Comparative effectiveness of alternative initial doses of opioid agonist treatment for individuals with opioid use disorder: a protocol for a retrospective population-based study using target trial emulation in British Columbia, Canada

Abstract

Introduction

Selecting an optimal initial dosage of opioid agonist treatment (OAT) balances effectiveness and safety, as initial doses that are too low may be insufficient, potentially prompting clients to seek unregulated drugs to alleviate withdrawal symptoms, which may increase the likelihood of treatment discontinuation. Conversely, initial doses that are too high carry a risk of overdose. As opioid tolerance levels have risen in the fentanyl era, linked population-level data capturing initial doses in the real world provide a valuable opportunity to refine existing guidance on optimal OAT dosing at treatment initiation. Our objective is to determine the comparative effectiveness of alternative initial doses of methadone, buprenorphine-naloxone and slow-release oral morphine at OAT initiation, as observed in clinical practice in British Columbia (BC), Canada.

Methods and analysis

We propose a population-level retrospective observational study with a linkage of nine provincial health administrative databases in BC, Canada (1 January 2010 to 31 December 2022). Our study includes two time-to-event primary outcomes: OAT discontinuation and all-cause mortality during follow-up. We propose ‘initiator’ target trial analyses for each medication using both propensity score weighting and instrumental variable analyses to compare the effect of different initial OAT doses on the hazard of time-to-OAT discontinuation and all-cause mortality. A range of sensitivity analyses will be used to assess the robustness of the results.

Ethics and dissemination

The protocol, cohort creation and analysis plan have been classified and approved as a quality improvement initiative by Providence Health Care Research Ethics Board and the Simon Fraser University Office of Research Ethics. Results will be disseminated to local advocacy groups and decision-makers, national and international clinical guideline developers, presented at international conferences and published in peer-reviewed journals electronically and in print.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The single-payer healthcare system in British Columbia is ideally suited for conducting direct comparisons of treatment strategies for opioid use disorder at the population level and within key subgroups.

• An ‘initiator’ approach, implemented with propensity score and instrumental variable approaches, will examine the effects of different initial doses on the time to opioid agonist treatment discontinuation and all-cause mortality among those initiating treatment, accounting for potential confounding effects at the time of initiation.

• Sensitivity analyses will assess potential uncontrolled confounding and the robustness of our results.

• Despite implementing a target trial emulation framework to reduce bias, we cannot rule out the potential for uncontrolled confounding.

Introduction

Opioid use disorder (OUD) is a chronic, recurrent illness affecting an estimated 26.8 million people, with a 47.3% increase in prevalence over the past 20 years worldwide.^{1 2} Opioid agonist treatment (OAT), including methadone (MET), buprenorphine/naloxone (BNX) and slow-release oral morphine (SROM), is intended as a long-term and potentially indefinite treatment for people with OUD. Sustained retention in treatment reduces the risk of infectious diseases, overdose and death.^{1 3} OAT commences with an induction phase, during which the initial medication dose plays a critical role in determining the safety of both the induction phase and the overall trajectory of the treatment. The optimal initial dosage of OAT balances effectiveness and safety. Clinical guidelines for outpatient OAT advise beginning with a low dosage and titrating gradually, as lower doses are well-tolerated and safe.⁴ However, lower starting doses may inadvertently prolong continued illicit opioid use and thus increase the risk of overdose and treatment discontinuation.^{5 6} Conversely, initial doses that are too high carry a risk of overdose, particularly in individuals with low opioid tolerance.

The goal of OAT is to ultimately eliminate opioid withdrawal symptoms and psychological symptoms (eg, craving) with an adequate dosage of medication.^{1 3} The induction period begins with the initial dose of OAT and may extend to over a month of early treatment, during which time an individual’s daily dose gradually increases to the point at which their opioid withdrawal and craving symptoms are eliminated. Therefore, initiation doses of OAT should ideally reduce opioid withdrawal symptoms while avoiding sedation or increased risk of overdose.⁷ OAT clients face the highest risk of overdose and mortality during the induction phase of treatment, necessitating careful considerations of safety in medication dosing decisions. A retrospective study of 5200 individuals in MET maintenance programmes in Amsterdam reported a mortality rate of 6.0 per 1000 person-years in the first 2 weeks following the initial MET dose, which was substantially higher than the 2.2 per 1000 person-years recorded during later ‘maintenance’ stages of treatment,⁸ which may be because of the slower metabolism of MET in clients who have recently initiated treatment compared with those who have achieved maintenance doses.⁹ BNX clients face the same high risk of overdose and mortality during the induction phase. Administering BNX before sufficient opioid withdrawal can display the full opioid agonist from receptors and trigger precipitated withdrawal, which may cause a rapid worsening of symptoms after the initial dose.¹⁰ However, there is limited evidence regarding SROM.

