge Abstract Databases (DAD) or no fixed address from Social Development and Poverty Reduction (SDPR). For ICD‐9/ICD‐10 codes used to identify concurrent chronic conditions, see Table S5. Major depressive disorder, bipolar disorder and schizophrenia. Antidepressants, antimanic/anti‐epileptic, antipsychotics and sedatives were excluded. Carries over 3 days (carry initiated prior to the current week and ended in the current week). The percentage of an individual’s billing records attributed to the OAT prescribing provider among all outpatient billing records in the past 6 months prior to OAT initiation. Cell sizes of <10 are suppressed owing to the data sharing agreement. We conducted separate analyses for completed taper and all‐cause mortality. For all‐cause mortality, we fitted a pooled logistic regression model to the expanded data set where each individual was cloned to each tapering strategy, with taper initiation strategy (a categorical covariate) and baseline variables at completed induction as covariates, incorporating the product of the two stabilised inverse probability of censoring weights. This provides an estimate of the effect of the timing of taper initiation on all‐cause mortality in the form of a hazard ratio (HR) for taper initiation strategy conditional on baseline covariates. Strictly speaking, this HR is an odds ratio (OR), but the OR well approximates the HR when the hazard during any 1 week is low, which is the case here (fewer than 10% of individuals died during the entire follow‐up interval [45]). To control for repeated treatment attempts in the prevalent new user design, we included the number of previous episodes and the total number of years of prior OAT use as covariates in the censoring models and the model for the hazard of all‐cause mortality. We additionally presented cumulative incidence curves using models that allowed for time‐varying effects of taper initiation strategy by including interaction terms (linear and quadratic) between the strategy and weeks since completed induction. Cumulative incidence was calculated as one minus the survival probability for each taper initiation strategy. Risk differences and risk ratios were calculated by comparing the cumulative incidence under each strategy with that under the reference strategy. We used a nonparametric percentile bootstrap with 1500 samples to compute 95% compatibility intervals (confidence intervals, CIs) [46, 47]. We used a weighted logistic regression model and a weighted Poisson regression model with a log link for the binary outcome of successful taper completion. Both models included the same baseline covariates as in the model for the hazard of death (which included the categorical taper initiation strategy variable) and were fitted using the product of the two stabilised inverse probability of censoring weights at the end of follow‐up among individuals who discontinued OAT. For the outcome of completed taper, adjusted odds ratios and risk ratios were estimated using weighted logistic regression and Poisson regression models, respectively. For all‐cause mortality, adjusted hazard ratios were estimated using a weighted pooled logistic regression model. Robust variance estimators were used to compute standard errors and the corresponding 95% CIs, accounting for the cloned replicates and repeated treatment attempts in the prevalent new user design [48]. We employed two‐sided compatibility intervals with a minimum compatibility level of 5% (P‐value) for interval inclusion. These intervals are numerically the same as 95% CIs; however, our phrasing appropriately refrains from implying a 95% certainty that the true value lies within the interval [47]. Sensitivity analyses We conducted subgroup and sensitivity analyses to assess the robustness of the primary analysis results and the heterogeneity of treatment effects among key subgroups. We established predetermined objectives that concentrated on limiting the study population, restricting the study time frame, reclassifying key measures and modifying the specifications of the analytic model (Table S8) [32]. We considered five alternative definitions of taper initiation to account for potential misclassification. Additionally, we reclassified the outcome of the completed taper to ≤10 mg for methadone and ≤4 mg/1 mg for buprenorphine/naloxone as these ‘final’ daily doses were necessarily arbitrary. We also reclassified the outcome to capture a measure of sustained success, with no indication of relapse (i.e. drug‐related acute care use, return to OAT or death within 18 