Abstract
Aims:
Injection drug use patterns are known to change and evolve over time, although such long-term changes have not been well described. We sought to characterize longitudinal trajectories of injection drug use and identify associated factors.
Design:
Data were derived from the Vancouver Injection Drug Users Study and AIDS Care Cohort to evaluate Exposure to Survival Services study, two prospective cohorts involving people who inject drugs in Vancouver, Canada, between 1996 and 2017. Growth mixture modeling was applied to identify distinct injection drug use trajectories. Multinomial logistic regression was used to identify baseline factors associated with each trajectory.
Setting:
Canada.
Participants:
A total of 2057 participants who reported having used illicit drugs via injection in the past six months at the baseline visit were included in the study. The median time since first injection drug use at baseline was 14.8 years (quartile 1 – quartile 3: 6.5 – 24.3).
Measurements:
Information regarding self-reported injection drug use during the past six months was collected at baseline and semi-annually thereafter via interviewer-administered questionnaires.
Findings:
Participants were followed for a median of 113.4 months (quartile 1 – quartile 3: 63.4 – 161.7). Five trajectories were identified: persistent high frequency injection (507, 24.6%); high frequency injection with late decrease (374, 18.2%); gradual cessation (662, 32.2%); early cessation with late relapse (227, 11.0%); and early cessation (287, 14.0%). Factors found to be associated with distinct trajectories included: daily heroin injection, binge injection drug use, age, not being in a stable relationship, and year of study enrollment.
Conclusions:
Five drug use trajectories were identified, with a significant portion of participants remaining high frequency injectors over the study period. Findings from this study may help provide scientific evidence to guide the development of strategies to reduce long-term injection drug use.
Keywords: Injection drug use, cessation, relapse, growth mixture modeling, latent trajectory groups, longitudinal studies
INTRODUCTION
The impact of injection drug use on the health of individuals and related social harms continues to be significant. Globally, there are around 13 million people who inject drugs, of whom approximately 13% are living with HIV and 67% are seropositive for hepatitis C [1]. People who inject drugs also suffer from high rates of preventable morbidity and mortality. A recent systematic review including 67 cohorts involving people who inject drugs demonstrated a pooled crude mortality rate of 2.4 deaths per 100 person-years, which constitutes approximately 14.7 times the risk of death among the general population [2]. Untreated substance dependence also inflicts substantive economic and social harm on communities, including unemployment, lost productivity, criminal activity, and excessive healthcare costs [3]. Therefore, reducing the prevalence of injection drug use remains an important public health challenge, which in turn requires a better understanding of the natural history of injection drug use.
Research demonstrates that over the course of a drug injection career, drug use patterns are rarely stable [4-9]. Multiple studies on drug use trajectories have focused on short-term (e.g., six months) injection drug use cessation and relapse, and suggested that factors including intensity of drug injection, frequent non-injection drug use, polydrug use, alcohol drinking, age at injection initiation, homelessness, incarceration, illegal income generating activities, and treatment involvement play an important role in influencing injection behaviour change [5, 8, 10-12].
However, substance dependence is a chronic relapsing condition, as a significant portion of people will experience multiple episodes of cessation of and relapse into active substance use over their drug-using career [13-15]. Few studies have provided insight into the factors associated with longitudinal patterns of injection drug use. Factors indicative of drug of choice and substance use severity (e.g., mixed heroin and cocaine injection, daily or more frequent injection, cumulative past use of stimulants), social stability (e.g., criminality, employment), treatment involvement (e.g., number of past treatment episodes), and psychological distress have been found to be predictors of change in long-term substance use patterns [14, 16-19]. Interestingly, factors associated with short-term changes in injection behaviour do not consistently predict long-term injection trajectories. For instance, numerous studies have documented that among those who cease use as a result of formal addiction treatment interventions, high rates of relapse are typical after treatment discharge [20-23]. Furthermore, for factors that are associated with both short-term and long-term changes in injection behaviour, the strength of the associations often varies. Specifically, a meta-analysis examined and summarized different types of predictors of continued drug use during and after treatment for opioid use disorder, and showed that for various factors (e.g., drug use history, criminal behaviour), the strength of concurrent and longitudinal associations differs considerably [24].
To date, only a small number of studies have examined longitudinal trajectories of injection drug use behaviour over periods of ten years or greater. Two studies using a sample of people who use drugs in the AIDS Linked to the Intravenous Experience (ALIVE) cohort in Baltimore, United States (US) [13, 19] consistently found an overall decreasing trend of injection drug use, with a substantial proportion of people stopping injection for extended time frames. Another recently published study assessed patterns of changes in injecting frequency in a European population and found similar longitudinal patterns [25], but the overall decreasing trend was not as strong as in the ALIVE cohort. In the face of escalating harmful illicit drug use, harm-reduction interventions, including needle and syringe programs [26], supervised consumption facilities [27], and methadone maintenance therapy [28, 29], have been implemented and scaled up in British Columbia, Canada. However, to our knowledge, there exist no studies that have examined the longitudinal trajectories of injection drug use in a Canadian setting.
Given that the long-term trajectories of injection drug use remain under-studied, there are limited empirical examinations of the natural history of injection drug use. There is, however, considerable practical and scientific importance associated with gaining greater understanding of longitudinal trajectories of injection drug use, which in turn will help provide scientific evidence to guide the development of strategies to reduce long-term injection drug use. Accordingly, this study aims to characterize long-term injection trajectories and identify factors associated with different injection drug use trajectories among longstanding cohorts of people who inject drugs in Vancouver, Canada.
METHODS
Design and participants
Data were derived from two ongoing open prospective cohort studies of people who use drugs in Vancouver: the Vancouver Injection Drug Users Study (VIDUS), and AIDS Care Cohort to evaluate Exposure to Survival Services (ACCESS). Detailed descriptions of these cohorts have been published elsewhere [30-32]. In brief, eligibility criteria for VIDUS includes being HIV-seronegative, at least 18 years of age at enrollment, residing in the Greater Vancouver area, and reporting injecting an illicit drug in the month prior to enrollment. Eligibility criteria for ACCESS includes being HIV-seropositive, at least 18 years of age, residing in the Greater Vancouver area, and reporting using an illicit drug (other than or in addition to cannabis) in the month prior to enrollment. Recruitment for both cohorts relies on extensive snowball sampling, self-referral, and street outreach. Individuals in VIDUS who seroconvert following recruitment are transferred to ACCESS for ongoing follow-up. The baseline and follow-up procedures for these studies, including the questionnaires, are harmonized to allow for combined analyses of the different cohorts.
At baseline and semi-annually thereafter, participants complete an interviewer-administered questionnaire that elicits a range of data. Nurses also assess participants for various health conditions, and obtain blood samples for HIV and HCV serologic testing, and HIV disease monitoring, as appropriate. Participants receive a $40 (CAD) honorarium for each visit. Both cohorts have received approvals from the University of British Columbia/Providence Health Care Research Ethics Board.
Eligibility criteria for the present study included: completing at least four follow-up visits between May 1, 1996 and November 30, 2017 to allow for polynomial growth curve analysis, and reporting having used illicit drugs via injection in the past six months at baseline.