International best practice guidelines are largely concordant on recommended ranges for starting doses of OAT; however, relatively little supporting evidence is provided for these recommendations. Of 13 national clinical guidelines on the clinical management of OUD in North America, Australia and the UK, initial dose recommendations varied from 5 to 40 mg for MET and 0.8–8 mg for BNX (online supplemental table 1). However, most existing guidelines were established prior to the introduction of fentanyl into the unregulated drug supply. Fentanyl and its analogues are estimated to be 50–100 times more potent than morphine and approximately 20–50 times more potent than heroin.11,13 In British Columbia (BC), the proportion of fentanyl and its analogues within the illicit drug supply increased from 4.9% (3.9%–6.1%) in 2012 to 84.7% (83.6–85.7%) in 2022.¹⁴ While OAT can reduce the risk of fentanyl-related overdose, individuals who use fentanyl are less likely than those who use prescription opioids to be successfully retained in OAT.^{15 16} People with OUD exhibit different levels of dependence and thus tolerance levels, based on their individual use patterns and the potency of the illicit drug supply. The global rise in the availability of potent synthetic opioids may further elevate tolerance, increasing the need for higher doses of OAT to support treatment retention.

Most guidelines made recommendations on starting doses based on clients’ opioid tolerance, risk of toxicity, withdrawal symptoms and risk of precipitated withdrawal. Withdrawal symptoms and precipitated withdrawal are typically measured by self-reported rating instruments with relevant threshold scores from the Clinical Opiate Withdrawal Scale (COWS) and the Subjective Opiate Withdrawal Scale (SOWS).^{17 18} In BC, individuals who have completed withdrawal management, individuals with heavy use of other sedating agents and people with severe comorbidities that increase toxicity risks are categorised as having low opioid tolerance; individuals who consume benzodiazepines or other sedatives and individuals with an alcohol use disorder are categorised as having moderate opioid tolerance; people who are actively using opioids are considered to have high opioid tolerance and a low risk of toxicity; and individuals with known very high tolerance, who face a very low risk of toxicity, are specifically characterised by previous experience with opioid agonist treatment and current use of fentanyl.¹ This also applies to individuals with multiple substance use and comorbidities, where a diagnosis of OUD has been confirmed using standardised diagnostic criteria supporting the indication of OAT. However, the self-reported level of opioid use provides only a rough approximation of tolerance.^{19 20} Individuals who use opioids may unknowingly consume fentanyl, particularly in settings where the illicit drug supply is unpredictable and poorly labelled, causing unknown exposure to fentanyl.²¹ Additionally, fentanyl has overtaken heroin in the illicit drug supply since 2016.²² Individuals with OUD increasingly seek out fentanyl explicitly, as their opioid tolerance has escalated to the level where traditional opioids no longer provide the desired effect or adequate relief. With MET, genetics play a significant role in determining the half-life, thus even if self-reported opioid tolerance was perfectly reliable, it remains difficult to translate this directly to an MET dose, given the natural variability in metabolism.

Most clinical guidelines determine the initial dosage for MET through clients’ opioid tolerance.13 23,26 The 2017 BC guideline recommends a 5–30 mg/day MET starting dose for individuals with different levels of opioid tolerance,³ while the 2023 BC guideline revised the maximum starting dose to 40 mg for individuals with known very high tolerance.¹ Other clinical guidelines from Canada, USA and Australia recommend that the initial dose cannot exceed 30 mg²³,2427,31 ³² 33,36 and those from the UK, New Zealand and New South Wales suggest 40 mg as the maximum initial dose.2526 37,39 The ranges of starting doses were set based on evidence from observational studies, and many guidelines cite a lack of evidence on the safety and effectiveness of initial doses above 40 mg.¹

The potential risk of precipitated withdrawal guides the initial dose for BNX in current national guidelines.^{1 3 23 26 31 39 40} Since BNX is a partial agonist with a high affinity for the opioid receptor, traditional induction has required periods of up to at least 24–72 hours of abstinence from opioids to avoid precipitated withdrawal.¹ This period is challenging for clients, as it requires clients to be in moderate withdrawal before induction. The 2023 BC guideline suggested 2 mg/0.5 mg BNX for people with concerns of precipitated withdrawal (including people who use fentanyl), and 4 mg/1 mg for those with a lower risk of precipitated withdrawal (ie, recently completed withdrawal management, known time of last opioid use or fentanyl-negative urine drug test (UDT)).¹ The 2017 BC guideline additionally proposes 6 mg/1.5 mg under supervised conditions for individuals experiencing severe withdrawal symptoms.³ In contrast, international guidelines advise initial dose ranges of BNX between 2 mg/0.5 mg and 8 mg/2 mg. Only the guideline from New South Wales reported a possibility of more than 8 mg, a level for which consultation with a specialist is required.²⁶ Limited evidence was cited in these guidelines to support the benefit of >8 mg/day of buprenorphine at OAT initiation.

A low-dose induction (also known as ‘micro-dosing’) strategy was first proposed in Switzerland in 2016, and this dosing strategy has been used in Canada since 2017. The low-dose induction gradually titrates small doses of buprenorphine without discontinuing other opioids until a therapeutic dose is achieved, which is used in clinical practice as it can reduce the risk of precipitated withdrawal and does not require clients to abstain from other opioids.¹ The 2023 BC guideline and the Canadian national guideline suggested 0.5 mg/0.125 mg two times per day, while the Ontario guideline advises the initial micro dose ranging from 0.25 to 0.5 mg.^{1 23 41} To date, there is no consensus on best practices, owing to the limited evidence base supporting this practice.

