First-line opioid agonist treatment as prevention against assisting others in initiating injection drug use: A longitudinal cohort study of people who inject drugs in Vancouver, Canada

Highlights • Most first-time injections involve assistance from people who inject drugs.• Having an untreated opioid use disorder may increase risk of providing assistance.• First-line opioid agonist treatment could therefore reduce assistance provision.• We found evidence to support this presumed protective treatment effect.• Effect magnitude uncertain due to imprecise estimation and observed heterogeneity.


Introduction
People who inject drugs are at increased risk of overdose, HIV, hepatitis C, and severe bacterial infections (Backmund et al., 2005;Brugal et al., 2002;Darke, 2003;Degenhardt et al., 2017Degenhardt et al., , 2016Garfein et al., 1996;Noroozi et al., 2020;Tagliaro et al., 1998;van Haastrecht et al., 1996), especially within the first few years after they begin injecting (Garfein et al., 1998;Goldsamt et al., 2010). Drug injecting is generally learned by injection-naïve people who use drugs through observing and interacting with people who inject drugs in shared social circles and drug use environments (Harocopos et al., 2009;Rhodes et al., 2011;Strike et al., 2014;Vashishtha et al., 2017;Werb et al., 2018). By engaging in injection-promoting behaviours (including injecting in front of injection-naïve individuals and speaking positively about injecting), people who inject drugs can normalize injecting and increase its appeal to others (Crofts et al., 1996;Des Jarlais et al., 2021;Frajzyngier et al., 2007;Harocopos et al., 2009;Khobzi et al., 2009;Small et al., 2009;Strike et al., 2014;Witteveen et al., 2006). Repeated exposure to injection practices combined with additional factors, such as increased drug tolerance and dependence, may influence injection-naïve people who use drugs to consider injecting and seek help initiating (Bryant and Treloar, 2007;Crofts et al., 1996;Strike et al., 2014;Witteveen et al., 2006). People who inject drugs may further facilitate others' transitions to injecting by providing assistance with first-time injections, either directly (i.e., they inject the initiate) or indirectly (i.e., they explain, describe, or demonstrate how to inject to an initiate who then injects themselves for their first time) (Des Jarlais et al., 2019;Des Jarlais et al., 2021;Gicquelais et al., 2020;Uusküla et al., 2018). In fact, between 76 and 90% of sampled people who inject drugs in Canada and the US report that they received assistance from an experienced person who injects drugs in initiating injecting, with most (range=55-86%) having been directly injected by someone else (Gicquelais et al., 2020).
Both frequent and public injecting have been associated with engaging in injection-promoting behaviours, being asked to help someone inject, and ultimately providing injection initiation assistance among people who inject drugs (Bluthenthal et al., 2015a(Bluthenthal et al., , 2015bMelo et al., 2018;Mittal et al., 2019b;Navarro et al., 2019;Rotondi et al., 2014;White et al., 2020). Injecting frequently in public spaces can increase the visibility of one's injecting practices to injection-naïve individuals, potentially resulting in more requests and opportunities to assist with first injections (Des Jarlais et al., 2021;Melo et al., 2018;Mittal et al., 2019b;Navarro et al., 2019;White et al., 2020). Though people who inject drugs consistently describe a group ethic or "moral code" against helping others initiate injecting due to injection-related harms, some report reluctantly providing such assistance in exchange for drugs (or money to obtain drugs) to meet their own drug use needs, including avoiding or mitigating withdrawal symptoms (Goldsamt et al., 2010;Guise et al., 2018;Guise et al., 2017;Kolla et al., 2015;Mittal et al., 2019a;Olding et al., 2019;Rhodes et al., 2011;Simpson et al., 2020). This commonly expressed motivation may further explain why frequent injecting, which is correlated with increased drug tolerance and dependence, is positively associated with providing injection initiation assistance Simpson et al., 2020;Werb et al., 2018).
The PReventing Injecting by Modifying Existing Responses (PRIMER) study is a consortium of existing cohorts across North America (including Vancouver, Canada; San Diego, US; and Tijuana, Mexico) that aims to identify interventions that might reduce the likelihood that people who inject drugs provide injection initiation assistance (Werb et al., 2016). Initial PRIMER evidence suggests that opioid agonist treatment (OAT) with methadone or buprenorphine/naloxone-the recommended first-line oral pharmacotherapies for opioid use disorder in Canada and many other countries (Bruneau et al., 2018;Canadian Research Initiative in Substance Misuse (CRISM), 2018; Centre for Addiction and Mental Health (CAMH), 2021; Dunlap and Cifu, 2016;Volkow et al., 2014)-might offer such a benefit (Marks et al., 2019(Marks et al., , 2021bMeyers-Pantele et al., 2022;Mittal et al., 2017Mittal et al., , 2019b. Specifically, prior cross-sectional studies of Vancouver-based people who inject drugs found that participants enrolled in first-line OAT within the past six months had reduced odds of helping someone initiate injecting over the same period (Marks et al., 2019(Marks et al., , 2021bMittal et al., 2019b). These observed associations are presumably owed to the known effectiveness of methadone and buprenorphine/naloxone in alleviating opioid cravings and preventing withdrawal symptoms in treated people who inject drugs, leading to reductions in non-medical opioid (and overall) injecting (Gowing et al., 2011;Karki et al., 2016;Mattick et al., 2009;Mittal et al., 2019b;Volkow et al., 2014). Due to these primary benefits, people who inject drugs on first-line OAT may be less likely to expose others to their injecting practices (because of reductions in overall and public injecting) versus untreated individuals, resulting in fewer requests from injection-naïve individuals to assist them in initiating injecting (Des Jarlais et al., 2021;Kolla et al., 2015;Mittal et al., 2019b). Furthermore, by preventing opioid cravings and withdrawal, people who inject drugs on a stable OAT dose may be less vulnerable to accepting requests to provide injection initiation assistance in exchange for drugs or money to procure drugs (Des Jarlais et al., 2021;Kolla et al., 2015;Mittal et al., 2019b). As an estimated 73% of people who inject drugs in North America primarily inject opioids but only 20% are on OAT Larney et al., 2017), these preliminary findings suggest that increasing OAT coverage in this population could substantially limit injection drug use initiation, in addition to reducing injecting and related harms in treated individuals (Marks et al., 2021b(Marks et al., , 2019Meyers-Pantele et al., 2021;Mittal et al., 2019b).
Though a secondary preventative effect of first-line OAT on injection initiation assistance provision in people who inject drugs seems plausible, further investigation is warranted due to limitations of the initial studies assessing this relationship (Marks et al., 2019(Marks et al., , 2021bMeyers-Pantele et al., 2022;Mittal et al., 2017Mittal et al., , 2019b. First, prior studies were not restricted to individuals with a suspected opioid use disorder, meaning OAT-ineligible participants were likely included (Marks et al., 2019(Marks et al., , 2021bMeyers-Pantele et al., 2022;Mittal et al., 2017Mittal et al., , 2019b. Second, the initial evidence is from cross-sectional analyses involving contemporaneous confounder, exposure, and outcome measures (Marks et al., 2019(Marks et al., , 2021bMeyers-Pantele et al., 2022;Mittal et al., 2017Mittal et al., , 2019b. Without clarity in the temporal sequence of these measures, there are other plausible explanations for previously observed associations beyond treatment preventing injection initiation assistance provision, including reverse causality (Marks et al., 2019(Marks et al., , 2021bMeyers-Pantele et al., 2022;Mittal et al., 2017Mittal et al., , 2019b. Third, earlier studies used participant-reported measures of both OAT and injection initiation assistance provision but did not account for misclassification bias (e.g., inaccurate recall and socially desirable responding) in these measures (Bouck et al., 2022).
In this study, we aimed to estimate the short-term effect of first-line OAT on injection initiation assistance provision within a cohort of people who inject drugs in Vancouver, Canada, while addressing the aforementioned issues by restricting to participants with suspected opioid use disorder; using longitudinal data to establish temporality between confounder, exposure, and outcome measures; and adjusting for misclassification in participant-reported treatment and outcome measures. We hypothesized that participants currently on first-line OAT with methadone or buprenorphine/naloxone would be less likely, on average, to help someone initiate injecting in the next six months compared to participants not on any medication for opioid use disorder.

