Trajectories of retention in opioid agonist therapy and overdose risk during a communitywide overdose epidemic in a Canadian setting

Abstract

Introduction:

Retention in opioid agonist therapy (OAT) has consistently been linked with improved outcomes among people with opioid use disorder (PWOUD). However, less is known about the links between patterns of engagement in OAT over the long term and overdose risk. We assessed the association of OAT retention trajectories with non-fatal overdose.

Methods:

Data were drawn from two community-recruited prospective cohorts of people who use drugs in Vancouver, Canada. Latent class growth analysis was used to identify trajectories of OAT retention among PWOUD initiating OAT, and generalized estimating equations to assess the association of these trajectories with non-fatal overdose events after OAT initiation.

Results:

Between 2005 and 2018 among 438 OAT initiators, we identified four retention trajectories: “consistently high” (35.6%), “increasing” (26.0%), “consistently low” (23.3%), and “decreasing” (15.1%) OAT engagement. Over the study period, there were 371 non-fatal overdose events, with 179 (40.1%) participants reporting at least one. In adjusted analysis, the “consistently low” (Adjusted Odds Ratio [AOR] = 1.73, 95% Confidence Interval [CI]: 1.10 – 2.71) and “decreasing” (AOR = 1.87, 95% CI: 1.18 – 2.95) retention trajectories were positively associated with increased odds of non-fatal overdose compared to the “consistently high” OAT retention class.

Conclusions:

We found that sub-optimal trajectories of OAT retention were associated with increased likelihood of non-fatal overdose. These findings suggest that reducing barriers to sustained engagement in OAT will be critical to address North America’s overdose epidemic.

INTRODUCTION

According to recent estimates, there were more than 26 million persons living with an opioid use disorder (PWOUD) in 2016, a 47% increase in prevalence compared to 1990.¹ North America has been particularly affected by harms arising from opioid use disorder (OUD), with approximately 15% of the global population of PWOUD, and a quarter of the global Disability-Adjusted Life Years and deaths attributable to OUD being concentrated in this region.¹ Since 2016, the opioid-related overdose crisis in the U.S. and Canada has continued to worsen, largely driven by an increasing contamination of the unregulated drug supply with illicitly-manufactured fentanyl and related analogues.² In 2018, there were more than 48,000 opioid-related overdose deaths (14.9 per 100,000 people) in the U.S.—a 12% increase from 2016.² Overdose death rates have also been consistently rising across Canada, with some provinces like British Columbia being particularly impacted, as indicated by an overdose mortality rate of 30.6 per 100,000 people in 2018.³

Research from different settings has demonstrated that individuals with at least one non-fatal overdose are three to eight times more likely to experience a subsequent non-fatal or fatal overdose than those with no overdose history.^4-6 Additionally, a dose-dependent relationship appears to exist whereby the risk of death increases with the number of overdose events.^{4, 5} Other studies indicate that around 10% of people who survived an overdose will die within a year, with over a third of these deaths being the result of another opioid overdose, further underscoring the close link between non-fatal overdose and risk of death.⁷

Although OUD is best described as a chronic relapsing condition with no effective cure, periods of retention in opioid agonist treatment (OAT) have been consistently associated with improvements in health outcomes, including reductions in all-cause and overdose-specific mortality,^{8, 9} highlighting the need to expand access to and retention in OAT as a key response to the overdose crisis. While many individuals cycle in and out of treatment, retention has typically been measured as the duration of a single episode,¹⁰ and, thus, the links between different OAT engagement patterns and overdose risk are not well understood. The identification of heterogenous subgroups of PWOUD, as it pertains to their longitudinal engagement in OAT and subsequent health outcomes, is critical to target interventions to those most at risk. Therefore, the aim of this study was to identify distinct longitudinal patterns of retention in OAT, and to assess their association with overdose risk among people initiating OAT in Vancouver, Canada, a setting with low-barrier access to OAT.

METHODS

Study design and participants

Data for this study were derived from two harmonized prospective and ongoing community-recruited cohorts of people who use drugs in Vancouver, Canada: the Vancouver Injection Drug Users Study (VIDUS) and the AIDS Care Cohort to evaluate Exposure to Survival Services (ACCESS). Detailed eligibility criteria and study procedures for each cohort have been published before.^{11, 12} Briefly, participants are recruited through self-referral, word-of-mouth, and street outreach in the Greater Vancouver Regional District. VIDUS includes HIV-negative adults (i.e., ≥ 18 years) who injected drugs in the month preceding enrollment, and ACCESS includes HIV-positive adults who used illicit drugs (other or in addition to cannabis, which was illegal for most of the study period) in the month prior.