Within North America, only Canadian jurisdictions offer, and thus have clinical guidelines for SROM.^{1 3 23 42} The 2023 BC guideline advises starting doses of SROM to be 50–300 mg/day,¹ while there is no recommendation about this in the 2017 BC guideline. Other guidelines from Canada recommend a dosage of 30–60 mg of SROM for clients with low opioid tolerance, and 100–200 mg for those with high tolerance or illicit fentanyl use.²³ For cases with extreme conditions, even a dosage of 200–400 mg is recommended for such individuals.^{1 3 42} There is extremely limited evidence cited in these related recommendations, as SROM was only recently introduced as a treatment option for OUD in BC in 2017, and the rest of Canada in 2018.¹ Although SROM is available as an alternative medication to MET in a range of European jurisdictions, there is a lack of established clinical guidelines.⁴³

Evidence regarding the optimal OAT starting doses is not only limited but likely outdated for settings in which fentanyl has been introduced into the illicit drug supply. Clinical guidelines for MET and BNX in Canada are broadly aligned with those in USA, UK and Australia, with initial dosing typically based on clients’ opioid tolerance. A higher initial dose of BNX may be associated with increased treatment retention. A retrospective study of 17 158 individuals initiating buprenorphine treatment in the US reported that compared with initiating a higher initial dose (>4 mg), a lower-dose buprenorphine initiation (≤4 mg) was associated with increased odds of treatment discontinuation within 60 days (adjusted odds ratio (aOR)=1.74, p<0.01), 90 days (aOR=1.73, p<0.01), 120 days (aOR=1.75, p<0.01) and 180 days (aOR)=1.79, p<0.01) of treatment initiation.⁴⁴ Another retrospective study in the US reported a consistent conclusion. The study, which included 17 329 individuals with OUD who initiated buprenorphine treatment indicated that a low initial dose of buprenorphine (≤4 mg) was independently associated with treatment discontinuation within 180 days (adjusted hazard ratio (aHR)=1.72, p<0.001).⁴⁵ However, low-dose inductions of buprenorphine may be beneficial for individuals using fentanyl, as it can reduce the risk of precipitated withdrawal thus increasing the likelihood of retention among clients.¹ However, the scarcity of comparative effectiveness studies on low-dose induction constrains our ability to draw definitive conclusions about its effectiveness. Additionally, although higher MET maintenance doses have been shown to reduce opioid use and increase treatment retention,⁶ limited evidence has demonstrated an association between starting doses and treatment retention. A retrospective study involving 126 clients in MET maintenance treatment reported that there was no difference in treatment retention between initial doses above and below 30 mg.⁹ Due to the widely varying pharmacokinetic profiles of MET, individual metabolism differs significantly in ways that are unpredictable yet crucial to account for.⁴⁶ The literature lacks consensus on standardised therapeutic initial doses of OAT, reinforcing the need for more personalised starting doses.

The starting dosage physicians prescribe may also be influenced by clients’ existing medical conditions such as cardiovascular disease and mental health disorders, as well as the potential risk for toxicity from other medications.⁴ MET may prolong the QTc interval (the measurement of the heart’s QT interval on an electrocardiogram, corrected for heart rate, representing the time required for ventricular depolarization and repolarization), particularly at higher dosages, which elevates the risk of torsade de pointes, a rare but potentially fatal cardiac arrhythmia.47,50 Conversely, BNX has not been associated with QTc prolongation.^{48 51 52} Additionally, individuals receiving OAT with chronic pain were administered a higher average daily dose of buprenorphine compared with those without chronic pain (11.1±7.1 vs 8.9±6.1 mg/day, p=0.04) in a cross-sectional study of 509 individuals with OUD in France.⁵³ However, there is scant evidence regarding initial OAT dosing for individuals with those medical conditions.

Taken together, the evidence used to guide strategies for OAT initial doses among clients with OUD is extremely limited. To our knowledge, no analysis of the potential variance in optimal starting doses for subgroups of individuals based on socioeconomic characteristics, substance use or medical conditions such as cardiovascular disease, chronic pain and health conditions, has been published to date, nor has the evidence base been updated for clients using illicit fentanyl. As such, our objective is to conduct a population-level retrospective observational study with nine linked provincial administrative databases to determine the comparative effectiveness of alternative starting doses on the time from OAT initiation to treatment discontinuation and all-cause mortality, as observed in clinical practice in BC, Canada. Though the proposed timeline of this study (2010–2022) predates the release of the 2023 provincial OUD guidelines, we consider starting dose thresholds from both guidelines within our analysis.

Methods

Study design

We aim to execute a population-level retrospective study to emulate a target trial using a linkage of nine provincial health administrative databases in BC, Canada from 1 January 2010 to 31 December 2022. The study start date was determined based on the availability of one of the administrative databases, the National Ambulatory Care Reporting System (NACRS) Database, which contains data starting from 1 January 2010. While our current data cut is updated to 2024, we will incorporate the most recent data available to us at the time of analysis. We propose to approximate an ‘initiator’ approach (strategy as assigned at initiation regardless of treatment adherence) among incident users (no prior OAT dispensation, which will be verified with historical prescription records dating back to 1 January 1996) and prevalent new users (including incident users and those with repeated attempts) of OAT. Table 1 summarises the key elements of the proposed study design.