Measures
The main outcome of interest was a time-varying dichotomous variable of self-reported injection drug use during the past six months (yes vs. no). This was assessed by asking the participants, “In the last 6 months, when you were using, which of the following drugs did you inject?” The outcome variable was defined as yes if participants indicated injecting substances including heroin alone, cocaine alone, heroin and cocaine, crystal methamphetamine, heroin and crystal meth, crack cocaine, fentanyl powder/pills, benzos, prescription opioids, and any other specified drugs. We also sought to evaluate whether specific baseline individual characteristics, substance use behaviours, and social-structural exposures were associated with different injection drug use trajectories. We included self-reported baseline characteristics, including age (per year decrease), sex (male vs. female), and ethnicity (white vs. others). Other sociodemographic factors included: education attainment (high school completion or higher vs. less than high school); not being in a stable relationship, defined as not being married, common law or having a regular partner; employment, defined as having a regular job, temporary job or self-employed; current housing status (unstable housing vs. stable housing). Unstable housing was defined as living in a single room occupancy hotel, shelter or other transitional housing, or living on the street. Substance use factors were categorized as daily vs. less than daily use, and included: heroin injection; stimulant (i.e., cocaine or crystal methamphetamine) injection; speedball injection (i.e., heroin and cocaine in combination); and prescription opioid injection. Non-injection drug use included crack cocaine use, cannabis use, and alcohol use. Binge injection drug use (yes vs. no) and years since injection drug use initiation were also included as substance use factors. Binge injection drug use was assessed using the survey question “In the last six months, did you go on runs or binges (that is, when you injected drugs more than usual)?” Factors related to substance use treatment experience included: opioid agonist therapy (e.g., methadone maintenance treatment or buprenorphine/naloxone, yes vs. no), any other addiction treatment or services except for opioid agonist therapy (yes vs. no), as well as being unable to access addiction treatment (yes vs. no). Other behavioural risk factors/outcomes and social-structural exposures were categorized as yes vs. no, and included: being attacked, assaulted, or suffered violence; drug dealing; sex work involvement; incarceration; and non-fatal overdose. All behavioural variables referred to the previous six months unless otherwise specified. We also included a variable asking participants if they have ever been diagnosed with a mental health issue (yes vs. no), calendar year of study enrollment (per year increase), and study cohort designation (ACCESS vs. VIDUS).
Statistical analysis
To identify the optimal number of groups, we applied growth mixture modeling (GMM), which is a semi-parametric, group-based analytical approach [33-35]. The objective of the model is to discover meaningful distinctive subpopulations with homogeneous longitudinal trajectories within the larger heterogeneous population.
The metric of time used for the GMM analysis was time since study enrollment due to the nature of our dynamic cohort study design. We started with a single-class latent growth curve model and continued until a six-class model was fitted. Linear, quadratic, and cubic parameters were fitted for the time trend for each trajectory group. To avoid local maximization and ensure successful convergence, 500 random sets of starting values are used for each model. The models were compared using the following fit indices: Akaike information criterion (AIC), Bayesian information criterion (BIC), and Lo-Mendell-Rubin likelihood ratio test (LRT) [35, 36]. Lower absolute values for the information criteria indices and significant LRT P value suggest better model fit. Averaged posterior probability of group membership and entropy were used to evaluate classification quality. Furthermore, to ensure interpretability and usefulness of the latent classes, sample size per latent class and substantive importance of the trajectory groups were also considered.
Then, we further evaluated whether or not any of the baseline individual characteristics, drug use behaviours, social-structural exposures would predict and help distinguish trajectory groups. To do so, we first summarized and compared the characteristics among trajectory groups, using Pearson’s chi-square test for categorical variables and Kruskal-Wallis test for continuous variables. Next, we applied the standard three-step method, in which class membership was merged with the original data and used as an outcome in a multinomial logistic regression analysis [36]. We used an a priori-defined backward model selection procedure based on examination of AIC to fit a multivariable model. The multivariable model with the lowest AIC score was selected as the final model.
As a sensitivity analysis, we defined lost to follow-up as being alive but not seen within three years before the end of study period. Information on death, including date and underlying causes of death, was obtained through a confidential data linkage with the British Columbia Vital Statistics Agency. To assess the extent to which losses to follow-up and deaths affected the trajectory membership, we proceeded with the Roy latent dropout pattern-mixture modeling [37, 38], which can take into account potential nonignorable dropout. For individuals who were lost to follow-up or died, their first missed study visit was treated as the time of dropout. Additionally, considering that participants with various follow-up time, we conducted a second sensitivity analysis to examine the injection drug use trajectories among people with at least ten years follow-up time.
Data manipulations and multinomial regression analysis were conducted using SAS 9.4 (SAS Institute, USA). GMM was applied using the software Mplus version 8 [39]. All P values were two-sided.
RESULTS
General characteristics
Between May 1996 and November 2017, a total of 3146 individuals enrolled in the cohorts. 1089 were excluded from the present study: 236 did not report having injected drugs at baseline and 853 did not complete at least four follow-up visits. Compared with participants in the analytical sample (n = 2057), those excluded were younger at enrollment (median age: 34 years; P < 0.001), having injected drugs for less years (median time since injection drug use initiation: 9.5 years; P < 0.001), but there was no difference by sex (P = 0.098), ethnicity (P = 0.166), and risk of death (22.9%, P = 0.487).
Among 2057 included participants, 1309 (63.6%) were male, 1195 (58.1%) self-reported white ethnicity, and the median age was 37 years (quartile 1 – quartile 3: 30 – 44). There were 740 participants who had ever been diagnosed with a mental health issue, and the most commonly reported diagnoses included depression (n = 441, 59.6%), anxiety (n = 163, 22.0%), bipolar disorder (n = 96, 13.0%), post-traumatic stress disorder (n = 90, 12.2%), and attention deficit disorder (n = 62, 8.4%). These participants contributed 36679 observations with the median number of months between study visits as 6.0 (quartile 1 – quartile 3: 5.7 – 6.8). The median time of follow-up per participant was 113.4 months (quartile 1 – quartile 3: 63.4 – 161.7) and the median number of observations per participant was 16 visits (quartile 1 – quartile 3: 9 – 23). Through the study period, the total number of identified deaths was 495, including 103 (20.8%) HIV-related deaths and 97 (19.6%) fatal overdose deaths. 462 (22.5%) participants were identified as lost to follow-up. Over time, there were 182 deaths and 181 dropouts observed within 5 years of follow-up, 188 deaths and 189 dropouts between the 6th to 10th years of follow-up, 81 deaths and 57 dropouts between the 11th to 15th years of follow-up, and 44 deaths and 35 dropouts after the 15th year. Baseline demographic characteristics, behavioural factors, and follow-up information are shown in Table 1.
Table 1.
Baseline demographic characteristics, drug use behaviours, social-structural exposures, and follow-up information among 2057 people who use injection drugs, the Vancouver Injection Drug Users Study and AIDS Care Cohort to Evaluate Exposure to Survival Services, Vancouver, British Columbia, Canada, 1996-2017.