After providing informed consent, at baseline and at each six-month follow-up visit, participants complete an interviewer-administered questionnaire, as well as provide blood samples for HIV and HCV serological testing and disease monitoring, as appropriate. The questionnaire collects information on socio-demographics, substance use patterns, access to and engagement with relevant health services, as well as other pertinent social-structural exposures. At each study visit, participants are compensated $40 Canadian Dollars. The VIDUS and ACCESS studies have been approved by the University of British Columbia/Providence Health Care Research Ethics Board.

The analytical sample for the present analysis was restricted to participants who initiated OAT after enrollment into VIDUS or ACCESS, and completed at least three study visits between December 2005 and November 2018, in order to have at least one year of follow-up data for each participant, and be able to model non-linear OAT retention trajectories.

Measures

The outcome of interest was non-fatal overdose, defined as a self-report of having experienced an overdose in the six months prior to the interview. We chose non-fatal overdose given the inconsistent availability of mortality data during the study period, and the well-established association between non-fatal and subsequent fatal overdose.^4-6

Our exposure of interest was membership in distinct trajectories of retention in OAT. These trajectories were identified through latent growth class analysis (LCGA), using a self-reported measure of being enrolled in OAT. Methadone was available during the whole study period, buprenorphine/naloxone became available in 2010, and slow-release oral morphine and injectable diacetylmorphine and hydromorphone in 2017 under an off-label indication. Retention in OAT for a given visit was defined as being on OAT at the time of the current study interview and the immediately preceding interview, a time interval of approximately six-months.¹³ The date of the first interview in which a participant reported being enrolled in an OAT program was defined as the baseline for these analyses. The probability of OAT retention over time was used to identify distinctive subpopulations with homogeneous longitudinal trajectories. To determine the optimal number of distinct OAT retention groups, we started with a single-class latent growth curve model and continued until a five-class model was fitted. Linear, quadratic, and cubic parameters were tested for each trajectory group. The model that best fit the data was selected based on the following criteria: the Bayesian and Akaike information criteria, the average posterior probabilities of groups, examination of the shapes of the trajectories for similarities, and group membership probabilities.^{14, 15} The estimates of parameters were adjusted for attrition using a full-information maximum -likelihood estimation under the assumption that the data was missing at random, which allows to include individuals with missing data. Model fit statistics for one- to five-class solutions are presented in Table S1 of the appendix.

We selected a set of covariates based on previous evidence that we hypothesized could be associated with both retention in OAT and non-fatal overdose.⁹ Socio-demographic characteristics considered included age at OAT initiation, sex (male vs. female), self-reported ancestry (white versus others), and HIV serostatus. Substance use-related covariates included high intensity (i.e., dichotomized at ≥ daily vs. < daily) use of illicit opioids (e.g., heroin, fentanyl, methadone, opioid analgesics used in an illicit fashion, i.e., from unregulated sources or not as medically prescribed), stimulants (i.e., cocaine, crack, methamphetamine), or cannabis, as well as high-risk drinking, as per the U.S. National Institute on Alcohol Abuse and Alcoholism’s definition (i.e., >4 drinks/day or >14/week for men, and >3 drinks/day or 7/week for women).¹⁶ To assess substance use patterns, at each visit, participants are presented with a list of substances and asked to indicate which substances they used, the mode of administration (injection versus not), and the frequency of use in the past six months (no use, less than once a month, 1–3 times a month, about once a week, 2–3 times a week, ≥daily). We also included a variable indicating cumulative history of overdose, which was lagged to avoid overfitting of the model. Other variables considered included year of OAT initiation, and a number of prevalent social-structural exposures, such as homelessness (defined as responding yes to the question “have you been homeless in the last six months”), incarceration (defined as self-reporting being held in secure custody [i.e., a municipal detention centre or jail, a provincial prison or a federal penitentiary] at least overnight in the last six months), and employment (including regular or temporary work and self-employment). Except for the socio-demographic characteristics and year of OAT initiation, all variables were time-updated, refer to the six-month period prior to the interview, and are consistent with previous work. The prevalence of missing data for covariates was <0.25%.