Data sources
We used questionnaire data collected from December 2014-May 2018 on participants from three prospective, open, and linked cohorts in Vancouver, British Columbia: the Vancouver Injection Drug Users Study (VIDUS); the AIDS Care Cohort to Evaluate exposure to Survival Services (ACCESS) study; and the At-Risk Youth Study (ARYS). This period corresponds with VIDUS/ACCESS/ARYS survey cycles 18-24 and overlaps with prior studies of OAT and injection initiation assistance provision in Vancouver (Marks et al., 2019(Marks et al., , 2021bMittal et al., 2019b). Beyond living in the Vancouver area and providing informed consent, inclusion criteria were: for VIDUS, that participants be ≥18 years old, HIV-negative, and report injecting drugs in the past month; for ACCESS, that participants be ≥18 years old, HIV-positive, and report illicit drug use in the past month; and for ARYS, that participants be 14-26 years old, "street-involved" (recently homeless or used street youth services), and report illicit drug use in the past month (Werb et al., 2016). At enrollment and semi-annual visits thereafter, VIDUS/ACCESS/ARYS participants complete harmonized interviewer-administered questionnaires that collect information on their sociodemographic characteristics, drug use and related behaviours, and history of treatment for substance use disorders (Werb et al., 2016). Under PRIMER, survey items asking about participants' experiences helping others initiate injecting were added to VIDUS/ACCESS/ARYS questionnaires in December 2014 (Werb et al., 2016). This study was approved by the University of California San Diego Human Research Protection Program, the University of British Columbia/Providence Health Care and the University of Toronto Research Ethics Boards (#27433).