Table 1. Key components of the proposed emulated target trial on starting doses of opioid agonist treatment using the observational data.

Component	Theoretical target trial	Emulating using observational data
Eligibility criteria	People ≥18 years old with a diagnosis of opioid dependence and no prior attempt at OAT who were not incarcerated, pregnant or had terminal cancer.	Incident user design: people ≥18 years old with a diagnosis of opioid dependence and no prior attempt at OAT who were not incarcerated, pregnant, had cancer or palliative care or initiated treatment in inpatient settings. Prevalent new-user design: selection criteria above, with no history of OAT within the past month.
Assignment procedures	Individuals are randomly assigned to a starting dose strategy at time zero.	Individuals will be classified into one of the starting dose strategies based on their dispensation records on the date of treatment initiation, with conditional exchangeability sought in subsequent analysis.
Treatment strategies	Starting dosages: dose≤30 mg, 30 mg<dose≤40 mg, 40 mg<dose≤60 mg, 60 mg<dose≤80 mg and dose >80 mg for methadone; dose≤0.5 mg, 0.5 mg<dose ≤4 mg, 4 mg<dose≤8 mg, 8 mg<dose≤12 mg, 12 mg<dose≤16 mg and dose>16 mg for buprenorphine/naloxone; and dose≤150 mg, 150 mg<dose≤200 mg, 200 mg<dose≤300 mg, 300 mg<dose≤400 mg and dose>400 mg for SROM.	Same, measured by the first dose at OAT initiation.
Time zero	Date of treatment assignment.	Date of OAT initiation.
Outcomes	OAT discontinuation; all-cause mortality.	OAT discontinuation will be defined as breaks in days dispensed lasting longer than 5 days for methadone or SROM, and longer than 6 days for buprenorphine/naloxone. All-cause mortality will be observed during the follow-up period; ascertained via linkage to Vital Statistics records.
Follow-up	From the date of randomisation until the earliest of OAT dropout (for outcome of OAT discontinuation), death, loss to follow-up or end of study period.	From the date of OAT initiation until the earliest of OAT dropout (for outcome of OAT discontinuation), death, loss to follow-up (defined as no records over 66 months after the last service date—for outcome of mortality) or the end of the study period.
Causal contrasts	Intention-to-treat analysis.	Initiator analysis (among OAT initiators instead of clients who intended to treat)
Analysis plan	Kaplan-Meier curve and Cox-PH regression models estimating associations between treatment strategies and OAT discontinuation and all-cause mortality.	PS (ie, IPTW and alternative hdPS) and IV, alongside Cox-PH regression models, will be used to control observed and unobserved confounders, respectively.

BC, British Columbia; BNX, buprenorphine/naloxone; Cox-PH, Cox-proportional hazard; hdPS, high-dimensional propensity score; IPTW, inverse probability of treatment weighting; IV, instrumental variable; MET, methadone; OAT, opioid agonist treatment; PS, propensity score; SROM, slow-release oral morphine.

We will use the BC PharmaNet database⁵⁴ (prescription dispensation records) to define OAT episodes. We note that dispensations provided to individuals in federal correctional facilities and during hospitalisation are not included in this database. This database is linked with data from eight population-level administrative databases, including the Discharge Abstract Database⁵⁵ (hospitalisation records), Client Roster⁵⁶ (demographic and geographical information), Medical Services Plan⁵⁷ (physician billing records), NACRS Database⁵⁸ (emergency department visits), BC Provincial Corrections⁵⁹ (imprisonment and release), Vital Statistics⁶⁰ (death records and their underlying causes), Perinatal Care Database⁶¹ (maternal and infant healthcare) and the BC Social Development and Poverty Reduction database⁶² (records of social supports). An individual’s personal health number serves as the unique identification linking these databases.⁶³

Study population

The study will include individuals aged 18 or older who have no history of cancer or palliative care, no current incarceration, no known pregnancy (up to 3 months of post partum) at OAT initiation and who initiated OAT with either MET, BNX or SROM between 1 January 2010 and 31 December 2022. Individuals with known pregnancy at OAT initiation will be excluded from the analysis, as there are different care protocols for pregnant women compared with the general population with OUD. Additionally, since our data do not capture OAT receipt in inpatient settings, we will exclude episodes directly preceded by hospitalisation (date of discharge is within 5/6 days of OAT initiation for MET or BNX, respectively) and assume individuals who began OAT prior to admission continued their treatment through discharge. Treatment episodes where multiple forms of OAT are initiated concurrently will be excluded from the study.

Study follow-up

Time zero (ie, aligned time of treatment assignment, treatment start and start of follow-up) is defined as the date of treatment initiation and start of follow-up, operationalised as the date of the first OAT dispensation from a community-based pharmacy. Incident users will include individuals who initiated OAT for the first time during the study period, with no prior OAT dispensations dating back to 1 January 1996. A 1 month (30 days) washout period will be applied in a prevalent new user analysis (ie, no OAT dispensations received in the 30 days prior to the episode initiation). All analyses will be stratified on the medication received at treatment initiation. Study follow-up will conclude at either OAT discontinuation (only for the outcome of OAT discontinuation), death or the end of the study follow-up (31 December 2022), whichever occurs first.