Statistical analyses

As a first step, we summarized characteristics of the analytic sample, stratified by having ever experienced an overdose during the study period. Then, we examined bivariable associations between OAT retention trajectories and other covariates with the outcome of interest, using generalized estimating equations (GEE) with a logit-link function to account for repeated measurements over time from the same participants. To determine the independent effect of OAT retention trajectories on overdose, we built a multivariable GEE model using an a priori approach to identify and account for confounding covariates,¹⁷ that we have used previously.^{13, 18} Starting with a full model containing the primary explanatory variable, as well as other covariates associated with the outcome in bivariable analysis at p<0.10, we constructed reduced models removing the variable that resulted in the smallest relative change in the OAT trajectories coefficient, in a stepwise manner. We continued this iterative process until the minimum change of the OAT trajectories coefficient exceeded 5%. Remaining co-variates were considered confounders. A complete case approach was used, where observations with missing values for covariates were not included in the analyses. Statistical analyses were performed using SAS version 9.4 (SAS, Cary, NC, USA), and the LCGA was conducted using the Proc Traj procedure.¹⁹ All p-values were two-sided.

RESULTS

Between December 2005 and November 2018, 573 participants initiated OAT after enrollment in the study cohorts. Of these, 438 (76.4%) completed at least three study visits and were included in the analytic sample. Characteristics of included and excluded participants are presented in Table S2 of the Appendix. Excluded participants were more likely to have initiated OAT in more recent years and have a history of non-fatal overdose, but less likely to be using illicit opioids at least daily. No other significant differences were found between included and excluded participants.

Characteristics of the study sample at the time of OAT initiation are presented in Table 1. The median age was 42 years (IQR 35–48), 269 (61.4%) were male, and 196 (44.8%) self-identified as white ancestry. The majority (411, 93.8%) started methadone-based OAT. Co-use of substances other than illicit opioids was common, ranging from 34 participants (7.8%) reporting at-risk drinking to 183 (41.8%) reporting at least daily stimulant use. Participants were followed for a median of 91 months (IQR 50–119) after treatment initiation and contributed a total of 3119 person-years of observation.

Table 1.

Baseline characteristics of OAT initiators, stratified by OAT engagement trajectory, Vancouver, British Columbia, 2005-2018.

	Total (N = 438)	OAT engagement trajectory groups, n (%)				p- value
		Consistently low (n = 102)	Increasing (n = 114)	Decreasing (n = 66)	Consistently high (n = 156)
Socio-demographics
Age (median, IQR)	42 (35–48)	42 (33–49)	42 (35–49)	43 (36–49)	42 (36–48)	0.515b
Male sex	269 (61.4)	64 (62.8)	72 (63.2)	42 (63.6)	91 (58.3)	0.805
White race	196 (44.8)	40 (39.2)	49 (43.0)	33 (50.0)	74 (47.4)	0.458
Comorbidities
HIV-positive	151 (34.5)	31 (30.4)	37 (32.5)	25 (37.9)	58 (37.2)	0.615
Substance use-related factorsa
≥ Daily illicit opioid use	426 (97.3)	100 (98.0)	109 (95.6)	64 (97.0)	153 (98.1)	0.626c
≥ Daily stimulant use	183 (41.8)	41 (40.2)	53 (46.5)	24 (36.4)	65 (41.7)	0.596
≥ Daily cannabis use	80 (18.3)	16 (15.7)	23 (20.2)	10 (15.2)	31 (19.9)	0.698
High-risk drinking	34 (7.8)	8 (7.8)	11 (9.7)	6 (9.1)	9 (5.8)	0.661
History of non-fatal overdose	245 (55.9)	60 (58.8)	70 (61.4)	38 (57.6)	77 (49.4)	0.209
Calendar-year of OAT initiation	2009 (2007–2012)	2009 (2007–2012)	2009 (2007–2012)	2009 (2007–2012)	2009 (2007–2012)	0.238†
Social-structural factorsa
Employment	83 (18.9)	23 (22.6)	24 (21.1)	17 (25.8)	19 (12.2)	0.049
Homelessness	160 (36.5)	31 (30.4)	47 (41.2)	24 (36.4)	58 (37.2)	0.428
Incarceration	58 (13.2)	17 (16.7)	10 (8.8)	7 (10.6)	24 (15.4)	0.257
Months of follow-up (median, IQR)	91 (50–119)	66 (41–111)	98 (60–119)	93 (57–117)	97 (53–124)	0.015b

Note: Boldface indicates statistical significance (p<0.05).

IQR, interquartile range; OAT, opioid agonist therapy

^aRefers to the six-month period prior to the baseline interview

^bKruskal-Wallis test

^cFisher’s exact test

Overall, the proportion of participants retained in OAT at any given visit ranged between 47.3% and 60.7%. A four-class model was determined to provide the optimal number of distinct OAT retention trajectories: “consistently high” (156, 35.6%), “consistently low” (102, 23.3%), “increasing” (114, 26.0%), and “decreasing” (66, 15.1%). These have been described in detail before.²⁰ In brief, the “consistently high” class showed approximately 90% probability of OAT retention at each visit; the “consistently low” class, less than 20% retention probability; the “increasing” class, an initial decreasing pattern during the first year of OAT (from <50% to 20% probability with a subsequent increase to >80% probability at year 5); and the “decreasing” class a steady decreasing engagement pattern (from around 85% probability of OAT retention to <20% probability at year 5). As indicated in Table 1, there were no significant differences among members of each trajectory at the time of OAT initiation, with the exception that those in the “consistently high” class were more likely to be unemployed. Participant follow-up differed by trajectory, with those in the “consistently low” group being more likely to have shorter times.