months of taper completion). We further redefined completed induction as at least 30 days of continuous OAT without dose changes to reduce potential misclassification of medication‐assisted withdrawal as early taper initiation. Finally, to address uncontrolled potential confounders, we calculated E‐values to determine the magnitude of association necessary for these confounders to explain the observed treatment effects [49]. All analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, NC, USA). Patient and public involvement Although patients were not explicitly involved in the design of this study, the prioritisation of this analysis relative to other clinical questions was informed by qualitative feedback received on this and other related objectives specified in the parent grant R01DA050629. The results will be shared with local advocacy organisations representing people who use drugs and people who have received OAT, in consultation with these groups. The potential impact on client engagement was a significant consideration in this decision. RESULTS Population characteristics The study population included 17 726 incident users, including 11 183 individuals who completed induction with methadone and 6543 individuals who completed induction with buprenorphine/naloxone (Figure 1; Table 2). The prevalent new user design included a total of 34 042 methadone episodes and 15 473 buprenorphine/naloxone episodes from 31 231 individuals. Among clients using methadone, 4246 (38.0%) incident users and 10 738 (31.5%) prevalent new users initiated a taper after completing OAT induction, while 1464 (22.4%) buprenorphine incident users and 2678 (17.3%) prevalent new users initiated a taper (Figure S5; Tables S9 and S10). Population selection for the target trial emulation. Cell sizes of <10 were suppressed. Most incident users were male (66.0%), with a median age of 34 years (Q1 = 27 years, Q3 = 45 years). Additionally, 86.5% resided in urban settings, 6.8% were recorded as unstably housed in the past year and 6.7% had been incarcerated in the provincial corrections system within the past year (for prevalent new users, see Table S11). Primary outcome: completed taper The rate of taper completion among those completing OAT induction was generally low, with 268 (4.1%) incident users and 633 (4.1%) prevalent new users of buprenorphine/naloxone completing the taper (Tables S12 and S13). Methadone clients exhibited slightly higher taper completion rates, with 563 (5.0%) incident users and 1844 (5.4%) prevalent new users completing the taper (Tables S14 and S15). Initiating a taper within the first 3 months of completing OAT induction was associated with a higher likelihood of taper completion (compared with taper initiation 12–48 months after induction) in both incident users (buprenorphine/naloxone, aOR = 6.69, 95% CI = 4.62–9.68, aRR = 6.12, 95% CI =4.28–8.76; methadone, aOR = 2.42. 95% CI = 1.88–3.12, aRR = 2.34, 95% CI = 1.83–2.99) and prevalent new users (buprenorphine/naloxone, aOR = 6.90, 95% CI = 5.19–9.16, aRR = 6.46, 95% CI =4.89–8.53; methadone, aOR = 3.09, 95% CI = 2.58–3.68, aRR = 3.00, 95% CI = 2.52–3.56), for either medication (Table 3). Initiating the taper at 3–6 months and at 6–12 months also improved the odds of a completed taper, albeit to a lesser degree. For methadone incident users, there was no notable difference in initiating the taper at 3–6 months (aOR = 1.21, 95% CI = 0.92–1.60, aRR = 1.21, 95% CI = 0.93–1.58) or at 6–12 months (aOR = 1.14, 95% CI = 0.88–1.48, aRR = 1.14m 95% CI = 0.88–1.47), compared with initiating the taper at 12–48 months, as the compatibility intervals crossed 1.00. TABLE 3Adjusted odds ratios and risk ratios for the effect of taper initiation strategies on completed taper, compared with taper initiation at 12–48 months after completed OAT induction.Incident usersPrevalent new usersOR (95% CI)RR (95% CI)OR (95% CI)RR (95% CI) MET taper initiation <3 months2.42 (1.88–3.12)2.34 (1.83–2.99)3.09 (2.58–3.68)3.00 (2.52–3.56)3–6 months1.21 (0.92–1.60)1.21 (0.93–1.58)1.31 (1.07–1.59)1.30 (1.07–1.58)6–12 months1.14 (0.88–1.48)1.14 (0.88–1.47)1.29 (1.07–1.56)1.29 (1.07–1.55)12–48 monthsReferenceReferenceReferenceReference BNX taper initiation <3 months6.69 (4.62–9.68)6.12 (4.28–8.76)6.90 (5.19–9.16)6.46 (4.89–8.53)3–6 months2.06 (1.36–3.13)2.02 (1.35–3.05)2.08 (1.50–2.87)2.05 (1.49–2.82)6–12 months1.77 (1.15–2.70)1.74 (1.15–2.65)1.67 (1.20–2.32)1.66 (1.20–2.29)12–48 monthsReferenceReferenceReferenceReference Note: We truncated weights at the 99th percentile. The number of events