Key measures

The primary exposure will be the OAT dosage on the date of the first treatment dispensation. All analyses will be stratified by medication received at OAT initiation. Pending sufficient sample sizes, we aim to compare starting doses according to international guidelines, as follows:

MET: dose≤30 mg (reference), 30 mg<dose≤40 mg, 40 mg<dose≤60 mg, 60 mg<dose≤80 mg and dose>80 mg.

BNX: dose≤0.5 mg (per dose, considering ≤1 mg/day daily dosages to capture micro-dosing practices as total daily dispensation is recorded in the database), 0.5 mg<dose≤4 mg (reference), 4 mg<dose≤8 mg, 8 mg<dose≤12 mg, 12 mg<dose≤16 mg and dose>16 mg.

SROM: dose≤150 mg (reference), 150 mg<dose≤200 mg, 200 mg<dose≤300 mg, 300 mg<dose≤400 mg and dose>400 mg. The reference groups will be selected based on guideline recommendations. We will compare distributions of covariates between exposure categories a priori to ensure positivity, that is, in all confounder levels, there must be a non-zero probability of every individual receiving any level of the starting doses.

We propose two primary outcomes: OAT discontinuation and all-cause mortality. OAT discontinuation is defined as breaks in dispensed doses lasting 5 days or more for MET or SROM and 6 days or more for BNX. The 5/6 day interval for treatment discontinuation was selected to align with BC guidelines recommending reversion to starting doses after these durations of interrupted treatment.⁶⁴ All-cause mortality will be assessed during the follow-up period, regardless of treatment discontinuation by linking to Vital Statistics records.⁶⁵

Analysis plan

We describe our analysis as an ‘initiator’ analysis because our study is among OAT initiators; we cannot observe the intended treatment or dosage when a prescription is written but no medication is dispensed. One of the main challenges in determining the causal relationship between OAT starting dose and outcome is accounting for the various factors that influence the choice of the starting dose, that is, confounding by indication. We will balance risk factors affecting the starting dose at baseline and focus on an individual’s outcome at the end of follow-up, adjusting for selection bias.⁶⁶ We propose to employ propensity score (PS) weighting to adjust for measured confounders and instrumental variable (IV) analysis as a sensitivity analysis to address unmeasured confounders that may affect the selection of alternative OAT initial dosing. Robust variance estimates will be used in the prevalent new user design to control for repeated treatment attempts. Consistency between these methods will support the robustness of our conclusions and strengthen our inferences. Inconsistencies may indicate the need for further research, potentially through experimental studies. All analyses will start in September 2025 and end by May 2026.

Propensity score estimation

PS estimation is a statistical technique designed to reduce bias due to confounding in estimating treatment effects in observational studies. It relies on the assumption that all confounders influencing both treatment assignment and outcomes are measured.⁶⁷ We propose to use PS weighting (inverse probability of treatment weighting, IPTW), as it can handle exposures with more than two categories. IPTW reweighs the population based on the inverse probability of receiving the starting dose actually assigned, given the observed baseline covariates, to create a pseudo-population where the assignment of initial doses is independent of observed baseline covariates at OAT initiation.

Applications using investigator-selected covariates have shown this approach adjusts for confounding as effectively as traditional multiple regression methods, though residual confounding is possible with both methods. We propose a high-dimensional propensity score approach (hdPS) as an alternative to PS estimation. We will employ the hdPS approach to generate a large number of empirically-derived covariates specific to each outcome from data primarily collected for billing and routine administrative tasks in the Medical Services Plan (International Classification of Diseases (ICD)-9 codes), NACRS (ICD- 10 codes), Discharge Abstract Database (DAD) (ICD-9/10 codes, inpatient procedure codes) and PharmaNet databases (drug dispensed). We will exclude proxies that have very low prevalence and a minimal likelihood of introducing bias.^{68 69} We will use multiple logistic regression models to calculate PSs, adjusting for both selected investigator-specified covariates and proxy variables identified as important by the hdPS algorithm. Stabilised weights, truncated at the 99th percentile, will be used to balance the baseline covariates for both the IPTW approach and the hdPS approach.⁷⁰ The hdPS algorithm will be implemented using the SAS macro available at www.drugepi.org.⁶⁹

Instrumental variable estimation

IV estimation is designed to control for unobserved confounding when selection into a treatment group (eg, those starting with <40 mg MET at OAT initiation) is influenced by factors that may not be observed. A systematic review of preference-based IVs in health services found that the most commonly used IVs are facility-level, physician-level and regional-level preferences in comparative effectiveness studies when the association between prescribing preference and treatment selection is strong and interpretable.71,75 We propose to use the best-fitting IV from six facility-level and physician-level preference-based IVs on prior initiation dose preferences to adjust for unobserved confounders affecting OAT starting doses. Continuous IVs will be defined as the proportion of a prescriber’s (or treatment facility’s) client load over the past 12 months who initiated a low (≤30 mg for MET, ≤0.5 mg for BNX and ≤150 mg for SROM) or a high (>60 mg for MET, >16 mg for BNX and >400 mg for SROM) starting dose. For facility-level IVs, we will implement social network analysis⁷⁶ to create a network with connections between prescribers based on their shared OAT clients. We will identify the prescribers who were within the same network in the year of the individual’s OAT initiation and use the history of the prevalent new OAT clients of those prescribers within 12 months prior to the individual’s OAT initiation. We also propose high-preference/moderate-preference/low-preference physician/facility groups as IVs based on the quartiles of prescribing preference for a high/low starting dose among the physicians/facilities within 12 months prior to the time the physician initiated the individual on OAT.