At the time of OAT initiation, over half of participants had a history of previous non-fatal overdose (245, 55.9%), of whom 32 (13.1%) had experienced a recent overdose. Over the study period, there were 371 non-fatal overdose events, with 179 (40.1%) participants reporting at least one non-fatal overdose (median 2, IQR 2–3). More specifically, 42 (41.2%) participants in the “consistently low” OAT engagement class experienced 91 overdose events; 54 participants (47.4%) in the “increasing” class, 119; 33 participants (50%) in the “decreasing” class, 75; and 50 (32.1%) participants in the “consistently high” class, 86.

As indicated in Table 2, in unadjusted analysis, participants with “consistently low”, “increasing” or “decreasing” OAT retention patterns had increased odds of experiencing a non-fatal overdose compared to members of the “consistently high” class. After adjusting for a range of socio-demographics, drug use patterns and social-structural exposures, compared to the “consistently high” OAT retention class, the “consistently low” (AOR=1.73, 95% CI: 1.10–2.71) and “decreasing” (AOR=1.87, 95% CI: 1.18–2.95) retention trajectories—but not the “increasing” group (AOR=1.48, 95% CI: 0.96–2.27)— remained positively associated with increased odds of non-fatal overdose.

Table 2.

Unadjusted and adjusted longitudinal generalized estimating equations analyses of OAT retention trajectories and non-fatal overdose risk among 438 OAT initiators (Vancouver, Canada, 2005-2018).

	Non-fatal overdosea
	Unadjusted OR (95% CI)	p - value	Adjusted OR (95% CI)d	p - value
OAT retention trajectories (ref: Consistently high)
Consistently low	2.38 (1.45 – 3.91)	<0.001c	1.73 (1.10 – 2.71)	0.017
Increasing	1.86 (1.19 – 2.90)	0.006 c	1.48 (0.96 – 2.27)	0.073
Decreasing	1.91 (1.19 – 3.09)	0.008 c	1.87 (1.18 – 2.95)	0.008
Covariates
Age (per year older)	1.02 (1.00 – 1.03)	0.102
Male gender	1.11 (0.78 – 1.58)	0.549
Caucasian ethnicity	0.99 (0.70 – 1.38)	0.940
HIV positive	0.91 (0.64 – 1.30)	0.609
≥ Daily illicit opioid usea	0.92 (0.66 – 1.27)	0.601
≥ Daily stimulant usea	1.05 (0.79 – 1.41)	0.721
≥ Daily cannabis usea	1.03 (0.76 – 1.39)	0.839
High-risk drinkinga	1.95 (1.44 – 2.65)	<0.001c
Cumulative history of non-fatal overdoseb	1.21 (1.15 – 1.28)	<0.001c	2.57 (1.92 – 3.44)	<0.001
Employmenta	0.99 (0.76 – 1.29)	0.913
Homelessnessa	2.39 (1.81 – 3.14)	<0.001c	2.57 (1.92 – 3.44)	<0.001
Incarcerationa	1.54 (1.04 – 2.28)	0.032c
Calendar year of OAT initiation	1.21 (1.15 – 1.28)	<0.001c	1.20 (1.14 – 1.26)	<0.001

Note: Boldface indicates statistical significance (p<0.05).

OR, odds ratio; OAT, opioid agonist therapy.

Level of heterogeneity between cohorts: p=0.0268

^aRefers to the 6-month period prior to the interview

^btime-updated every 6 months

^cp<0.10 in the unadjusted analyses and considered for inclusion in the multivariable model

^dOnly the variables included in the final multivariable confounder model are presented in this column

As shown in Table S3 in the appendix, a sensitivity analysis including the 135 participants with less than three study visits, showed similar results, with those in the “consistently low” and “decreasing” OAT retention groups having increased odds of experiencing a non-fatal overdose when compared to the “consistently high” group (AOR=1.74, 95% CI: 1.11–2.72, and AOR=1.86, 95% CI: 1.19–2.90, respectively). There were no significant differences in the risk of non-fatal overdose between participants with less than three visits and the “consistently high” OAT retention group.