Participants
We included VIDUS/ACCESS/ARYS participants who attended a visit in each of the first two survey cycles involving PRIMER questions, i.e., cycles 18 (December 2014-May 2015) and 19 (June 2015-November 2015. Hereafter, we refer to the first and second of these visits by date as a participant's 'look-back' and 'baseline' visits, respectively. We then restricted to those participants who, at their baseline visit, reported ever injecting drugs and frequent (i.e., ≥weekly) non-medical opioid use via injection and/or non-injection over the past six months. These criteria were intended to select participants with injecting experience (at-risk for providing injection initiation assistance) and a probable opioid use disorder (presumably OAT-eligible) (Socías et al., 2018). To facilitate comparisons between individuals on first-line OAT versus no OAT, we excluded participants who reported OAT with an alternative opioid agonist medication including slow-release oral morphine, hydromorphone (oral or injectable), or injectable diacetylmorphine at their look-back visit or within the six months preceding, and inclusive of, their baseline visit (Appendix 1) (Oviedo-Joekes et al., 2016). Last, we excluded participants with missing baseline exposure, covariate, or outcome data (see Measures section).
Each participant contributed up to 5 follow-up visits from December 2015-May 2018, beginning with their first semi-annual visit after baseline and ending with the last visit they attended before (1) their first missed outcome measurement (due to missed visit, invalid or nonresponse), (2) their first visit where they reported recent OAT with an alternative opioid agonist medication (Appendix 1), or (3) June 2018 (administrative censoring), whichever came first.

Measures
Current first-line OAT was assessed at all visits and defined as a "yes" response to the question: "In the last six months, have you been in any kind of alcohol or drug treatment?" with specification of enrollment, as of the interview, in OAT with methadone or buprenorphine/naloxone (Bouck et al., 2022;Socías et al., 2019;Socías et al., 2017). Recent injection initiation assistance provision was measured at all visits and defined as a "yes" response to the question: "In the last six months, have you helped anybody inject who had never injected before?" (Werb et al., 2016).
We additionally measured time-independent and time-varying covariates that were presumed to confound the relationship between first-line OAT and subsequent injection initiation assistance provision and influence the likelihood of premature censoring during follow-up ( Fig. 1). Measured time-independent baseline covariates included: age (in years; treated as continuous), self-identified gender (cisgender man; yes/no), and cohort (VIDUS, ACCESS, or ARYS) (Ben Hamida et al., 2018;Kerr et al., 2005;Marks et al., 2019Marks et al., , 2021bMelo et al., 2018;Meyers-Pantele et al., 2021). Measured time-varying covariates, which were assessed at baseline and follow-up visits, included: recent homelessness; recent incarceration (jailed, imprisoned, or detained minimally overnight); indicators of recent income from the following sources (note: multiple sources could be selected): paid legal work, social assistance, street-based activities (i.e., panhandling, squeegeeing, or recycling), sex work, or illegal activities (i.e., drug dealing, theft, or other criminal activity); recent non-fatal overdose; recent frequency (daily, weekly, less-than-weekly, or none) of both opioid and non-opioid (e.g., cocaine, crystal methamphetamine, benzodiazepines) injection drug use; and recent public injection (Bouck et al., 2020;Bowles et al., 2021;Kerr et al., 2005;Luongo et al., 2017;Marks et al., 2021a;Mittal et al., 2019b;Navarro et al., 2019;White et al., 2020). Covariates qualified as 'recent' reflect behaviours, experiences, and/or activities over the past six months. If a time-varying covariate value was missing (true for <0.5% of uncensored person-visit observations), we carried the last observed value forward.