| Characteristics | Total n = 2057 (100%) |
Persistent injection n = 507 (24.6%) |
Persistent injection with late cessation n = 374 (18.2%) |
Gradual cessation n = 662 (32.2%) |
Early cessation with late relapse n = 227 (11.0%) |
Early cessation n = 287 (14.0%) |
P |
|---|---|---|---|---|---|---|---|
| Sociodemographic factors | |||||||
| Age (years), Median (quartile 1 – quartile 3) | 37 (30 – 44) | 36 (28 – 43) | 38 (30 – 44) | 37 (31 – 44) | 39 (29 – 44) | 38 (31 – 45) | 0.080 |
| Male | 1309 (63.6) | 296 (58.4) | 255 (68.2) | 434 (65.6) | 145 (63.9) | 179 (62.4) | 0.031 |
| White ethnicity | 1195 (58.1) | 289 (57.0) | 212 (56.7) | 388 (58.6) | 148 (65.2) | 158 (55.1) | 0.170 |
| High school completion or higher | 1387 (67.4) | 345 (68.1) | 236 (63.1) | 419 (63.3) | 176 (77.5) | 211 (73.5) | <0.001 |
| Not being in a stable relationship a | 1482 (72.1) | 378 (74.6) | 281 (75.1) | 480 (72.5) | 155 (68.3) | 188 (65.5) | 0.018 |
| Employment status (regular/temporary job; self-employed) a | 401 (19.5) | 87 (17.2) | 72 (19.3) | 143 (21.6) | 46 (20.3) | 53 (18.5) | 0.416 |
| Unstable housing | 1453 (70.6) | 361 (71.2) | 280 (74.9) | 466 (70.4) | 148 (65.2) | 198 (69.0) | 0.110 |
| Substance use a | |||||||
| Daily heroin injection | 810 (39.4) | 224 (44.2) | 127 (34.0) | 275 (41.5) | 83 (36.6) | 101 (35.2) | 0.008 |
| Daily stimulant injection | 678 (33.0) | 182 (35.9) | 98 (26.2) | 228 (34.4) | 88 (38.8) | 82 (28.6) | 0.002 |
| Daily speedball injection | 249 (12.1) | 80 (15.8) | 33 (8.8) | 79 (11.9) | 26 (11.5) | 31 (10.8) | 0.029 |
| Daily prescription opioid injection | 78 (3.8) | 12 (2.4) | 20 (5.4) | 29 (4.4) | 7 (3.1) | 10 (3.5) | 0.177 |
| Daily non-injection crack cocaine use | 497 (24.2) | 111 (21.9) | 119 (31.8) | 167 (25.2) | 37 (16.3) | 63 (22.0) | <0.001 |
| Daily cannabis use | 393 (19.1) | 93 (18.3) | 87 (23.3) | 112 (16.9) | 45 (19.8) | 56 (19.5) | 0.164 |
| Daily alcohol use | 372 (18.1) | 90 (17.8) | 54 (14.4) | 122 (18.4) | 47 (20.7) | 59 (20.6) | 0.219 |
| Binge injection drug use b | 840 (40.8) | 226 (44.6) | 128 (34.2) | 297 (44.9) | 93 (41.0) | 96 (33.5) | <0.001 |
| Years since injection drug use (quartile 1 – quartile 3) | 14.8 (6.5 – 24.3) | 13.4 (5.2 – 24.1) | 15.4 (7.4 – 24.5) | 15.5 (7.2 – 23.7) | 14.8 (6.6 – 25.9) | 15.0 (5.7 – 24.9) | 0.244 |
| Treatment experience a, b | |||||||
| Opioid agonist therapy | 520 (25.3) | 116 (22.9) | 111 (29.7) | 171 (25.8) | 54 (23.8) | 68 (23.7) | 0.178 |
| Any other addiction treatment or services | 191 (9.3) | 42 (8.3) | 40 (10.7) | 56 (8.5) | 21 (9.3) | 32 (11.2) | 0.518 |
| Unable to access addiction treatment | 302 (14.7) | 74 (14.6) | 47 (12.6) | 92 (13.9) | 39 (17.2) | 50 (17.4) | 0.341 |
| Behavioural risk factors a, b | |||||||
| Attacked, assaulted, or suffered violence | 233 (11.3) | 47 (9.3) | 62 (16.6) | 81 (12.2) | 14 (6.2) | 29 (10.1) | 0.684 |
| Drug dealing | 299 (14.5) | 61 (12.0) | 74 (19.8) | 118 (17.8) | 16 (7.1) | 30 (10.5) | <0.001 |
| Sex work involvement | 505 (24.6) | 144 (28.4) | 75 (20.1) | 166 (25.1) | 51 (22.5) | 69 (24.0) | 0.060 |
| Incarceration | 283 (13.8) | 76 (15.0) | 51 (13.6) | 93 (14.1) | 27 (11.9) | 36 (12.5) | 0.786 |
| Non-fatal overdose | 272 (13.2) | 77 (15.2) | 54 (14.4) | 84 (12.7) | 28 (12.3) | 29 (10.1) | 0.291 |
| Other factors | |||||||
| Ever been diagnosed with a mental health issue | 740 (36.0) | 179 (35.3) | 167 (44.7) | 233 (35.2) | 65 (28.6) | 96 (33.5) | 0.001 |
| Year of enrollment, median (quartile1 – quartile3) | 1999 (1996 – 2006) | 1998 (1996 – 2006) | 2006 (1996 – 2007) | 2003 (1996 – 2007) | 1996 (1996 – 2006) | 1998 (1996 – 2006) | <0.001 |
| Study cohort designation (ACCESS) | 370 (18.0) | 65 (12.8) | 98 (26.2) | 143 (21.6) | 21 (9.3) | 43 (15.0) | <0.001 |
| Follow-up information | |||||||
| Length of follow-up (months), median (quartile1 – quartile3) | 113.4(63.4 – 161.7) | 124.9(50.6 – 211.6) | 118.8(73.0 – 149.5) | 107.1(58.1 – 135.5) | 131.3(96.6 – 211.3) | 102.8(57.2 – 138.9) | <0.001 |
| Number of deaths | 495 (24.1) | 126 (24.9) | 73 (19.5) | 180 (27.2) | 50 (22.0) | 66 (23.0) | 0.071 |
| Lost to follow-up | 462 (22.5) | 102 (20.1) | 71 (19.0) | 126 (19.0) | 67 (29.5) | 96 (33.4) | <0.001 |
Denotes behaviours and events in the previous six months, measured at baseline.
Frequency refers to the number of individuals who responded yes to the measure.
Identify injection drug use trajectories
As shown in Table 2, different model fit statistics were compared with an increasing number of trajectories. Both AIC and BIC values continued to decrease, which was expected given the complexity of longitudinal data derived from this large sample. However, the reduction was relatively small when comparing models with four to six trajectory classes. LRT compared the likelihood of the model being tested with a model with one fewer class and suggested that model was no longer improved with 6 classes (P = 0.515). Also taking into account the classification quality and interpretability, a five-class solution was chosen. The averaged posterior probability of group membership for the five-class solution is presented in appendix Table S1. The trajectories of injection drug use are visualized in Figure 1.
Table 2.
Model comparison with an increasing number of trajectories.
| Number of Classes | 1 | 2 | 3 | 4 | 5 | 6 |
|---|---|---|---|---|---|---|
| AIC | 28579.5 | 27082.1 | 26480.5 | 26273.1 | 26108.3 | 26004.8 |
| BIC | 28607.7 | 27138.4 | 26565.0 | 26385.6 | 26249.0 | 26173.6 |
| LMR LRT P value | 0.00 | 0.00 | 0.02 | 0.01 | 0.52 | |
| Entropy | 0.446 | 0.455 | 0.446 | 0.440 | 0.411 | |
| Averaged posterior probability of group membership (range) | 0.79 – 0.82 | 0.59 – 0.81 | 0.59 – 0.76 | 0.51 – 0.74 | 0.47 – 0.71 | |
| Sample size per class based on the estimated model (%, range) | 45.8 – 54.2 | 26.7 – 41.7 | 11.9 – 36.1 | 13.4 – 28.0 | 11.4 – 22.2 |
LMR LRT = Lo-Mendell-Rubin likelihood ratio test.
Figure 1.
Injection drug use trajectory classes using growth mixture modeling among 2057 people who use injection drugs in Vancouver, British Columbia, Canada, 1996-2017.
After assigning participants to each trajectory class based on their most likely latent class membership, we characterized the five classes as: (1) “persistent injection” (n=507, 24.6%); (2) “persistent injection with late cessation” (n=374, 18.2%); (3) “gradual cessation” (n=662, 32.2%); (4) “early cessation with late relapse” (n=227, 11.0%); and (5) “early cessation” (287, 14.0%). The probability of injection drug use among the “persistent injection” group remained high, ranging from 65.1% to 95.3% over the study period. For the “persistent injection with late cessation” group, the probability of injection stayed relatively high (>70.0%) till around the 18th year since baseline, and then showed a decreasing trend. We also observed three trajectory classes which represented distinctive injection cessation and relapse patterns. The probability of injection for “gradual cessation” group declined steadily to around 25.0% over the study period. As shown in Figure 1, the probability for the “early cessation” group rapidly dropped to as low as 13.1% at the 6th year, then stayed below 10.0% after the 14th year, and maintained less than 1.0% from the 16th year. The probability for the “early cessation with late relapse” group also declined significantly at an early stage with the lowest of 7.6% at the 12th year. However, it increased gradually afterwards showing a tendency towards relapse.