IVs must satisfy three key assumptions: (1) they must be related to the client’s exposure assignment; (2) they must not have an independent influence on the outcome other than through the exposure (exclusion restriction assumption); and (3) they must not be confounded by unmeasured common factors related to the outcome (the IV is not itself confounded).^{75 77} To validate our IVs, we will conduct F-tests from the first-stage regression to empirically assess assumption 1, and rely on the expertise of a scientific advisory committee to evaluate assumption 2.^{71 72 75} For assumption 3, we will demonstrate empirically that the IVs are not associated with selected covariates.^{71 72 75}

For both primary outcomes, we will present HRs obtained from a Cox proportional hazard model (which will include the dosing strategy variable and covariates) with IPTW or estimated with IV analysis. We will then compute the subject-specific predicted time-to-event (survival) probabilities under the weighted Cox proportional hazard model, and then average the survival probabilities across individuals by exposure category. We will evaluate the proportional hazards assumption of the Cox proportional hazards model using statistical tests and graphical diagnostics based on scaled Schoenfeld residuals.⁷⁸ The absolute risk of outcome (cumulative incidence) will be calculated as one minus the probability of survival, and the corresponding risk differences will then be calculated. We will use sandwich variance estimators to conduct inference for HRs, while using a non-parametric bootstrap of 500 samples to calculate 95% compatibility intervals (‘confidence intervals’, CIs) for risk differences and cumulative incidence. We will employ 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, we will appropriately refrain from suggesting a 95% certainty that the true value lies within the interval.⁷⁹

Covariate selection

We previously identified potential confounders by conducting a systematic literature review for articles published up to 2 September 2019 on factors associated with OAT retention.⁷⁷ We propose to augment the listing with additional variables from recent research up to 25 March 2024 (table 2). The variable list includes socio-demographics (sex, age at OAT initiation, region (urban versus rural), unstable housing, receipt of income assistance), medical conditions (asthma or Chronic Obstructive Pulmonary Disease (COPD), dispensation of psychiatric or sedative medication, dispensations of opioids other than OAT, non-opioid substance use disorder, alcohol use disorder, tobacco use disorder, mental health conditions, hepatitis C, chronic pain, the Charlson Comorbidity Index, drug-related acute care visits), OAT-related variables (attachment to the OAT prescriber, urine drug test receipt) and COVID-19-related measures (use of virtual care, receipt of OAT by pharmacy delivery, concurrent opioid prescription) (figure 1). To account for potential differences in initial OAT dosages prescribed at withdrawal management facilities compared with other outpatient settings, social network analysis will be used on historical data to identify withdrawal management sites, which will be identified on the basis of caseload and indication that clients exclusively transfer to other clinics within 3 weeks of OAT initiation. We will also adjust for COVID-19 restrictions using the COVID-19 stringency index developed by the Bank of Canada.⁸⁰ These variables will be defined at OAT initiation as time-fixed covariates.

Table 2. Factors associated with opioid treatment retention.