Statistical analysis
To evaluate our hypothesis and account for repeated outcome measurements, we fit a weighted log-binomial model using generalized estimating equations with an unstructured working correlation matrix for within-participant responses (Ballinger, 2004;Hernán et al., 2002;Hernán and Robins, 2020):  Robins et al., 2000); C it =0 reflects that the analysis was restricted to uncensored person-visit observations (Moodie et al., 2008); β 0t represents visit-specific intercepts; and β 1 expresses the focal exposure-outcome association on the log probability scale. Assuming exchangeability (no unmeasured confounding and ignorable censoring), consistency (treatment levels under comparison correspond with well-defined interventions and measured levels), positivity (participants have non-zero probabilities of being treated or untreated conditional on their measured treatment and covariate histories), and no weight or outcome model misspecifications, the model 1 coefficient β 1 is an unbiased estimate of the causal parameter γ 1 from the following repeated measures marginal structural model (Cole and Hernán, 2008;Hernán et al., 2002;Hernán and Robins, 2020;Robins et al., 2000): where Y ait− 1 it represents a participant's potential outcome at visit t had their treatment at visit t-1 been set, possibly contrary to their observed exposure, to a it− 1 (current OAT: 1=yes, 0=no) and exp(γ 1 ) is the treatment effect of interest-expressed as a relative risk (RR)-which compares the risk of recent injection initiation assistance provision at visit t had everyone (all uncensored participants by visit t) been on firstline OAT at visit t-1 versus had everyone not been on any medication for opioid use disorder at visit t-1, conditional on everyone having the same treatment status at visit t-2 (Hernán et al., 2001;Hernán and Robins, 2020;Keogh et al., 2018;Robins et al., 2000). A 95% confidence interval (CI) was calculated using a robust, sandwich-type variance estimator (Hernán et al., 2002). In anticipation of a rare outcome, we analyzed repeated outcome measures to increase precision in estimation ( Van-derWeele et al., 2011).
To account for confounding and non-ignorable censoring (i.e., differential loss-to-follow-up) by measured time-independent and timevarying covariates, we fit model 1 using inverse probability-oftreatment-and-censoring (IPTC) weights (Cole and Hernán, 2008). Informally, the IPTC weight for participant i at visit t was inversely proportional to that participant's probability of remaining uncensored by visit t and receiving their reported treatment at visit t-1 conditional on their observed treatment and covariate histories (Cole and Hernán, 2008). Formally, the IPTC weight for the it th person-visit contributing to model 1 was calculated as the product of their stabilized inverse probability-of-treatment (IPT) weight, SW A it , and inverse-probability-of-censoring (IPC) weight, SW C it : where t=(1-5); C ik flags whether a participant was censored by visit k (1=yes, 0=no) (Moodie et al., 2008); V i0 represents the vector of measured time-independent baseline covariates; L ik− 1 represents the vector of measured time-varying covariates at visit k-1; and an overbar signifies the observed history of a time-varying variable, e.g., . Weight components were estimated using pooled logistic regression (Appendix 2). Application of the IPTC weights creates a pseudo-population where treatment selection and censoring are statistically independent of measured time-independent and time-varying covariates featured exclusively in the denominator of the weights (Cole and Hernán, 2008;Hernán et al., 2002;Hernán and Robins, 2020;Howe et al., 2016). In other words, the IPTC weights remove the arrows in Fig. 1 from the V 0 and L t− 1 variables into A t− 1 and C t where t=(1-5) (Appendix 2). With a rare outcome and more common exposure, weighting to control for confounding facilitates model convergence and reduces overfitting versus traditional covariate adjustment (Greenland et al., 2016;Richardson et al., 2015;Schuster et al., 2016). While our eligibility criteria were intended to select OAT-eligible participants, we visually inspected estimated weight distributions over time for means far from one or extreme values, which could indicate positivity violations or weight model misspecifications (Cole and Hernán, 2008;Hernán and Robins, 2020). To explore the influence of measured confounding and non-ignorable censoring, we alternatively fit model 1 using IPT weights, IPC weights, and no weights for comparison with the IPTC-weighted estimate.