for each category is presented in Table S14. We adjusted for baseline covariates and weighted for time‐varying confounders, including sex, age, region, OAT dose at completed induction, above minimum effective dose, receipt of income assistance, unstable housing, incarcerated in past year, cumulative years of prior OAT exposure (for prevalent new users), opioid dispensations for pain, drug‐related acute care visits, Charlson Comorbidity Index, dispensation of psychiatric or sedative medication in the past months, other substance use disorder, alcohol use disorder, serious mental disorder, hepatitis C, chronic pain, treated tobacco use disorder, receipt of asthma or COPD care in the past year, attachment to the OAT prescriber, OAT take‐home doses, urine drug test in past week, and linear and quadratic terms for weeks since completed induction. The global test indicated that global P‐values for all taper initiation strategies were <0.001 across all models.Abbreviations: BNX = buprenorphine/naloxone; COPD = chronic obstructive pulmonary disease; MET = methadone; OAT = opioid agonist treatment. Adjusted odds ratios and risk ratios for the effect of taper initiation strategies on completed taper, compared with taper initiation at 12–48 months after completed OAT induction. Note: We truncated weights at the 99th percentile. The number of events for each category is presented in Table S14. We adjusted for baseline covariates and weighted for time‐varying confounders, including sex, age, region, OAT dose at completed induction, above minimum effective dose, receipt of income assistance, unstable housing, incarcerated in past year, cumulative years of prior OAT exposure (for prevalent new users), opioid dispensations for pain, drug‐related acute care visits, Charlson Comorbidity Index, dispensation of psychiatric or sedative medication in the past months, other substance use disorder, alcohol use disorder, serious mental disorder, hepatitis C, chronic pain, treated tobacco use disorder, receipt of asthma or COPD care in the past year, attachment to the OAT prescriber, OAT take‐home doses, urine drug test in past week, and linear and quadratic terms for weeks since completed induction. The global test indicated that global P‐values for all taper initiation strategies were <0.001 across all models. Abbreviations: BNX = buprenorphine/naloxone; COPD = chronic obstructive pulmonary disease; MET = methadone; OAT = opioid agonist treatment. Sensitivity analyses were consistent with the main findings across a range of population subgroups, study time periods, and alternative classifications of exposure and outcome (Figure S6; Table S16). While all point estimates in the sensitivity analyses aligned with the primary analysis, compatibility intervals included 1.00 in smaller population subsets. Among individuals with chronic pain, no notable difference was observed when the taper was initiated at 3–6 months or at 6–12 months, compared with 12–48 months, after completed OAT induction in both incident users and prevalent new users, regardless of medication. Additionally, the results were sensitive to the OAT dosage received at time zero. For buprenorphine/naloxone recipients with a minimum effective dose above 16 mg/4 mg at completed OAT induction, initiating the taper at 3–6 months or at 6–12 months did not differ from initiating the taper at 12–48 months, across treatment episodes. However, for methadone clients with <80 mg or buprenorphine/naloxone clients with <16 mg/4 mg, the results were consistent with the primary analysis for prevalent new users. Restricting the study period from the first detection of fentanyl in the provincial coroner’s reports (1 April 2012) yielded consistent results with the primary analysis for both incident users and prevalent new users of buprenorphine/naloxone and prevalent new users of methadone. Results were largely consistent across alternative definitions of taper initiation and episode discontinuation. Notably, for incident users of methadone, initiating the taper between 3 and 6 months and between 6 and 12 months of completed induction was associated with an increased odds of taper completion when taper initiation was defined as two consecutive dose decreases with no reversion within 14 days (3–6 months, aOR = 1.60, 95% CI = 1.22–2.11; 6–12 months, aOR = 1.34, 95% CI = 1.03–1.75) or 28 days (3–6 months, aOR = 1.50, 95% CI = 1.14 = 1.97); 6–12 months, aOR = 1.32, 95% CI = 1.02–1.71), compared with initiating a taper between 12 and 48 months. When the definition of a completed taper was