Baseline predictors of trajectory group membership
The results of bivariable and multivariable multinomial logistic regression analyses of factors associated with injection trajectories are presented in Table 3 and Table 4. Bivariable multinomial logistic regression analyses with the alternative trajectory group as the reference group are presented in appendices Table S2-S5. In the adjusted model, compared to the “early cessation” group, participants of younger age were more likely to be in the “persistent injection” group (adjusted odds ratio (AOR) = 1.02, 95% confidence interval (CI): 1.00, 1.04) and the “persistent injection with late cessation” group (AOR = 1.03, 95% CI: 1.01, 1.04). Further, not being in a stable relationship was positively associated with being in the “persistent injection” group (AOR = 1.64, 95% CI: 1.18, 2.27), the “persistent injection with late cessation” group (AOR = 1.59, 95% CI: 1.12, 2.27), and the “gradual cessation” group (AOR = 1.44, 95% CI: 1.06, 1.97).
Table 3.
Bivariable multinomial logistic regression analyses of baseline demographic characteristics, drug use behaviours, and structural-social exposures associated with different trajectory classes among 2057 people who use injection drugs, the Vancouver Injection Drug Users Study and AIDS Care Cohort to Evaluate Exposure to Survival Services, Vancouver, British Columbia, Canada, 1996-2017.
| Odds Ratio (95% Confidence Interval), early cessation as the reference group |
||||||||
|---|---|---|---|---|---|---|---|---|
| Characteristics | Persistent injection |
P | Persistent injection with late cessation |
P | Gradual cessation |
P | Early cessation with late relapse |
P |
| Sociodemographic factors | ||||||||
| Younger age, per year | 1.02 (1.00, 1.03) | 0.019 | 1.00 (0.99, 1.02) | 0.640 | 1.00 (0.99, 1.02) | 0.577 | 1.00 (0.99, 1.02) | 0.563 |
| Male | 0.85 (0.63, 1.14) | 0.271 | 1.29 (0.94, 1.79) | 0.119 | 1.15 (0.86, 1.53) | 0.346 | 1.07 (0.74, 1.53) | 0.725 |
| White ethnicity | 1.08 (0.81, 1.45) | 0.595 | 1.07 (0.78, 1.46) | 0.675 | 1.16 (0.87, 1.53) | 0.309 | 1.53 (1.07, 2.19) | 0.020 |
| High school completion or higher | 0.79 (0.57, 1.09) | 0.146 | 0.65 (0.46, 0.91) | 0.011 | 0.63 (0.46, 0.85) | 0.003 | 1.25 (0.83, 1.89) | 0.284 |
| Not being in a stable relationship a | 1.56 (1.14, 2.15) | 0.006 | 1.59 (1.13, 2.24) | 0.007 | 1.43 (1.06, 1.93) | 0.019 | 1.12 (0.77, 1.63) | 0.542 |
| Employment status (regular/temporary job; self-employed) a | 0.92 (0.63, 1.33) | 0.643 | 1.05 (0.71, 1.56) | 0.799 | 1.22 (0.86, 1.73) | 0.274 | 1.12 (0.72, 1.74) | 0.608 |
| Unstable housing | 1.12 (0.82, 1.54) | 0.477 | 1.35 (0.96, 1.91) | 0.086 | 1.09 (0.81, 1.48) | 0.576 | 0.83 (0.58, 1.21) | 0.333 |
| Substance use a | ||||||||
| Daily heroin injection | 1.46 (1.09, 1.97) | 0.013 | 0.95 (0.69, 1.31) | 0.741 | 1.31 (0.98, 1.74) | 0.067 | 1.06 (0.74, 1.53) | 0.747 |
| Daily stimulant injection | 1.42 (1.03, 1.94) | 0.030 | 0.88 (0.63, 1.25) | 0.481 | 1.32 (0.97, 1.78) | 0.075 | 1.58 (1.09, 2.28) | 0.016 |
| Daily speedball injection | 1.56 (1.00, 2.42) | 0.051 | 0.80 (0.48, 1.34) | 0.402 | 1.14 (0.73, 1.77) | 0.569 | 1.07 (0.62, 1.86) | 0.815 |
| Daily prescription opioid injection | 0.67 (0.29, 1.58) | 0.361 | 1.57 (0.72, 3.40) | 0.258 | 1.27 (0.61, 2.64) | 0.524 | 0.88 (0.33, 2.35) | 0.800 |
| Daily non-injection crack cocaine use | 1.00 (0.70, 1.41) | 0.985 | 1.66 (1.16, 2.37) | 0.005 | 1.20 (0.86, 1.67) | 0.275 | 0.70 (0.45, 1.10) | 0.120 |
| Daily cannabis use | 0.93 (0.64, 1.34) | 0.695 | 1.25 (0.86, 1.83) | 0.246 | 0.84 (0.59, 1.20) | 0.337 | 1.02 (0.66, 1.58) | 0.930 |
| Daily alcohol use | 0.84 (0.58, 1.21) | 0.338 | 0.65 (0.43, 0.98) | 0.039 | 0.88 (0.62, 1.24) | 0.450 | 1.02 (0.66, 1.56) | 0.947 |
| Binge injection drug use | 1.59 (1.18, 2.15) | 0.003 | 1.07 (0.77, 1.48) | 0.689 | 1.62 (1.21, 2.16) | 0.001 | 1.37 (0.96, 1.97) | 0.085 |
| Years since injection drug use | 0.99 (0.98, 1.01) | 0.236 | 1.01 (0.99, 1.02) | 0.523 | 1.00 (0.99, 1.02) | 0.802 | 1.00 (0.99, 1.02) | 0.893 |
| Treatment experience a | ||||||||
| Opioid agonist therapy | 0.95 (0.68, 1.34) | 0.785 | 1.36 (0.96, 1.94) | 0.084 | 1.12 (0.81, 1.55) | 0.496 | 1.00 (0.67, 1.51) | 0.997 |
| Any other addiction treatment or services | 0.72 (0.44, 1.17) | 0.181 | 0.95 (0.58, 1.56) | 0.850 | 0.74 (0.47, 1.16) | 0.187 | 0.81 (0.45, 1.45) | 0.474 |
| Unable to access addiction treatment | 0.81 (0.55, 1.20) | 0.288 | 0.68 (0.44, 1.05) | 0.081 | 0.76 (0.52, 1.11) | 0.159 | 0.98 (0.62, 1.55) | 0.929 |
| Behavioural risk factors a | ||||||||
| Attacked, assaulted, or suffered violence | 0.88 (0.51, 1.52) | 0.645 | 1.04 (0.62, 1.76) | 0.886 | 0.80 (0.48, 1.32) | 0.377 | 0.77 (0.37, 1.61) | 0.485 |
| Drug dealing | 1.17 (0.74, 1.86) | 0.503 | 2.11 (1.34, 3.33) | 0.001 | 1.86 (1.21, 2.85) | 0.005 | 0.65 (0.35, 1.22) | 0.182 |
| Sex work involvement | 1.26 (0.90, 1.76) | 0.173 | 0.79 (0.55, 1.15) | 0.219 | 1.06 (0.77, 1.47) | 0.707 | 0.92 (0.61, 1.38) | 0.675 |
| Incarceration | 1.23 (0.80, 1.88) | 0.342 | 1.10 (0.70, 1.74) | 0.681 | 1.14 (0.76, 1.72) | 0.529 | 0.94 (0.55, 1.60) | 0.824 |
| Non-fatal overdose | 1.60 (1.01, 2.51) | 0.044 | 1.51 (0.93, 2.43) | 0.095 | 1.30 (0.83, 2.03) | 0.254 | 1.26 (0.73, 2.18) | 0.415 |
| Other factors | ||||||||
| Ever been diagnosed with a mental health issue | 1.08 (0.80, 1.47) | 0.621 | 1.61 (1.17, 2.21) | 0.004 | 1.08 (0.81, 1.45) | 0.596 | 0.80 (0.55, 1.17) | 0.246 |
| Year of enrollment | 1.01 (0.98, 1.04) | 0.483 | 1.07 (1.04, 1.10) | <0.001 | 1.04 (1.02, 1.07) | <0.001 | 0.95 (0.92, 0.98) | 0.001 |
| Study cohort designation (ACCESS) | 0.83 (0.55, 1.27) | 0.394 | 2.02 (1.35, 3.00) | <0.001 | 1.56 (1.08, 2.27) | 0.019 | 0.58 (0.33, 1.01) | 0.053 |
Denotes behaviours and events in the previous six months, measured at baseline.