Sensitivity analysis
We assessed the joint impact of unmeasured confounding and nonignorable censoring by prior outcome response (Y it− 1 ) by adding this variable to the treatment and censoring weight denominators and refitting model 1 with the updated IPTC weights.
We conducted quantitative bias analyses to explore the impact of self-reported treatment misclassification bias on the direction and magnitude of our effect estimate. First, we assumed self-reported exposure (current OAT at visit t-1; A it− 1 ) misclassification was nondifferential by outcome (Y it ) and assigned beta distributions for sensitivity (α=116.6, β=22.2; mean=84%, standard deviation [SD]=3.1%) and specificity (α=245.1, β=36.6; mean=87%, SD=2.0%) based on validation data from a sample of people who inject drugs in Toronto, Ontario (Appendix 3) (Bouck et al., 2022). Using a method described by Fox et al. and adapting open-access code (Appendix 4), we: (1) randomly sampled sensitivity and specificity (bias parameters) from the Fig. 1. Assumed relationships among study measures. Notes: t denotes semi-annual study visit, where t=(− 1,0,1,2,3,4,5) with t=− 1 indicating a participant's lookback visit six months before their baseline visit (t = 0); V 0 denotes time-independent covariates measured at baseline (t = 0); A t denotes exposure (current opioid agonist treatment) at visit t; L t denotes time-varying covariates measured at visit t; Y t denotes outcome measured at visit t; and C t denotes whether a participant was censored as of visit t (the boxes around C t reflect that our analyses were restricted to uncensored person-visit observations, i.e., C t =0). Time-independent baseline covariates (V 0 ) were assumed to affect all other measured variables. The subscript number(s) in parentheses provided beside each variable (or set of variables) denotes the period or point of assessment (in months from baseline) captured by these variables (e.g., Y 1 (0,6] is the outcome measured at t = 1 [first follow-up visit] and it captures whether participants provided injection initiation assistance in the six months between baseline (exclusive) and their first follow-up visit [inclusive]). beta distributions describing our uncertainty in their values; (2) used the selected parameter values to adjust person-visit observations for exposure misclassification; (3) re-fit model 1 (unweighted) in the bias-adjusted dataset; and (4) repeated steps (1)-(3) for 10,000 iterations (Appendix Figure 1) Fox et al. 2021). The median bias-adjusted RR was reported with 95% simulation intervals (SI) incorporating both systematic error (uncertainty in the bias parameter distributions) and random error (Fox et al., 2021). Second, we followed the same steps as the preceding analysis but additionally updated the value of the treatment covariate (current OAT at visit t-2; A it− 2 ) for person-visit observations where t=(1-5) based on the lagged value of a participants' bias-corrected exposure (A it− 1 ) within the same simulated dataset. We reported the median RR with 95% SI after adjusting for joint exposure and covariate misclassification (Appendix 4).

Post hoc analysis
As individuals with severe opioid use disorder characterized by highfrequency opioid injecting may not adequately benefit from first-line OAT for various reasons (e.g., cannot reach therapeutic dose or persistent cravings) (Fairbairn et al., 2019), we investigated whether the short-term effect of first-line OAT on injection initiation assistance provision differs by baseline opioid injecting frequency. Specifically, we added terms to the primary IPTC-weighted outcome model for daily baseline opioid injecting (V * i0 : 1=yes, 0=no)-intended as an indicator of severe, potentially treatment-refractory opioid use disorder-and its product with current OAT at visit t-1 (Cole and Hernán, 2008): All analyses were conducted in SAS V9.4 (SAS Institute Inc.; Cary,

Fig. 2. Flow of participants into study. Notes: VIDUS=Vancouver Injection Drug Users Study; ACCESS=AIDS Care Cohort to Evaluate exposure to Survival Services study;
ARYS=At-Risk Youth Study; *Participants who reported current treatment with second-or third-line opioid agonist medications (e.g., slow-release oral morphine, injectable hydromorphone, or injectable diacetylmorphine) at their look-back visit were also excluded. NC). Fig. 2 shows that 334 VIDUS/ACCESS/ARYS participants were identified as eligible and included in the study cohort. Table 1 summarizes baseline participant characteristics, overall and by current first-line OAT status. At baseline, the median age of participants was 44 years (interquartile range [IQR]=32-52; range=18-70); 61% were cisgender men; 54% were white; and 50%, 30%, and 20% were VIDUS, ACCESS, and ARYS participants, respectively. Within the past six months, 98% of participants reported income from social assistance, 23% experienced homelessness, 16% experienced an overdose, 7% were incarcerated, and 97% injected drugs, of whom 98% injected opioids (64% daily, 34% weekly, and 2% less-than-weekly). Over half (55%) of participants reported recently injecting opioids and non-opioid drugs. One quarter of participants reported previously providing injection initiation assistance, with 25 (7%) recently providing such assistance. Prior (including current) first-line OAT was reported by 86% of participants.