modified, either as a final dosage below 10 mg/day (3–6 months, aOR = 1.55, 95% CI = 1.23–1.96; 6–12 months, aOR = 1.42, 95% CI = 1.13–1.78) or as no treatment initiation, acute care visit or death within 18 months after taper completion (3–6 months, aOR = 1.55, 95% CI = 1.03–2.33; 6–12 months, aOR = 1.68, 95% CI = 1.15–2.46), initiating a taper between 3 and 6 months and between 6 and 12 months was associated with an increased probability of taper completion among methadone incident users. Finally, we calculated E‐values to assess the potential impact of uncontrolled confounding, which ranged from 1.55 to 13.27 for point estimates in our primary analysis (Table S17). Primary outcome: all‐cause mortality Among buprenorphine/naloxone recipients who completed induction, 268 deaths were observed in incident users (4.1%; 15.6 per 1000 person‐years) and 634 deaths in prevalent new users (4.1%; 16.2 per 1000 person‐years). Results were similar among methadone clients, with 565 deaths in incident users (5.1%; 12.8 per 1000 person‐years) and 1854 deaths in prevalent new users (5.4%; 14.9 per 1000 person‐years). Initiating a taper within 3 months, between 3 and 6 months and between 6 and 12 months after completing induction was associated with an increased hazard of mortality compared with initiating a taper after 1 year for both incident users and prevalent new users of both methadone and buprenorphine/naloxone (Table 4). The cumulative incidence of mortality was lowest for those initiating the taper between 12 and 48 months but with no visibly distinguishable differences for both methadone and buprenorphine/naloxone recipients (Figure S7; Table S18). TABLE 4Adjusted hazard ratios and 66‐month risk ratios for the effect of taper initiation strategies on all‐cause mortality, compared with taper initiation at 12–48 months after completed OAT induction.Incident usersPrevalent new usersHR (95% CI)RR (95% CI)HR (95% CI)RR (95% CI) MET taper initiation <3 months1.21 (1.08–1.35)1.20 (1.07–1.33)1.18 (1.12–1.25)1.17 (1.12–1.24)3–6 months1.20 (1.11–1.31)1.19 (1.09–1.29)1.12 (1.08–1.17)1.12 (1.08–1.17)6–12 months1.11 (1.04–1.19)1.11 (1.04–1.18)1.08 (1.04–1.11)1.07 (1.04–1.11)12–48 monthsReferenceReferenceReferenceReference BNX taper initiation <3 months1.13 (1.02–1.26)1.14 (1.03–1.25)1.12 (1.05–1.19)1.12 (1.06–1.18)3–6 months1.17 (1.09–1.25)1.17 (1.09–1.23)1.11 (1.07–1.15)1.12 (1.07–1.16)6–12 months1.08 (1.04–1.13)1.08 (1.04–1.13)1.07 (1.04–1.10)1.08 (1.05–1.10)12–48 monthsReferenceReferenceReferenceReference Note: We truncated weights at the 99th percentile. The number of events for each category is presented in Table S14. We adjusted for baseline covariates and weighted for time‐varying confounders, including sex, age, region, OAT dose at completed induction, above minimum effective dose, receipt of income assistance, unstable housing, incarcerated in past year, cumulative years of prior OAT exposure (for prevalent new users), opioid dispensations for pain, drug‐related acute care visits, Charlson Comorbidity Index, dispensation of psychiatric or sedative medication in the past months, other substance use disorder, alcohol use disorder, serious mental disorder, hepatitis C, chronic pain, treated tobacco use disorder, receipt of asthma or COPD care in past year, attachment to the OAT prescriber, OAT take‐home doses, urine drug test in past week, and linear and quadratic terms for weeks since completed induction. The global test indicated that global P‐values for all taper initiation strategies were <0.001 across all models.Abbreviations: BNX = buprenorphine/naloxone; COPD = chronic obstructive pulmonary disease; MET = methadone; OAT = opioid agonist treatment. Adjusted hazard ratios and 66‐month risk ratios for the effect of taper initiation strategies on all‐cause mortality, compared with taper initiation at 12–48 months after completed OAT induction. Note: We truncated weights at the 99th percentile. The number of events for each category is presented in Table S14. We adjusted for baseline covariates and weighted for time‐varying confounders, including sex, age, region, OAT dose at completed induction, above minimum effective dose, receipt of income assistance, unstable housing, incarcerated in past year, cumulative years of prior OAT exposure (for prevalent new users), opioid dispensations for pain, drug‐related acute care visits, Charlson Comorbidity Index, dispensation of psychiatric or sedative medication in the past months, other substance use disorder, alcohol use disorder, serious mental disorder, hepatitis C, chronic