Table 4.
Multivariable multinomial logistic regression analyses of baseline demographic characteristics, drug use behaviours, and social-structural exposures associated with different trajectory classes among 2057 people who use injection drugs, the Vancouver Injection Drug Users Study and AIDS Care Cohort to Evaluate Exposure to Survival Services, Vancouver, British Columbia, Canada, 1996-2017.
| Adjusted Odds Ratio (95% Confidence Interval) |
||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Characteristics | Persistent injection vs. Early cessation |
P | Persistent injection with late cessation vs. Early cessation |
P | Gradual cessation vs. Early cessation |
P | Early decrease with late relapse vs. Early cessation |
P | Persistent injection vs. Early decrease with late relapse |
P |
| Younger age, per year | 1.02 (1.00, 1.04) | 0.041 | 1.03 (1.01, 1.04) | 0.007 | 1.01 (1.00, 1.03) | 0.133 | 1.00 (0.98, 1.02) | 0.847 | 1.02 (1.00, 1.04) | 0.036 |
| Male | 0.86 (0.63, 1.18) | 0.358 | 1.28 (0.91, 1.81) | 0.162 | 1.19 (0.87, 1.62) | 0.278 | 1.10 (0.75, 1.61) | 0.635 | 0.79 (0.55, 1.11) | 0.172 |
| Not being in a stable relationship a | 1.64 (1.18, 2.27) | 0.003 | 1.59 (1.12, 2.27) | 0.010 | 1.44 (1.06, 1.97) | 0.020 | 1.12 (0.76, 1.64) | 0.566 | 1.46 (1.02, 2.09) | 0.038 |
| Daily heroin injection a | 1.41 (1.03, 1.93) | 0.030 | 1.01 (0.72, 1.42) | 0.949 | 1.36 (1.01, 1.83) | 0.046 | 1.01 (0.70, 1.47) | 0.956 | 1.40 (1.00, 1.95) | 0.051 |
| Binge injection drug use a | 1.63 (1.19, 2.22) | 0.002 | 1.26 (0.90, 1.77) | 0.183 | 1.77 (1.31, 2.40) | <0.001 | 1.23 (0.85, 1.78) | 0.277 | 1.32 (0.95, 1.84) | 0.096 |
| Year of enrollment | 1.03 (1.00, 1.06) | 0.061 | 1.08 (1.05, 1.12) | <0.001 | 1.06 (1.03, 1.09) | <0.001 | 0.95 (0.92, 0.99) | 0.006 | 1.08 (1.05, 1.12) | <0.001 |
| Adjusted Odds Ratio (95% Confidence Interval) |
||||||||||
| Characteristics | Persistent injection with late cessation vs. Early decrease with late relapse |
P | Gradual cessation vs. Early decrease with late relapse |
P | Persistent injection vs. Gradual cessation |
P | Persistent injection with late cessation vs. Gradual cessation |
P | Persistent injection vs. Persistent injection with late cessation |
P |
| Younger age, per year | 1.03 (1.01, 1.05) | 0.007 | 1.02 (1.00, 1.03) | 0.112 | 1.01 (0.99, 1.02) | 0.451 | 1.01 (1.00, 1.03) | 0.100 | 0.99 (0.98, 1.01) | 0.366 |
| Male | 1.17 (0.80, 1.70) | 0.421 | 1.08 (0.77, 1.52) | 0.653 | 0.73 (0.56, 0.94) | 0.015 | 1.08 (0.81, 1.44) | 0.608 | 0.67 (0.50, 0.91) | 0.010 |
| Not being in a stable relationship a | 1.42 (0.97, 2.09) | 0.071 | 1.29 (0.92, 1.82) | 0.146 | 1.13 (0.86, 1.49) | 0.375 | 1.10 (0.81, 1.49) | 0.528 | 1.03 (0.74, 1.42) | 0.871 |
| Daily heroin injection a | 1.00 (0.70, 1.44) | 0.998 | 1.34 (0.97, 1.86) | 0.075 | 1.04 (0.81, 1.33) | 0.761 | 0.75 (0.56, 0.98) | 0.037 | 1.40 (1.04, 1.87) | 0.025 |
| Binge injection drug use a | 1.02 (0.72, 1.46) | 0.895 | 1.44 (1.05, 1.99) | 0.025 | 0.92 (0.72, 1.17) | 0.480 | 0.71 (0.54, 0.93) | 0.014 | 1.29 (0.97, 1.72) | 0.081 |
| Year of enrollment | 1.14 (1.10, 1.18) | <0.001 | 1.12 (1.08, 1.15) | <0.001 | 0.97 (0.95, 0.99) | 0.004 | 1.02 (1.00, 1.05) | 0.064 | 0.95 (0.92, 0.97) | <0.001 |
Denotes behaviours and events in the previous six months, measured at baseline.
Baseline substance use factors, including at least daily heroin injection and binge injection drug use, were found to be predictive of long-term trajectories. Specifically, compared to the “early cessation” group, participants who injected heroin daily were more likely to be in the “persistent injection” group (AOR = 1.41, 95% CI: 1.03, 1.93) and the “gradual cessation” group (AOR = 1.36, 95% CI: 1.01, 1.83). Moreover, participants who engaged in binge injection drug use at baseline were at 63% increased odds of being in the “persistent injection” group (AOR = 1.63, 95% CI: 1.19, 2.22), and 77% increased odds of being in the “gradual cessation” group (AOR = 1.77, 95% CI: 1.31, 2.40).
Sensitivity analysis results
In sensitivity analysis, the 462 participants who were lost to follow-up and the 495 participants who died were treated as dropouts at the time of their first missed study visit. When applying the Roy latent dropout pattern-mixture model, we found a similar class membership distribution, with: 489 (23.8%) participants in the “persistent injection” group, 461 (22.4%) participants in the “persistent injection with late cessation” group, 555 (27.0%) participants in the “gradual cessation” group, 211 (10.2%) participants in the “early cessation with late relapse” group, and 341 (16.6%) participants in the “early cessation” group. Results regarding class membership and the trajectory plot comparing to the standard GMM are presented in appendices Table S6 and Figure S1.
In the second sensitivity analysis, 974 participants were identified as having at least ten years follow-up time. Model fit statistics are presented in appendices Table S7. As shown in Figure S2, with a five-class solution, 433 (44.5%) participants injected drugs persistently over the study period, with 76 (7.8%) participants in the “persistent injection” group and 357 (36.7%) participants in the “persistent injection with late cessation” group. Overall, 541 (55.5%) participants demonstrated different cessation patterns. With four-class solution, 502 (51.5%) participants showed relatively persistent injection trajectories, and 49.5% of the participants demonstrated early and delayed cessation patterns (figure not shown).