Baseline characteristics
At baseline, a lower proportion of participants on first-line OAT (versus those not on OAT) reported recent provision of injection initiation assistance, homelessness, overdose, daily injecting of opioids or non-opioids, and any public injecting over the past six months. Over 80% of participants on OAT at baseline reported current treatment at their look-back visit versus only 19% of participants not on OAT at baseline.

Follow-up and censoring
Overall, 50 (15%) participants were censored before their first semiannual visit after baseline. The other 284 (85%) participants contributed 1114 follow-up visits (median [IQR]=5 [3-5] visits per participant) ( Table 2). Of the 158 prematurely censored participants, 148 were censored for a missing outcome measurement (97% due to missed visit) and 10 were censored for initiating treatment with an alternative OAT medication. Table 2 summarizes the frequency of recent injection initiation assistance provision among participants by follow-up visit and current first-line OAT status (lagged to prior visit). The proportion of participants reporting current first-line OAT at their preceding visit ranged between 54 and 64% per follow-up visit. Of the 64 possible treatment trajectories between participants' look-back visit and the final exposure assessment (t=4), the most frequently observed treatment history among uncensored participants was reporting current first-line OAT at every visit (41%; 79/192) followed by never reporting current OAT (21%; 41/192).

Opioid agonist treatment and injection initiation assistance provision
The proportion of participants reporting recent injection initiation assistance provision was low, irrespective of follow-up visit (range=3.4-6.9%). In total, there were 50 follow-up visits where recent injection initiation assistance provision was reported across 36 unique participants (median [IQR]=1 [1-2], range=1-4 follow-up visits with outcome per participant). Fig. 3 visualizes estimated weight distributions for the 1114 personvisit observations by 284 participants contributing to our outcome models, though all 334 eligible participants contributed to weight  OAT = Opioid agonist treatment (methadone or buprenorphine/naloxone). a Number of remaining (uncensored) participants. b Helped someone initiate injection drug use in the past six months. c Row percentages (sum may exceed 100% due to rounding). estimation. At each follow-up visit, the stabilized IPT, IPC, and IPTC weights appear well behaved, as mean values are (approximately) equal to 1 and no extreme weights were observed. Table 3 presents our repeated measures log-binomial regression model results estimating the average effect of first-line OAT at visit t-1 on the risk of providing injection initiation assistance in the following six months (i.e., between visits (t-1, t]), conditional on treatment status at visit t-2. Based on the primary IPTC-weighted estimate, participants on first-line OAT at a given visit were, on average, 50% less likely than participants not on OAT to report helping someone initiate injecting in the following six months (RR=0.50, 95% CI=0.23-1.11). Weighting minimally impacted estimation, as the IPTC-weighted and unweighted estimates (RR=0.54, 95% CI=0.25-1.17) are of comparable magnitude and precision. Compared to the IPTC-weighted estimate, the IPTweighted estimate was further from the null (RR=0.47, 95% CI=0.21-1.05) whereas the IPC-weighted estimate was closer to the null (RR=0.56, 95% CI=0.25-1.23). Non-parametric percentile bootstrapping generated 95% CI of similar width to the robust 95% CI (Appendix Table 1).
In a post hoc analysis, current first-line OAT (versus no OAT) was associated with a substantial reduction in the risk of providing injection initiation assistance within the next six months in participants with lessthan-daily baseline opioid injecting (IPTC-weighted n = 468 personvisits [17 with outcome]; RR=0.15, 95% CI=0.05-0.44) but not in participants with daily baseline opioid injecting (IPTC-weighted n = 644 person-visits [36 with outcome]; RR=0.86, 95% CI=0.35-2.11).

Table 3
Estimates of the average effect of current first-line opioid agonist treatment (OAT) at visit t-1 on risk of providing injection initiation assistance over the next six months (i.e., between visits (t-1, t]) based on 1114 person-visits (before weighting) by 284 a people who inject drugs with suspected opioid use disorder from Vancouver, Canada-December 2014 to May 2018.