DISCUSSION

This study identified four distinct patterns of longitudinal engagement in OAT (i.e., “consistently high”, “consistently low”, “increasing”, and “decreasing”) among more than 400 PWOUD initiating OAT in Vancouver, Canada, between 2005 and 2018. After adjusting for a number of socio-demographic and substance use-related variables, membership in the “consistently low” and “decreasing” retention classes remained associated with an almost doubled increased likelihood of experiencing a non-fatal overdose after OAT initiation. To our knowledge, this is the first study to investigate the differential impacts of distinct long-term OAT retention trajectories on non-fatal overdose risk.

Overall, research shows that retention in OAT is consistently associated with reduced risk of non-fatal and fatal overdose. For example, a recent meta-analysis reported that people with untreated OUD have approximately three-times increased risk of dying than those engaged in OAT, with the period immediately after OAT discontinuation being of particular high risk.⁸ Likewise, in our setting, the prevalence of individuals exposed to unknown substances (e.g., fentanyl-laced substances) and not engaged in OAT at the time of non-fatal opioid overdoses has sharply increased since 2015.²¹ Our findings build on these previous studies, as indicated by higher likelihood of overdose among participants with sub-optimal longitudinal OAT engagement patterns. Specifically, our study suggests that not only PWOUD who have a sustained poor retention in OAT are at higher risk of overdose (i.e., “consistently low” class), but also that even those who may have demonstrated good attendance in the first year of OAT initiation may be at risk of overdose if engagement in OAT is not sustained over time (i.e., “decreasing” class).

Collectively, our findings have implications for OAT programs. Specifically, our results suggest that sustained engagement in OAT is critical to decrease overdose risk among PWOUD. Unfortunately, and consistent with our results, typically less than half of PWOUD who initiate OAT are typically retained in care for more than six months.²² Common challenges contributing to low OAT retention rates include strict programmatic requirements, as well as prevalent socioeconomic marginalization, including high rates of homelessness (>30% in our sample). For example, the need for frequent (e.g., daily) visits to the pharmacy for witnessed ingestion of some of the OAT medications (e.g., methadone),²³ could be a barrier for employed individuals that may in turn impact adherence or retention in care, as previously reported.^{20, 24} Similarly, competing priorities among homeless individuals, including securing basic needs such as shelter, food and safety, may partially explain their low use of addiction services.^{25, 26} Accordingly, there is an urgent need to address individual, health system and structural barriers that contribute to these high attrition rates. From a clinical perspective, focusing on longitudinal patterns of engagement in OAT, rather than single treatment episodes may help to identify individuals at higher risk of overdose and provide adequate support to facilitate quick re-engagement in care of PWOUD with missed clinical visits before they are lost-to-follow-up.

Limitations

Some limitations should be taken into account when interpreting findings from this study. First, our sample was not randomly selected and only included participants with at least three study visits, which may limit the generalizability of these findings. Likewise, the majority of our study population initiated methadone-based OAT, and thus findings for other OAT should be taken with caution. Second, we relied on self-reported data which may be prone to recall and social desirability bias. However, previous studies have demonstrated PWUD reports to be valid and reliable,^{27, 28} and there is no reason to believe there would be differential reporting between participants in different OAT trajectories with respect to reporting or not an overdose. That said, given the varying and subjective nature of overdose symptoms, we may have inadvertently captured non-fatal overdose from other substances. Third, to construct the measure of OAT retention we used two time points: the participant’s self-report of being on OAT at the time of the current and the immediately preceding interview. Thus, we cannot ensure that the participant was enrolled in OAT for the entire interval between the two visits. Fourth, given limitations in mortality data during the study period, we did not investigate the relationship between OAT retention trajectories and fatal overdose. Although non-fatal overdose is a risk factor for fatal overdose, future research should consider the impacts of longitudinal engagement patterns of OAT retention on mortality risk. Fifth, given the observational nature of this study and the statistical approach used to define the OAT retention trajectories (i.e., need to use the five years of data for each participant), we cannot neither exclude the possibility that the positive association between two of the sub-optimal OAT retention trajectories and non-fatal overdose is the result of unmeasured confounding nor that some of the overdose events may have preceded some of the study visits used to define OAT engagement patterns.

CONCLUSIONS

In conclusion, our study found that sub-optimal trajectories of OAT retention were associated with an increased likelihood of non-fatal overdose among PWOUD initiating OAT in Vancouver, Canada. These findings highlight the importance of identifying barriers to OAT retention, and further suggest that providing additional and continued support to PWOUD at risk of treatment drop out will be key to reduce their risk of non-fatal overdose and subsequent risk of death, and maximize the potential impact of evidence-based treatment on the ongoing overdose crisis.