pain, treated tobacco use disorder, receipt of asthma or COPD care in past year, attachment to the OAT prescriber, OAT take‐home doses, urine drug test in past week, and linear and quadratic terms for weeks since completed induction. The global test indicated that global P‐values for all taper initiation strategies were <0.001 across all models. Abbreviations: BNX = buprenorphine/naloxone; COPD = chronic obstructive pulmonary disease; MET = methadone; OAT = opioid agonist treatment. Sensitivity analyses largely supported our primary findings (Figure S8; Table S19). Findings were similar to the primary results after the initial fentanyl detection and the declaration of a public emergency for both methadone and buprenorphine/naloxone clients. Results were, however, sensitive to adjustments in the minimum effective dose at completed induction. For prevalent new users of methadone who completed induction with a minimum effective dose above 80 mg/day, the timing of initiating the taper was not associated with mortality risk. Additionally, when individuals were censored at treatment discontinuation, we did not observe clear evidence that the hazard of mortality on treatment across four taper initiation strategies among either incident users or prevalent new users, regardless of medication, as the compatibility intervals included null. Finally, E‐values for the point estimates ranged from 1.35 to 1.71 for the primary analysis. DISCUSSION We conducted a population‐level study including 31 231 OAT recipients to emulate target trials comparing the effectiveness of alternative times for initiating a taper. The likelihood of completing a taper was low among incident users and prevalent new users of both methadone and buprenorphine/naloxone. We found that initiating tapers soon after completing OAT induction typically resulted in an increased likelihood of a completed taper but also increased the risk of all‐cause mortality compared with initiating tapers after 1 year. A range of subgroup and sensitivity analyses supported the notion with consistent results. Current clinical OAT guidelines explicitly advise against tapering off OAT, owing to the increased risk of overdose and relapse associated with the discontinuation of OAT. For individuals who still have a strong desire to eventually discontinue treatment, BC and national guidelines specifically discourage taper initiation within 1 year of stabilised OAT, yet the recommendation has thus far lacked robust supporting evidence [1, 2, 18, 20, 22, 23, 50, 51]. This recommendation may be predicated on the evidence that an extended time on OAT reduces the risk of mortality; however, no evidence cited for this recommendation in BC guidelines has directly linked the time on OAT prior to taper initiation to either death or successful taper completion. Individuals with sustained engagement with OAT may have developed a deeper familiarity and routine with treatment than those with less experience. Therefore, when faced with withdrawal discomfort during tapering, they may be more inclined to return to OAT as a viable strategy, especially if they lack a strong intent to taper off OAT. Additionally, our findings indicated an increased risk of mortality among those who initiate an early taper for both medications, regardless of treatment adherence; however, we observed no association between the timing of taper initiation and mortality while on treatment. OAT reduces the risk of mortality while individuals are actively undergoing treatment [1, 2]. However, the risk of death rises after discontinuing OAT, including for those who have completed tapers. Therefore, initiating a taper is typically not recommended, whether initiated early or late in the course of treatment. Our primary findings suggest higher odds of taper completion when tapers are initiated early among incident users and prevalent new users, contradicting previous research on this topic. Some studies reported that initiating the taper after longer treatment durations is linked to a greater likelihood of taper completion. For example, a retrospective study involving 853 individuals aiming to taper off methadone in Guangdong, China, found that initiating a taper after 13 months was associated with higher odds of taper completion compared with tapers initiated in the first 4 months (aOR = 2.81, 95% CI = 1.48–5.34) [52]. Another retrospective study of 5774 individuals who tapered off buprenorphine reported that people who initiated a taper after 1 year of OAT treatment had a lower hazard of opioid overdose within 18 months