Effect of current OAT at visit t-1
Model specifications a n b (# with outcome b ) Relative Risk 95% CI c IPC = inverse-probability-of-censoring; IPT = inverse-probability-of-treatment; IPTC = inverse-probability-of-treatment-and-censoring; CI = confidence interval. a All estimates obtained from log-binomial generalized estimating equations regression models with recent injection initiation assistance provision (outcome) regressed on fixed terms for follow-up visit (t= [1,2,3,4,5]), current OAT at visit t-1 (exposure), and current OAT at visit t-2 (covariate), with an unstructured working correlation structure for within-participant responses. All models based on 1114 person-visit observations (before weighting) by 284 unique participants. All other eligible participants (50/334) were censored by t = 1 (see Table 2) and only contributed to weight estimation. b For weighted models, n and # with outcome rounded to nearest integer after application of weights. c Based on robust variance estimator with clustering by subject.

Discussion
In this longitudinal cohort study, we evaluated whether first-line OAT (liquid methadone or sublingual buprenorphine/naloxone) in people who inject drugs with suspected opioid use disorder--characterized by frequent non-medical opioid use at baseline-reduces their short-term likelihood of helping others initiate injection drug use. According to our primary inverse-probabilityweighted estimate, participants currently on first-line OAT were 50% less likely, on average, to help someone initiate injecting in the following six months versus participants not on medication for opioid use disorder; however, the corresponding 95% CI was imprecise, exhibiting compatibility with both a stronger beneficial effect (77% relative risk reduction) and weak, likely negligible detrimental effect (11% relative risk increase). Notably, the magnitude and precision of the estimated treatment effect were relatively consistent across unweighted and alternatively-weighted secondary analyses exploring violations of the no weight model misspecifications, no unmeasured confounding, and ignorable censoring assumptions underlying our inferences. Additionally, adjustments for self-reported outcome and treatment misclassification yielded slightly more protective effect estimates.
Our study is the first to use incidence data to demonstrate support for an effect of first-line OAT on subsequent injection initiation assistance provision in people who inject drugs with frequent non-medical opioid use. Despite intentional differences in design and analysis, the direction and magnitude of our primary effect estimate fits with prior crosssectional associations involving the same Vancouver-based cohorts and outcome (recent injection initiation assistance provision) (Marks et al., 2021b;Mittal et al., 2019b). Using VIDUS/ACCESS/ARYS questionnaire data from 2014 to 2017, both Mittal et al. (1740 participants;4.6% with outcome) and Marks et al. (1825 participants;4.8% with outcome) found that people who inject drugs enrolled in first-line OAT within the past six months were less likely to report helping someone initiate injecting over the same period (odds ratio=0.52, 95% CI=0.31-0.87 and prevalence ratio=0.65, 95% CI=0.43-1.00, respectively), after adjusting for a similar set of potential confounders including age, gender, and recent injection drug use (Marks et al., 2021b;Mittal et al., 2019b). This relative consistency in point estimates suggests that the sources of potential bias we identified in previous studies-specifically, including treatment-ineligible participants and contemporaneously measured variables (Marks et al., 2021b;Mittal et al., 2019b)-and addressed in our primary analysis to improve validity may have minimally biased estimation. Collectively, this mounting evidence indicates support for increasing OAT coverage in people who inject drugs as a strategy to both reduce injection and associated harms among treated individuals and prevent incident injection events in the broader population of people who use drugs (Marks et al., 2021b(Marks et al., , 2019Meyers-Pantele et al., 2021;Mittal et al., 2019b).
While the direction and magnitude of our primary estimate offer support for a protective treatment effect, the imprecision of the corresponding 95% CI represents a key limitation that could be due to several factors. First and foremost, precision was clearly limited by the low number of outcomes contributing to the analysis as a result of our small sample size and suspected underreporting of injection initiation assistance provision because of stigma and poor recall (Guise et al., 2018;Mittal et al., 2019a). Precision could have been compromised further by self-reported treatment (exposure) misclassification, which can inflate standard errors and widen confidence intervals for treatment effect estimates (van Smeden et al., 2020).
Another explanation for the uncertainty in our primary result is treatment effect heterogeneity across subgroups constituting the overall study population (Corraini et al., 2017). Supporting this notion, our post hoc analysis results imply that current first-line OAT might substantially reduce the short-term risk of injection initiation assistance provision in participants with less-than-daily baseline opioid injecting but not in participants with daily baseline opioid injecting. While promising, the magnitude of the estimated treatment effect in the lower frequency opioid injecting subgroup is surely exaggerated due to an inadequate number of injection initiation assistance provision reports within OAT and covariate strata (sparse data bias) (Greenland et al., 2016). Conversely, the approximately null treatment effect in participants with daily opioid injecting could be owed to this subgroup being largely comprised of individuals with more severe opioid use disorder for whom methadone or buprenorphine/naloxone may not adequately reduce their opioid cravings, withdrawal symptoms, and non-medical opioid injecting (Fairbairn et al., 2019). Therefore, some participants in this subgroup may have continued injecting opioids frequently despite current or past first-line OAT (86% ever enrolled at baseline), thereby remaining at increased risk of providing injection initiation assistance versus the overall study population Mittal et al., 2019b;Navarro et al., 2019;White et al., 2020). Such participants may have instead benefited from emerging alternative OAT medications (e. g., slow-release oral morphine or injectable hydromorphone) versus initiating, restarting, or continuing first-line OAT (Centre for Addiction and Mental Health (CAMH), 2021; Fairbairn et al., 2019). Although these alternative OAT medications were not a focus of this study as their use was heavily restricted during follow-up, they have become increasingly prescribed in Canada in recent years to expand treatment coverage amid an ongoing fentanyl-driven opioid crisis (Ivsins et al., 2020;Young et al., 2022). Future investigations might evaluate the effectiveness of alternative OAT medications on reducing non-medical opioid injecting and injection initiation assistance provision in people who inject opioids daily, including those with prior history of first-line OAT, given their elevated risk of exposing non-injectors to injection practices and helping them initiate injecting (Fairbairn et al., 2019).
There are other study limitations worth noting. The generalizability of our results to the intended target population (Vancouver-based people who inject drugs with opioid use disorder) is questionable, as the study cohort represents a convenience sample in which the prevalence of current first-line OAT (range=54-64%) far exceeds prior estimates of current OAT enrollment in North American people who inject drugs with opioid use disorder (range=11-34%) . Relatedly, Table 4 Results of probabilistic bias analyses adjusting for nondifferential misclassification of current opioid agonist treatment-measured at both visit t-1 (as exposure) and visit t-2 (as covariate)-and differential misclassification of recent injection initiation assistance provision (outcome).  Table 3) with recent injection initiation assistance provision (outcome) regressed on terms for time (t; follow-up visit), current opioid agonist treatment at visit t-1 (exposure), and current opioid agonist treatment at visit t-2 (covariate). b Upper limit of interval (97.5th percentile value) divided by lower limit of interval (2.5th percentile value). c For conventional result, 95% Interval is 95% CI from Table 3. d For the bias-adjusted results, the reported RR is the median value across all converging iterations and the 95% Interval is Simulation Interval that incorporates systematic error (and random error where explicitly noted). e Updated covariate value for a given person-visit observation where t= (1,2,3,4,5) based on lagged, bias-adjusted value of exposure for same participant at prior visit.
we could not adjust for sources of bias affecting selection of participants into the study through weighting, which would require information on the sampling frame that is unavailable. Lastly, our study includes observations both before and during (2016-present) the fentanyl era of the opioid crisis in British Columbia. Knowledge of increasing adulteration of the unregulated drug supply with high-potency fentanyl and its analogues may have reinforced some participants' attitudes against providing injection initiation assistance, resulting in reduced likelihood of engaging in this behavior over time independent of treatment status (Olding et al., 2019).

