Declining Mortality Rates in HIV-Infected People Who Inject Drugs During a Seek-and-Treat Initiative in Vancouver, Canada, 1996–2014: A Prospective Cohort Study

Abstract

We estimated rates and predictors of death among a community-recruited prospective cohort of 961 human immunodeficiency virus (HIV)–infected people who inject drugs in Vancouver, Canada, between 1996 and 2014. The results demonstrated significant declines in age-adjusted all-cause and HIV-related mortality rates since 2010, coincident with the scale-up of a community-wide “seek-and-treat” campaign.

injection drug use, HIV, antiretroviral therapy, gender, mortality

High mortality rates among people who inject drugs (PWID) living with human immunodeficiency virus (HIV) has been well documented, with drug overdose and HIV/AIDS-associated conditions the leading causes of death, largely as a result of suboptimal access to HIV care [1]. However, there is scant literature examining the impact of recent efforts to expand access to antiretroviral therapy (ART), as prioritized in the United Nations Joint Programme on HIV/AIDS (UNAIDS) 2014 global HIV/AIDS strategy [2], on mortality among this population.

In Vancouver, British Columbia, Canada, extensive “seek-and-treat” HIV treatment interventions have been implemented through the local universal healthcare system over the past 2 decades. We sought to identify rates, causes, and predictors of mortality among HIV-infected PWID in this setting between 1996 and 2014.

METHODS

Study Design

Data were derived from the AIDS Care Cohort to evaluate Exposure to Survival Services (ACCESS), a prospective cohort study of HIV-infected people who use illicit drugs in Vancouver, Canada. Recruitment and data collection procedures have been described in detail elsewhere [3]. In brief, beginning in 1996, participants were recruited through self-referral and street-based outreach. Inclusion criteria included age ≥18 years, residence in the greater Vancouver area, HIV seropositivity, use of illicit drugs (other than cannabis) in the previous month, and written informed consent.

At baseline and semiannually thereafter, participants completed an interviewer-administered questionnaire soliciting data on demographics, drug use patterns, and other characteristics and exposures. At each visit, individuals also provided blood samples for serologic analyses (ie, hepatitis C virus [HCV] antibody seropositivity) and HIV clinical monitoring, and were compensated 30 Canadian dollars. Through a confidential linkage with the provincial Drug Treatment Program, a complete clinical profile of all CD4 T-cell counts, plasma viral load (VL) observations, exposure to antiretroviral agents, and treatment outcomes for each participant were obtained. In British Columbia, provision of all HIV treatment is centralized through a province-wide dispensary. All HIV/AIDS treatment and related care including all medications are provided at no cost through the province’s universal healthcare system. The ACCESS study has been approved by the University of British Columbia/Providence Health Care Research Ethics Board.

For this study, participants who completed the baseline visit and at least 1 follow-up visit between 1 May 1996 and 31 May 2014 and who reported having ever injected drugs at baseline or having initiated injecting during follow-up were eligible.

Measures

We ascertained mortality rates and underlying causes of death through a confidential record linkage with the British Columbia Vital Statistics Agency using each participant’s unique and persistent identifier issued to all residents of British Columbia. The Vital Statistics database recorded causes of death according to the International Classification of Diseases, 10th Revision. Based on previous studies [1, 4], we classified causes of death into the following 6 categories: HIV-related, overdose, liver-related, other accidental, other nonaccidental, and unknown causes (Supplementary Table 1).

For the analysis of predictors of death, the primary endpoint was all-cause mortality. We considered a range of explanatory variables that we hypothesized might be associated with time to mortality. Fixed variables included calendar year of study enrollment; self-reported ethnicity/ancestry (white vs others); and time since first initiation of injection drug use at baseline. The remaining variables referred to the 6-month period prior to each interview and were treated as time-varying variables. Of these, demographic and drug-using variables included age (per 10 years older); at least daily heroin injection; at least daily cocaine injection; at least daily methamphetamine injection; at least daily prescription opioid use (eg, street methadone, morphine); at least daily crack smoking; any cannabis use; and at least daily alcohol use. Social, structural, and environmental exposures included unstable housing; engagement in sex work; the cumulative number of incarceration events between baseline and up to the follow-up assessment (>5 times, 3–5 times, 1–2 times vs never); enrollment in opioid agonist therapy (OAT) including methadone and buprenorphine-naloxone; and enrollment in addiction treatment other than OAT (eg, outpatient treatment, counseling, detox, residential treatment programs, drug treatment courts). In British Columbia, naltrexone is not widely used as opioid antagonist therapy, compared with OAT, as oral naltrexone is the only formulation available and is not covered through the universal health insurance. Enrollment in naltrexone treatment would be captured in the variable “enrollment in addiction treatment other than OAT.” Clinical variables included HCV seropositivity; ART access and plasma HIV-1 RNA VL (defined as ≥1 day dispensed ART and VL of <50 copies/mL vs ≥1 day dispensed ART and VL of ≥50 copies/mL vs no days dispensed ART in the past 6 months); years since the first record in the registry; and initiating highly active antiretroviral therapy. We measured VL in each 6-month period as the median of all observations, or if there were none in a specific period, the closest observation to the 6-month period (within 365.25 days). The derived VL value was deemed as either nondetectable (<50 copies/mL) or detectable (≥50 copies/mL). Other variable definitions were also consistent with our previous work [3, 4].