post‐treatment, compared with those who started a taper within 1 year (aHR = 0.69, 95% CI = 0.48–0.997) [14]. However, some findings contradict the benefits of extended OAT prior to tapering. A retrospective study with results similar to ours involved 14 602 clients initiating methadone tapering in BC, and reported that the duration of methadone treatment did not independently affect sustained drug abstinence 18 months after taper completion, after adjusting for other covariates [5]. This discrepancy might arise because neither of these studies fully adjusted for the inherent confounding by indication in the comparison; individuals who are able to remain on treatment longer have inherently different risks of discontinuation and mortality compared with those initiating tapers early. The clone–censor–weight approach we employed corrects the error by assigning each individual to all strategies at time zero and artificially censoring them according to their actual taper initiation time, with weighting accounting for this artificial censoring [53]. Our analyses estimate the average effects of alternative taper initiation strategies at the population level among individuals who completed OAT induction. Earlier taper initiation was associated with a higher hazard of all‐cause mortality, suggesting that longer retention in OAT may be important for safety at the population level. These findings therefore contribute evidence relevant to population‐level treatment strategies. However, decisions on the timing of taper initiation in clinical practice are inherently individualised. In clinical practice, a full tapering protocol is dynamic, encompassing both when the taper is initiated and how the taper proceeds thereafter. Our study focused only on the timing of taper initiation after completing OAT induction, defined using prespecified initiation windows, and did not evaluate the taper schedule after taper initiation. Accordingly, our findings should be interpreted as evidence regarding the timing of taper initiation in real‐world care rather than the effects of a fully specified dynamic tapering protocol. This study largely adhered to the specifications outlined in a previously published protocol [32]. To enhance the statistical power, we combined taper initiation strategies of 12–24 months and 24–48 months into 12–48 months for methadone and extended 24 months to 48 months for buprenorphine. For the outcome of all‐cause mortality in the primary analysis, we adjusted it from death occurring during treatment to any time during follow‐up, regardless of OAT discontinuation. We included all‐cause mortality on treatment in the sensitivity analysis to account for deaths occurring within 4 or 5 days of OAT discontinuation. For the outcome of completed taper, we considered OAT discontinuation prior to taper initiation as treatment dropout, instead of censoring at OAT discontinuation, as individuals would be allowed to drop out of treatment during the study period in a randomised trial. This study has several limitations. Despite implementing a target trial emulation framework to reduce bias, we cannot rule out the potential for uncontrolled fixed or time‐varying confounding. Nevertheless, the computed E‐values suggest that strong uncontrolled confounding would be required to fully explain the associations between the timing of taper initiation and the completed taper, whereas moderate uncontrolled confounding could potentially account for the smaller observed association with all‐cause mortality [49]. Importantly, similar to any other observational study, the validity of our estimates relies on the untestable assumption that the measured covariates were sufficient to adjust for confounding by indication. As we can only observe the dispensed doses of OAT prescriptions in our administrative databases, we cannot determine whether the individual fully reduced the doses. Despite applying alternative definitions of taper completion in sensitivity analysis, measurement bias may persist. Additionally, because the available administrative data began in 1996, earlier OAT episodes could not be observed. Therefore, some individuals classified as incident users may have had prior treatment before 1996 and may have been misclassified. Lastly, the data in this study were sourced from BC and may not be generalisable to settings with different co‐payment structures, which may affect both the likelihood of taper initiation and completion. CONCLUSION Opioid agonist treatment (OAT) is regarded as an open‐ended treatment. However, for individuals