DISCUSSION
In the present study, we identified five distinct injection drug use trajectories among people who inject drugs in Vancouver, Canada. Given that almost half of the study participants had injected drugs for 15 years at study enrollment, around one quarter of the participants remained persistent injectors during the study period, and only a small portion of participants achieved sustained cessation by the end of the study period. These trajectories displayed associations with several individual characteristics and drug use behaviours measured at baseline.
Our findings regarding injection drug use trajectories are comparable to studies conducted in the US [13, 19] and Netherlands [25], in which researchers also identified five trajectory groups. Further, our findings are consistent with previous findings indicating that the majority of participants experienced at least once cessation and relapse episode (either at early or late follow-up periods) during the study period. Specifically, three groups (i.e., “early cessation”, “early cessation with late relapse”, “gradual cessation”) demonstrated different rates of declining probability of injection. In the ALIVE cohort in the US, although 31.9% of the participants were categorized as persistent injectors, their estimated probability of injection dropped below 50.0% at the end of the study period. Whereas in our study, the estimated probability of injection for 24.6% of the participants remained above 65.0% over the study period. This difference could be potentially explained by distinct drug use patterns, cultural settings, or other social-structural conditions operating in the study sites.
The first sensitivity analysis took into account potential nonignorable dropout. The trajectory shapes as well as the interpretation of the latent classes were in line with the primary findings. Overall, for the Roy model, 46.2% of participants showed relatively persistent injection trajectories (i.e., the “persistent injection” and “persistent injection with late cessation” groups) over time, which was higher than the main result of 42.8%. Therefore, the Roy model indicated a slightly worse assessment of the injection drug use trajectories among the sample compared to the primary findings. In the second sensitivity analysis, the result was consistent with the primary findings in that over half of the entire sample showed a decreasing trend of injection drug use over time. However, with the restriction on longer follow-up time, the decreasing rate of injection drug use among the “early cessation” group was much slower. For example, in primary analysis, the estimated probability of injection drug use dropped below 30% around the 3rd year, whereas in sensitivity analysis, it was around the 6th year of follow-up. We also observed increased proportions for the “persistent injection with late cessation”, “early cessation”, and “delayed cessation” groups. In the primary analysis, the estimated probability of injection drug use for the “gradual cessation” group appeared to increase slightly from around 25% to 40% at the end of the three years of follow-up. This small increase did not indicate a clear relapsing pattern. Further, we did not observe such an increase in the “gradual cessation” group in the second sensitivity analysis with the restriction on longer follow-up time. Therefore, it is possible that the inflation was due to the small proportion of participants retained at the end of the study or a potential misclassification bias with the “early cessation with late relapse” group.
We observed both age and year of study enrollment were associated with different injection trajectories. Compared to the “early cessation” group, participants of younger age were at higher risk of engaging in persistent injection patterns (i.e., “persistent injection” and “persistent injection with late cessation”). This is consistent with the finding that older age was associated with six-month injection cessation in previous studies utilizing ACCESS and VIDUS cohorts [6, 40]. Studies in other settings also demonstrated that older age increased the likelihood of experiencing long-term cessation and decreased the risk of relapse [8, 41]. Compared to the “persistent injection” group, participants enrolled in more recent years were more likely to be in the “persistent injection with late cessation” and “gradual cessation” groups. It is possible that these findings are reflective of the continuing implementation and expansion of various harm reduction strategies in Canada. For instance, beginning in the year 2000, efforts have been made to transform the centralized syringe exchange program to a decentralized, multi-site syringe distribution program [42]. One study demonstrated a significant increase in the proportion of people in Vancouver reporting a six-month injection drug use cessation during the needle and syringe expansion period [7]. In 2003, North America’s first legal supervised injection site opened in Vancouver. Studies have indicated the potential role of supervised injecting facility in promoting the uptake of addiction treatment and increasing the likelihood of six-month injection cessation [9]. Together, these findings suggest that the age of participants and year of enrollment may be predictive of both short-term and long-term injection behavioural change, and indicate potential cohort and period effects.
Being in a stable relationship at baseline was positively associated with being in the “early cessation” group, suggesting that being married or having a close relationship is associated with a better long-term drug use outcome. It is possible that individuals in stable relationships may enjoy greater social support, which may serve to mitigate stress and thereby reduce the use of illicit drugs. This finding is further supported by an extensive literature examining relationship status and substance use patterns [43-45].
Binge injection drug use at baseline was associated with an increased likelihood of being in the “persistent injection” and “gradual cessation” groups. This finding is consistent with observations made in other studies that high intensity injecting not only impedes injection cessation, but is also negatively associated with sustained injection cessation [5, 7, 8, 19, 46]. We also found that drug of choice appears to have an impact on long-term drug use trajectories. Various types of substance use were associated with injection trajectories in bivariable analyses. However, only daily heroin injection remained positively associated with being in the “persistent injection” and “gradual cessation” groups in the multivariable model. Studies have shown that polydrug use is associated with change in injecting behaviour [12, 19, 25]. In the current study, speedball injection was only marginally associated with “persistent injection” in bivariable analyses, which provided limited information to help differentiate long-term drug use trajectories. Future studies investigating the relationship between early poly substance use (e.g., cumulative number of types of substances reported) and later trajectories would potentially be of benefit.
Baseline exposures were examined in our study in order to identify predictors of trajectory group membership. However, we recognize that injection behaviour could change under the influence of important events or cumulative exposure to these events over time. Therefore, there is a need for future study to more closely examine how these exposures, as time-varying covariates, would influence the shape of injection trajectories.
Limitations in the current study need to be considered. First, the use of self-report, especially for socially stigmatized and criminalized behaviours (e.g., illicit substance use), could introduce errors of recall and social-desirability bias. Second, the dichotomous variable of self-reported injection drug use (yes vs. no) was used for GMM, therefore, the identified trajectory classes could not differentiate between variability in frequency and quantity of injection drug use. Third, there could be potentially important baseline characteristics that were not included when predicting trajectory groups. For example, we were not able to include tobacco smoking since the information was not routinely measured in the survey questionnaires. Fourth, this study is not based on a random sample of people who inject drugs and is conducted in BC, Canada, where the social-structural conditions (e.g., harm reduction strategies, attitudes towards injection drug use) could be different compared to other settings. Therefore, these factors may limit the generalizability of the findings. Further, to explore the predictive factors for distinct drug use trajectories, we used the three-step method due to the practical reason that we had a large number of covariates and the recognized disadvantages of joint model estimation approach for our research purpose [36]. However, we recognize that this approach does not account for potential classification errors, although modified three-step approaches by Wang et al [47] and Vermunt [48] make allowances for issues such as errors in classification and bias adjustment.
Recent concerns have been noted regarding the validity of GMM analysis due to the “cat’s cradle” effect, which is a strong tendency to identify four prototypical classes in substance use research: stable high use, stable low use, increasing use, and decreasing use [49]. However, it is notable that in the current study, two relatively persistent injection trajectories and three different levels of decreasing trajectories were found. Furthermore, the “fling” trajectories (i.e., “persistent injection with late cessation”, “early cessation with late relapse”) were also identified. Together, these more interesting trajectories deviate from the prototypical classes, which suggest that the current study may be less susceptible to the cat’s cradle effect.
Finally, with the reported entropy value and the averaged posterior probability of group membership, there could be classification bias introduced when assigning classes. However, it is worth mentioning that entropy is not an ideal measure of model fit, nor should be solely used to select the number of latent classes [50]. Entropy value could be negatively influenced by chance misclassification when having a higher number of classes [51], be sensitive to the patterns of growth in each class [52], and potentially depend on the context and the variables used in the study [53, 54]. Even with this classification uncertainty, summarizing 20 years’ follow-up data at individual level in relation to several injection drug use trajectories has proven useful. Further, the primary findings were consistent with studies from other settings [13, 19, 25]. Several factors were found to be predictive of these trajectory classes, which further indicate that the identified trajectory classes were different from each other. Further, the 462 participants who were lost to follow-up, and 495 participants died, were treated as dropouts during the study period, which might also introduce bias regarding group classification. To account for attrition, our primary findings from GMM were based on maximum-likelihood estimation, which utilized all of the available data to generate parameter estimates. In both sensitivity analyses, the distribution of the trajectory class membership was consistent with the primary research findings that more than half of the sample demonstrated different cessation patterns. Recent developments offer a variety of more sophisticated methodologies (e.g., Muthén-Roy pattern-mixture model, Diggle-Kenward model) to understand missing data mechanisms in growth mixture modeling framework [38].