Covariate	Association	Quality of evidence* (source)	Available in our database?
Individual-related characteristics
Demographics
Age	+ MET retention	Level I85	Yes
Marital status (married)	+ MET retention	Level I85	No
Employment status (employed)	+ MET retention	Level I85	Yes
Gender (female)	+ MET retention	Level I85	Yes
Number of treatment episode	− MET/BNX retention	Level II86,88	Yes
Duration of treatment in an episode	+ MET retention	Level I85	Yes
Cumulative days of OAT	+ MET/BNX retention	Level II (HIV population)89	Yes
Ethnicity (Hispanic or African American)	− BUP retention	Level II90	No
Living in rural area	− MET retention	Level II91	Yes
Move house/transition	+ BUP retention	Level II92	Yes
Family history of addiction	− MET retention	Level II93	No
Homelessness	− MET/BNX retention	Level II94	Yes
Incarceration	− MET/BNX retention	Level II94	Yes
Community density			Yes
Clinics density in community	Risk factor for overdose	Level II95	Yes
History of overdose	Risk factor for overdose	Level III96	Yes
Numbers of overdose	Risk factor for overdose		Yes
Concurrent conditions
Psychiatric comorbidity: major depression	+ BUP retention	Level II97	Yes
Serious mental illness	+ MET/BUP retention	Level II98 99	Yes
Schizophrenia	− BUP retention	Level II97 100	Yes
Personality disorders	− BUP retention	Level II97	Yes
Severe withdrawal at beginning of treatment	− BUP retention	Level I101	No
Hepatitis C virus	+ BUP retention	Level II94	Yes
HIV	− MET/BNX retention	Level II102	Yes
Other substance use disorders	− BUP retention	Level II103	Yes
Severe chronic pain	Risk factor for overdose	Level III96	Yes
Respiratory disease	Risk factor for overdose	Level III96	Yes
Cocaine use on admission to OAT	− BNX retention	Level II104	No
Past-month injection drug use	− BNX retention	Level II105	No
Fentanyl use	− MET retention	Level II16	Yes
Medication history
Use of sedatives within past 30 days of OAT	− BUP retention	Level II106	Yes
Previous receipt of BUP	+ BUP retention	Level II107	Yes
Receipt of psychiatric medication	+ BUP retention	Level II108	Yes
Receipt of prescription opioids	+ MET/ BNX retention	Level II99 100 102	Yes
Receiving high opioid prescription doses	Risk factor for overdose	Level III96	Yes
Receipt of benzodiazepine	+ MET/BNX retention	Level II99 100 109	Yes
Receipt of Z-drug	+ MET/BNX retention	Level II110	Yes
Receipt of stimulants	− MET/BNX retention	Level II99 109	Yes
Reverting dosages after missing dosing	− MET/BNX retention		Yes
Healthcare utilisation
Emergency department visits	− BUP retention	Level II99 100 103	Yes
Psychiatric hospitalisations	− BUP retention	Level II103	Yes
All-cause hospitalisations	− BUP retention	Level II100 102	Yes
Treatment-related and contextual factors
Service provision
OAT in integrated care	+ BUP retention	Level I111	Yes
Behavioural therapy	+ BUP/MET retention	Level I112 113	Yes
Positive relationships with service staff	+ MET retention	Level II114	No
Prescriber’s social network strength			Yes
Contextual factors
Poor availability and quality of heroin in drug supply	+ MET/BUP retention	Level II115	No
COVID-19 pandemic	No difference	Level II116	Yes
Telemedication during COVID-19	+ BUP retention	Level II117	No
Virtual care/telehealth	− Risk factor for overdose	Level II118	Yes
Receipt of prescriber safer supply	+ MET/BNX retention	Level II119	Yes
OAT site
Community-based health centre	+ MET retention	Level II120	Yes
Physician’s office-based setting	+ BUP retention	Level III121	Yes
Clinic onsite pharmacy (compared with community offsite pharmacy)	+ MET/ BUP retention	Level II120 122	Yes
Mixed sites	+ MET retention	Level II87	Yes
Patient loads	+ MET retention	Level II123	Yes

^*Quality of evidence ratings: level I: systematic reviews, meta-analyses and randomised controlled trials; level II: cohort studies, case–control studies, case studies; level III: case reports, ideas, editorials, opinions (source: Cochrane review library https://consumers.cochrane.org/levels-evidence).

−, negative association; +, positive association; BC, British Columbia; BNX, buprenorphine/naloxone; BUP, buprenorphine; MET, methadone.

Figure 1. Directed acyclic graph detailing the relationship between starting dose and outcome for individuals who initiated OAT in British Columbia, Canada Panel A represents the scenario in which we assume there is no unmeasured confounding in the primary analysis. Panel B shows the scenario in which unmeasured confounding exists, and we propose to use the instrumental variable approach to eliminate it in sensitivity analysis. (A) Exposure strategies of starting dose of OAT assigned at time zero; Y: outcomes including OAT discontinuation and all-cause mortality; L₀: any baseline confounders that are time-fixed and that impact OAT starting dose and outcomes, including age at OAT initiation, sex, receipt of urine drug testing, carry receipt, unstable housing, receipt of income assistance, mental health conditions, alcohol use disorder, non-opioids substance use disorder, HCV, chronic pain, tobacco use disorder, sedative medication, prescription opioid, psychiatric medication, Charlson Comorbidity Index, Chronic Disease Score, drug overdose in the last 30 days, drug-related hospitalisation or emergency department visits, psychiatric hospitalisations, incarceration in a provincial corrections facility in the past year, physician attachment, receipt of an opioid-based medication for pain, use of virtual care, receipt of treatment by delivery and whether prescribed safer supply; Z: preference-based instrumental variables; U: unmeasured confounders that may impact OAT starting dose and outcomes and may be affected by baseline confounders, such as individuals’ ongoing illicit drug usage. We do not observe the unmeasured confounders and plan to control their impact via the preference-based instrumental variable. HCV, Hepatitis C virus; OAT, opioid agonist treatment.

Subgroup and sensitivity analysis

We propose a range of subgroup and sensitivity analyses to assess the robustness of our results and explore the heterogeneity of treatment effects among key subgroups. Our predetermined objectives focus on restricting the study population, restricting the time frame of the study, reclassifying key exposures and outcomes and re-specifying the analytical model (table 3). As individuals with concurrent medical conditions may have different initial dosing requirements, we propose to include subgroup analyses for people with concurrent mental health conditions, cardiovascular disease and chronic pain. We also propose to include individuals with a history of prescription opioid use prior to OUD diagnosis. Additionally, we plan to restrict the study follow-up period to 17 March 2020, which provides a comprehensive follow-up history of individuals prior to the COVID-19 pandemic. If sample sizes are sufficient for model convergence, we will further restrict a sensitivity analysis to between 17 March 2020 and 31 December 2022 to confirm how the impact of initial dosing strategies may have changed after COVID-19. Analyses will otherwise be stratified by OAT initiation at inpatient withdrawal management sites. Additionally, we consider two secondary outcomes: completed induction (reaching a minimum of 2 weeks of continuous OAT with no dose changes) and overdose-related acute care visits. Further, we propose hdPS and instrumental variable analyses as described previously among initiators. We plan to truncate follow-up at different times, since the initial dose may be important for a short while after time zero, the treatment effects may depend heavily on the time of follow-up. We also propose no treatment weights truncation for IPTW and hdPS approach. Relevant findings from each analytic approach will be presented in a forest plot. Any deviations from this protocol will be documented in final reports.