who strongly desire to taper off OAT, at least 1 year of stability on OAT before initiating a taper is recommended by guidelines. This population‐based study demonstrated that while initiating a taper early after completing induction may increase the likelihood of taper completion, compared with initiating tapers after 1 year of OAT, it may also result in an increased risk of mortality. AUTHOR CONTRIBUTIONS Ruyu Yan: Conceptualization; methodology; software; data curation; formal analysis; visualization; writing—original draft. Jeong Eun Min: Methodology; conceptualization; data curation; writing—review and editing. Megan Kurz: Methodology; writing—review and editing; conceptualization. Shaun R. Seaman: Methodology; writing—review and editing. Paxton Bach: Writing—review and editing. Sander Greenland: Writing—review and editing. Paul Gustafson: Writing—review and editing. Mohammad Ehsanul Karim: Writing—review and editing. Lawrence McCandless: Writing—review and editing. Robert William Platt: Writing—review and editing. Uwe Siebert: Writing—review and editing. María Eugenia Socías: Writing—review and editing. Hui Xie: Writing—review and editing. Bohdan Nosyk: Conceptualization; supervision; project administration; funding acquisition; writing—review and editing. DECLARATION OF INTERESTS The authors have no interests to declare. Supporting information Appendix S1. OAT episode construction. Appendix S2. Clone–censor–weight approach. Figure S1. Diagram describing OAT episode construction. Figure S2. Example of replicated strategies and artificial censoring for clone‐censor approach. Figure S3. Distribution of stabilised weights for all‐cause mortality over a period of up to 66 months of follow‐up among methadone clients, comparing weights with and without truncation. Figure S4. Distribution of stabilised weights for all‐cause mortality over a period of up to 66 months of follow‐up among buprenorphine/naloxone clients, comparing weights with and without truncation. Figure S5. Number of individuals following each monitoring strategy, by medication type. Figure S6. Subgroup and sensitivity analyses on the effect of the timing of taper initiation, compared with initiating the taper between 12 and 48 months, on completed taper. Figure S7. Adjusted cumulative incidence curve for all‐cause mortality. Figure S8. Subgroup and sensitivity analyses on the effect of timing of taper initiation, compared with initiating the taper between 12 and 48 months, on all‐cause mortality. Table S1. List of linked administrative health data sources. Table S2. Drug identification numbers for identification of opioid agonist treatment from PharmaNet. Table S3. Identification of psychiatric and sedative medications. Table S4. Diagnostic codes used to identify concurrent chronic conditions. Table S5. Variables and definitions. Table S6. Contributions to the stabilised weights at different time points for strategies that initiate a taper at the end of the grace period. Table S7. Censoring indications in the primary analysis. Table S8. Proposed subgroup and sensitivity analyses. Table S9. Characteristics measured at completed induction among incident users, stratified by whether taper was initiated during follow‐up. Table S10. Characteristics measured at completed induction among prevalent new users, stratified by whether taper was initiated during follow‐up. Table S11. Characteristics of prevalent new users who completed opioid agonist treatment induction in British Columbia, Canada, from 1 January 2010 to 17 March 2020. Table S12. Characteristics measured at completed induction among incident users, stratified by status of taper completion. Table S13. Characteristics measured at completed induction among prevalent new users, stratified by status of taper completion. Table S14. Durations of included OAT episodes after completed induction by taper initiation strategy. Table S15. Summary statistics on primary outcomes. Table S16. Event rates and adjusted odds ratios for the effect of timing of taper initiation on completed taper in sensitivity analyses. Table S17. Adjusted hazard ratios, adjusted odds ratios, CIs, and E‐values for the effect of timing of taper initiation on primary outcomes. Table S18. Adjusted cumulative incidence, risk differences and risk ratios for the effect of timing of taper initiation on all‐cause mortality at selected follow‐up time points following completed induction. Table S19. Event rates and adjusted hazard ratios for the effect of timing of taper initiation on all‐cause mortality in sensitivity analyses.