CONCLUSIONS
The current study took advantage of over 20 years of rich information from over 2000 individuals by creating a unique cohort combining ACCESS and VIDUS, which are among the longest-standing community-recruited cohort studies of people who inject drugs in the world. The study identified the existence of five distinct injection drug use trajectories among people who use drugs in Vancouver, Canada. Despite the fact that at recruitment many participants has already been injecting drugs for some 5-25 years, around three-fourths of them reduced the frequency of injection drug use at the end of the study period, and this estimate might be taken as an encouraging sign. This study represents a relatively comprehensive investigation of individual, social and structural factors for different trajectories of injection drug use across long-term periods. The findings from this study further highlight the importance of identifying targeted interventions for long-term injection drug use and the need to identify factors that support cessation and protect against relapse.
Supplementary Material
Acknowledgements
This work was supported by the US National Institutes of Health (U01DA038886, U01DA021525). Huiru Dong is supported through a Canadian Institutes of Health Research (CIHR) Doctoral Award. Dr. Thomas Kerr is supported by a foundation grant from the CIHR (20R74326). Dr. Evan Wood receives support through a Tier 1 Canada Research Chair in Inner City Medicine and the CIHR Canadian Research Initiative on Substance Misuse (SMN–139148). Dr. Kanna Hayashi is supported by a CIHR New Investigator Award (MSH-141971), a Michael Smith Foundation for Health Research (MSFHR) Scholar Award, and the St. Paul’s Foundation. Dr. Kora DeBeck is supported by a MSFHR/ St. Paul’s Hospital Foundation-Providence Health Care Career Scholar Award and a CIHR New Investigator Award. Dr. M-J Milloy is supported in part by the United States National Institutes of Health (U01DA021525), a New Investigator Award from the CIHR, and a Scholar Award from the MSFHR. His institution has received an unstructured gift to support his research from NG Biomed, Ltd., an applicant to the Canadian federal government for a license to produce medical cannabis. He is the Canopy Growth professor of cannabis science at the University of British Columbia, a position created by an unstructured gift to the university from Canopy Growth, a licensed producer of cannabis, and the Government of British Columbia’s Ministry of Mental Health and Addictions.
Funding sources had no role in the design of this study; collection, analysis, and interpretation of the data; writing of the report; or the decision to submit the paper for publication. The authors thank the study participants for their contribution to the research, as well as current and past researchers and staff.
Footnotes
Declare of interests: None.
References
- 1.World Health Organization. People who inject drugs. http://www.who.int/hiv/topics/idu/about/en/. Accessed September 10, 2018. [Google Scholar]
- 2.Mathers BM, Degenhardt L, Bucello C, Lemon J, Wiessing L, Hickman M Mortality among people who inject drugs: a systematic review and meta-analysis. Bull World Health Organ 2013; 91: 102–23. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Rehm J, Gnam W, Popova S, Baliunas D, Brochu S, Fischer B, et al. The costs of alcohol, illegal drugs, and tobacco in Canada, 2002. J Stud Alcohol Drugs 2007; 68: 886–95. [DOI] [PubMed] [Google Scholar]
- 4.Shah NG, Galai N, Celentano DD, Vlahov D, Strathdee SA Longitudinal predictors of injection cessation and subsequent relapse among a cohort of injection drug users in Baltimore, MD, 1988–2000. Drug Alcohol Depend 2006; 83: 147–56. [DOI] [PubMed] [Google Scholar]
- 5.Bruneau J, Brogly SB, Tyndall MW, Lamothe F, Franco EL Intensity of drug injection as a determinant of sustained injection cessation among chronic drug users: the interface with social factors and service utilization. Addiction 2004; 99: 727–37. [DOI] [PubMed] [Google Scholar]
- 6.DeBeck K, Kerr T, Li K, Milloy MJ, Montaner J, Wood E Incarceration and drug use patterns among a cohort of injection drug users. Addiction 2009; 104: 69–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Werb D, Kerr T, Buxton J, Shoveller J, Richardson C, Montaner J, et al. Patterns of injection drug use cessation during an expansion of syringe exchange services in a Canadian setting. Drug Alcohol Depend 2013; 132: 535–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Evans JL, Hahn JA, Lum PJ, Stein ES, Page K Predictors of injection drug use cessation and relapse in a prospective cohort of young injection drug users in San Francisco, CA (UFO Study). Drug Alcohol Depend 2009; 101: 152–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.DeBeck K, Kerr T, Bird L, Zhang R, Marsh D, Tyndall M, et al. Injection drug use cessation and use of North America's first medically supervised safer injecting facility. Drug Alcohol Depend 2011; 113: 172–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Linton SL, Celentano DD, Kirk GD, Mehta SH The longitudinal association between homelessness, injection drug use, and injection-related risk behavior among persons with a history of injection drug use in Baltimore, MD. Drug Alcohol Depend 2013; 132: 457–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Steensma C, Boivin JF, Blais L, Roy E Cessation of injecting drug use among street-based youth. J Urban Health 2005; 82: 622–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Horyniak D, Strathdee SA, West BS, Meacham M, Rangel G, Gaines TL Predictors of injecting cessation among a cohort of people who inject drugs in Tijuana, Mexico. Drug Alcohol Depend 2018; 185: 298–304. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Galai N, Safaeian M, Vlahov D, Bolotin A, Celentano DD, ALIVE Study. Longitudinal patterns of drug injection behavior in the ALIVE Study cohort, 1988-2000: description and determinants. Am J Epidemiol 2003; 158: 695–704. [DOI] [PubMed] [Google Scholar]
- 14.Hser YI, Hoffman V, Grella CE, Anglin MD A 33-year follow-up of narcotics addicts. Arch Gen Psychiatry 2001; 58: 503–8. [DOI] [PubMed] [Google Scholar]
- 15.Termorshuizen F, Krol A, Prins M, Geskus R, van den Brink W, van Ameijden EJ Prediction of relapse to frequent heroin use and the role of methadone prescription: an analysis of the Amsterdam Cohort Study among drug users. Drug Alcohol Depend 2005; 79: 231–40. [DOI] [PubMed] [Google Scholar]
- 16.Nosyk B, Anglin MD, Brecht ML, Lima VD, Hser YI Characterizing durations of heroin abstinence in the California Civil Addict Program: results from a 33-year observational cohort study. Am J Epidemiol 2013; 177: 675–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Darke S, Ross J, Mills KL, Williamson A, Havard A, Teesson M Patterns of sustained heroin abstinence amongst long-term, dependent heroin users: 36 months findings from the Australian Treatment Outcome Study (ATOS). Addict Behav 2007; 32: 1897–906. [DOI] [PubMed] [Google Scholar]
- 18.Darke S, Ross J, Teesson M, Ali R, Cooke R, Ritter A, et al. Factors associated with 12 months continuous heroin abstinence: findings from the Australian Treatment Outcome Study (ATOS). J Subst Abuse Treat 2005; 28: 255–63. [DOI] [PubMed] [Google Scholar]