Conclusions
Our findings suggest that first-line OAT may meaningfully reduce the short-term likelihood that people who inject drugs with frequent nonmedical opioid use help others initiate injecting, particularly among individuals with less-than-daily opioid injecting. However, the extent of this presumed effect remains uncertain due to imprecise estimation and observed heterogeneity by baseline opioid injection frequency. Future investigations on this topic should prioritize recruiting a larger cohort, consider effect modification by baseline opioid injection frequency (and previous OAT history), and evaluate emerging alternative OAT medications, including in people who inject opioids daily given their increased risk of providing injection initiation assistance.

Authors' contributions
ZB, ACT, LR, and DW formulated the research question and hypothesis. ZB conducted the statistical analyses and wrote the initial draft. All authors contributed to the design, interpretation of the results, and manuscript revisions.

Role of funding source
None.

Ethics approval and consent to participate
This study was approved by the University of California San Diego Human Research Protection Program, the University of British Columbia/Providence Health Care and University of Toronto Research Ethics Boards (#27433).

Consent for publication
Not applicable.

Availability of data and materials
Data sharing and privacy agreements for VIDUS/ACCESS/ARYS prohibit the investigators from making the data set publicly available. Partial code has been provided in the Supplemental Materials.

Declaration of Competing Interest
The authors declare that they have no conflicts of interest.