Table 3. Proposed subgroup and sensitivity analyses.

Proposed sensitivity analysis	Rationale
1. Population restriction
PWOUD with chronic pain	To account for individuals with prior indications of medical conditions may require different initial dosing strategies based on care requirements.
PWOUD with cardiovascular diseases
PWOUD with a history of mental health conditions
Individuals with a history of prescription opioid use prior to OUD diagnosis	Prescription opioid use may influence tolerance of OAT and affect dosing requirements.
Individuals initiating OAT in inpatient withdrawal management facilities	To account for individuals initiating OAT for rapid detoxification.
Exclude individuals initiating OAT in inpatient withdrawal management facilities	To remove individuals initiating OAT for rapid detoxification.
OAT episodes without a washout period	To account for episodes initiated within 30 days of previous OAT dispensation.
OAT episodes with a 14-day washout period	To assess the sensitivity of gap times between episodes and account for any changes that may be affected by more recent OAT experience and capturing different phases of treatment adherence.
OAT episodes with a 90-day washout period
2. Timeline restriction
The date of the first death for which fentanyl was detected in the province (1 April 2012)	Fentanyl in the illicit drug supply may lead to effects on OAT dropout and OAT starting doses.
The date of the BC public health opioid overdose emergency (14 April 2016)	The opioid-related deaths have increased since 2016, which may affect OAT starting doses.
Restrict timeline prior to the COVID-19 pandemic (Until 17 March 2020)	To explore the robustness of our finding prior to the COVID-19 era.
During the COVID-19 pandemic (from 18 March 2020)
3. Outcome re-classification
Episode discontinuation defined as at least 14 days without OAT	Alternative discontinuation thresholds have been defined in other studies and guidelines.124,126
Episode discontinuation is defined as at least 30 days without OAT
All-cause mortality on treatment	To account for the results that are not sensitive to the defined time during specific treatment phase.15
Truncation at 1 month of follow-up	To account for the treatment effects will depend strongly on follow-up time since the first dose.
Truncation at 3 months of follow-up
Truncation at 12 months of follow-up
Second outcome: success of completed induction (reaching a minimum of 2 weeks of continuous OAT with no dose changes)	To assess the effect of different starting doses on completed induction.
Second outcome: overdose-related acute care visits	To assess the effect of different starting doses on time to overdose-related emergency department visits.
4. Model specification
Treatment weights using hdPS	To account for potential uncontrolled confounding or mismeasured confounding by accounting for proxies (empirical covariates).
No treatment weights truncation	To confirm effects of baseline analysis, despite weight instability associated with non-truncation.
Prescriber preference-based IV approach	To account for unobserved confounders at baseline.
5. Bias analysis
Calculating e-value	To measure the association necessary to explain the observed exposure–outcome association attributable to unmeasured factors identified.127

BC, British Columbia; hdPS, high-dimensional propensity score approach; IV, instrumental variable; OAT, opioid agonist treatment; OUD, opioid use disorder; PWOUD, people with opioid agonist treatment.

Limitations

While we will employ target trial emulation, our study has several limitations. First, we will exclude individuals who initiated OAT in inpatient settings as dosing information in acute care settings is unavailable. Second, we can only assess the universe of strategies executed and observed in clinical practice within the study setting and timeline. Additionally, we cannot rule out the potential for uncontrolled confounding (eg, amount of illicit opioid consumption at OAT initiation). Nevertheless, we propose the instrumental variable as a sensitivity analysis to adjust for unmeasured confounding. Finally, we will use an ‘initiator’ approach rather than an intention-to-treat approach, thus excluding individuals who are prescribed, but never receive treatment; inferences are thus conditional on treatment receipt.

Ethics and dissemination

The BC Ministries of Health and Mental Health and Addiction provided the research team with access to this connected database as a response to the opioid overdose public health crisis in the province. The database was designated as a quality improvement initiative. The analysis conducted by the Providence Health Care Research Institute and the Simon Fraser University Office of Research Ethics met the criteria for exemption as stated in Article 2.5 of the Tri-Council Policy Statement: Ethical Conduct for Research Involving Humans.⁸¹

This study will adhere to international guidelines for conducting and reporting research, including the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.⁸² Additionally, a multidisciplinary scientific advisory committee will use the ‘Risk of Bias in Non-Randomized Studies-of Interventions’ tool for post-study evaluation. The findings will be published both electronically and physically in peer-reviewed academic journals. The study aims to produce robust evidence on the safety and effectiveness of different starting doses at OAT initiation in real-world settings over the long-term, with the goal of enhancing OAT retention⁸³ for these essential and life-saving⁸⁴ medications.

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.

Data availability statement

Study datasets: Not available. Statistical code: Available from Dr Bohdan Nosyk (bnosyk@sfu.ca)