- 19.Genberg BL, Gange SJ, Go VF, Celentano DD, Kirk GD, Mehta SH Trajectories of injection drug use over 20 years (1988-2008) in Baltimore, Maryland. Am J Epidemiol 2011; 173: 829–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Brecht ML, Herbeck D Time to relapse following treatment for methamphetamine use: a long-term perspective on patterns and predictors. Drug Alcohol Depend 2014; 139: 18–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Gossop M, Green L, Phillips G, Bradley B Lapse, relapse and survival among opiate addicts after treatment. A prospective follow-up study. Br J Psychiatry 1989; 154: 348–53. [DOI] [PubMed] [Google Scholar]
- 22.Hubbard RL, Marsden ME Relapse to use of heroin, cocaine, and other drugs in the first year after treatment. NIDA Res Monogr 1986; 72: 157–66. [PubMed] [Google Scholar]
- 23.Simpson DD, Joe GW, Fletcher BW, Hubbard RL, Anglin MD A national evaluation of treatment outcomes for cocaine dependence. Arch Gen Psychiatry 1999; 56: 507–14. [DOI] [PubMed] [Google Scholar]
- 24.Brewer DD, Catalano RF, Haggerty K, Gainey RR, Fleming CB A meta-analysis of predictors of continued drug use during and after treatment for opiate addiction. Addiction 1998; 93: 73–92. [PubMed] [Google Scholar]
- 25.Mikolajczyk RT, Horn J, Prins M, Wiessing L, Kretzschmar M Trajectories of injecting behavior in the Amsterdam Cohort Study among drug users. Drug Alcohol Depend 2014; 144: 141–7. [DOI] [PubMed] [Google Scholar]
- 26.Kerr T, Small W, Buchner C, Zhang R, Li K, Montaner J, et al. Syringe sharing and HIV incidence among injection drug users and increased access to sterile syringes. Am J Public Health 2010; 100: 1449–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Marshall BD, Milloy MJ, Wood E, Montaner JS, Kerr T Reduction in overdose mortality after the opening of North America's first medically supervised safer injecting facility: a retrospective population-based study. Lancet 2011; 377: 1429–37. [DOI] [PubMed] [Google Scholar]
- 28.Nosyk B, Marsh DC, Sun H, Schechter MT, Anis AH Trends in methadone maintenance treatment participation, retention, and compliance to dosing guidelines in British Columbia, Canada: 1996-2006. J Subst Abuse Treat 2010; 39: 22–31. [DOI] [PubMed] [Google Scholar]
- 29.Socias ME, Wood E, McNeil R, Kerr T, Dong H, Shoveller J, et al. Unintended impacts of regulatory changes to British Columbia Methadone Maintenance Program on addiction and HIV-related outcomes: An interrupted time series analysis. Int J Drug Policy 2017; 45: 1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Strathdee SA, Palepu A, Cornelisse PG, Yip B, O'Shaughnessy MV, Montaner JS, et al. Barriers to use of free antiretroviral therapy in injection drug users. JAMA 1998; 280: 547–9. [DOI] [PubMed] [Google Scholar]
- 31.Wood E, Stoltz JA, Montaner JS, Kerr T Evaluating methamphetamine use and risks of injection initiation among street youth: the ARYS study. Harm Reduct J 2006; 3: 18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Wood E, Kerr T, Marshall BD, Li K, Zhang R, Hogg RS, et al. Longitudinal community plasma HIV-1 RNA concentrations and incidence of HIV-1 among injecting drug users: prospective cohort study. BMJ 2009; 338: b1649. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Muthén B, Muthén LK Integrating person-centered and variable-centered analyses: growth mixture modeling with latent trajectory classes. Alcohol Clin Exp Res 2000; 24: 882–91. [PubMed] [Google Scholar]
- 34.Nagin DS, Tremblay RE Analyzing developmental trajectories of distinct but related behaviors: a group-based method. Psychol Methods 2001; 6: 18–34. [DOI] [PubMed] [Google Scholar]
- 35.Jung T, Wickrama KAS An Introduction to Latent Class Growth Analysis and Growth Mixture Modeling. Social and Personality Psychology Compass 2008; 2: 302–17. [Google Scholar]
- 36.van d.S., Sijbrandij M, Winter SD, Depaoli S, Vermunt JK The GRoLTS-Checklist: Guidelines for Reporting on Latent Trajectory Studies. Structural Equation Modeling: A Multidisciplinary Journal 2017; 24: 451–67. [Google Scholar]
- 37.Roy J Modeling longitudinal data with nonignorable dropouts using a latent dropout class model. Biometrics 2003; 59: 829–36. [DOI] [PubMed] [Google Scholar]
- 38.Muthén B, Asparouhov T, Hunter AM, Leuchter AF Growth modeling with nonignorable dropout: alternative analyses of the STAR*D antidepressant trial. Psychol Methods 2011; 16: 17–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Muthén LK, Muthén BO MPlus User's Guide. Eighth Edition. Los Angeles, CA: Authors; 1998-2017. [Google Scholar]
- 40.Luchenski S, Ti L, Hayashi K, Dong H, Wood E, Kerr T Protective factors associated with short-term cessation of injection drug use among a Canadian cohort of people who inject drugs. Drug Alcohol Rev 2016; 35: 620–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Xia Y, Seaman S, Hickman M, Macleod J, Robertson R, Copeland L, et al. Factors affecting repeated cessations of injecting drug use and relapses during the entire injecting career among the Edinburgh Addiction Cohort. Drug Alcohol Depend 2015; 151: 76–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Hyshka E, Strathdee S, Wood E, Kerr T Needle exchange and the HIV epidemic in Vancouver: lessons learned from 15 years of research. International Journal of Drug Policy 2012; 23: 261–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Moos RH, Nichol AC, Moos BS Risk factors for symptom exacerbation among treated patients with substance use disorders. Addiction 2002; 97: 75–85. [DOI] [PubMed] [Google Scholar]
- 44.Winters J, Fals-Stewart W, O'farrell TJ, Birchler GR, Kelley ML Behavioral couples therapy for female substance-abusing patients: Effects on substance use and relationship adjustment. Journal of Consulting and Clinical psychology 2002; 70: 344. [DOI] [PubMed] [Google Scholar]
- 45.Heinz AJ, Wu J, Witkiewitz K, Epstein DH, Preston KL Marriage and relationship closeness as predictors of cocaine and heroin use. Addictive behaviors 2009; 34: 258–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Huo D, Bailey SL, Ouellet LJ Cessation of injection drug use and change in injection frequency: the Chicago Needle Exchange Evaluation Study. Addiction 2006; 101: 1606–13. [DOI] [PubMed] [Google Scholar]
- 47.Wang C, Hendricks Brown C, Bandeen-Roche K Residual diagnostics for growth mixture models: Examining the impact of a preventive intervention on multiple trajectories of aggressive behavior. Journal of the American Statistical Association 2005; 100: 1054–76. [Google Scholar]
- 48.Vermunt JK Latent class modeling with covariates: Two improved three-step approaches. Political analysis 2010; 18: 450–69. [Google Scholar]
- 49.Sher KJ, Jackson KM, Steinley D Alcohol use trajectories and the ubiquitous cat's cradle: cause for concern? J Abnorm Psychol 2011; 120: 322–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50.Tein JY, Coxe S, Cham H Statistical power to detect the correct number of classes in latent profile analysis. Struct Equ Modeling 2013; 20: 640–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 51.Collins LM, Lanza ST Latent class and latent transition analysis: With applications in the social, behavioral, and health sciences. John Wiley & Sons; 2010. [Google Scholar]
- 52.Morgan G, Beaujean A An investigation of growth mixture models for studying the Flynn Effect. Journal of Intelligence 2014; 2: 156–79. [Google Scholar]
- 53.Feldman BJ, Masyn KE, Conger RD New approaches to studying problem behaviors: A comparison of methods for modeling longitudinal, categorical adolescent drinking data. Dev Psychol 2009; 45: 652. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54.Muthén B, Asparouhov T Growth mixture modeling with non-normal distributions. Stat Med 2015; 34: 1041–58. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.