Statistical Analyses

Given the known sex/gender-based differences in mortality among PWID [1], all analyses were stratified by sex (female vs male). For each cause of death, we calculated a crude mortality rate and a 95% confidence interval (CI) for the entire and sex-stratified samples, using the Poisson distribution. Then, we fit age-adjusted Poisson regression models including sex as the primary independent variable to calculate age-adjusted mortality rate ratio for each specific cause among women compared to men. To determine if the age-adjusted rates of mortality from all causes and 3 common causes have changed since the expansion of HIV treatment interventions in this setting, we subsequently fit sex-stratified age-adjusted Poisson regression models including calendar-year intervals (2010–2014 vs 2004–2009 vs 1996–2003) as the primary independent variable. Calendar-year intervals denoted a phased expansion of HIV treatment interventions in this setting: the initial rollout of ART (1996–2003), a steady expansion of ART access (2004–2009), and a community-wide scale-up of “seek-and-treat” interventions (2010–2014) [5, 6].

Next, we used Kaplan–Meier methods and the log-rank test to estimate and compare survival probabilities between men and women. Then, we used sex-stratified extended Cox regression to identify factors associated with time to all-cause mortality. As in a previous study [4], we used a priori–defined manual backward modeling procedure to construct multivariable extended Cox regression models. We selected models with the lowest Akaike information criterion value. All P values were 2-sided. All statistical analyses were performed using SAS software version 9.4 (SAS Institute, Cary, North Carolina).

RESULTS

Study Sample

In total, 961 participants were eligible for the present analyses and were followed for a median of 63.8 (interquartile range [IQR], 41.4–97.9) months. The median age at baseline was 39.4 (IQR, 32.2–45.6) years, and 353 (36.7%) were female. Other baseline characteristics stratified by sex and ART initiation before and after 2010 are presented in Supplementary Tables 2 and 3.

Rates, Causes, and Trends of Mortality

During follow-up, 297 deaths occurred, yielding a crude mortality rate of 4.6 (95% CI, 4.1–5.1) deaths per 100 person-years. The 3 most common causes of death were identical in both sexes: HIV-related (43.4%), other nonaccidental causes (eg, respiratory and circulatory diseases, neoplasms; 20.5%), and overdose (19.2%). The Kaplan–Meier analysis showed no statistically significant difference in the cumulative survival probability between men and women (log-rank P = .71). There was also no statistically significant difference in age-adjusted all-cause mortality rates between the sexes (adjusted rate ratio [ARR], 1.04 [95% CI, .82–1.31]). However, liver-related mortality rates were significantly lower among women (ARR, 0.28 [95% CI, .08–.99]). Detailed rates and causes of mortality are presented in Supplementary Table 4.

Supplementary Figure 1 depicts mortality rates for all-cause and the 3 common specific causes over the 3 calendar-year intervals. As shown, all-cause mortality rates per 100 person-years dropped from 6.0 (95% CI, 5.1–7.1) in 1996–2003 and 5.3 (95% CI, 4.4–6.4) in 2004–2009 to 3.0 (95% CI, 2.4–3.9) in 2010–2014. Table 1 presents the results of age-adjusted Poisson regression analyses. As shown, the calendar-year interval of 2010–2014 (compared to that of 1996–2003) was significantly associated with reduced all-cause and HIV-related mortality rates among both men and women, as well as reduced overdose and other nonaccidental mortality rates among men (all P < .05). Compared to 1996–2003, there was no significant reduction in HIV-related mortality rates in 2004–2009 among either sex, although all-cause and other nonaccidental mortality rates significantly declined among men (all P < .05).

Table 1.

Poisson Regression Analyses of Age-Adjusted Mortality Rates Among 961 HIV-Infected People Who Inject Drugs in Vancouver, Canada, 1996–2014

Calendar Year Intervals	Females (n = 353)		Males (n = 608)
	ARR (95% CI)	P Value	ARR (95% CI)	P Value
All-cause mortality
2010–2014 vs 1996–2003	0.24 (.14–.42)	<.001	0.20 (.13–.31)	<.001
2004–2009 vs 1996–2003	0.75 (.50–1.11)	.146	0.69 (.49–.97)	.033
HIV-related mortality
2010–2014 vs 1996–2003	0.15 (.06–.41)	<.001	0.12 (.05–.28)	<.001
2004–2009 vs 1996–2003	0.65 (.34–1.24)	.193	0.77 (.46–1.30)	.334
Overdose mortality
2010–2014 vs 1996–2003	0.30 (.08–1.10)	.070	0.16 (.05–.53)	.003
2004–2009 vs 1996–2003	0.81 (.33–1.98)	.651	0.61 (.26–1.40)	.241
Other nonaccidental mortalitya
2010–2014 vs 1996–2003	0.42 (.13–1.31)	.134	0.18 (.07–.46)	<.001
2004–2009 vs 1996–2003	1.19 (.48–2.98)	.702	0.37 (.16–.83)	.015

All models adjusted for age.

Abbreviations: ARR, adjusted rate ratio; CI, confidence interval; PWID, people who inject drugs.

^a Includes circulatory disease, respiratory disease, neoplasms, etc.

Predictors of All-Cause Mortality

In multivariable extended Cox regression analyses of all-cause mortality (Table 2), compared to those not receiving ART, those receiving ART and having undetectable VL (<50 copies/mL) had a 42% lower hazard of all-cause mortality among men (adjusted hazard ratio [AHR], 0.58 [95% CI, .39–.88]), and a 45% lower hazard among women (AHR, 0.55 [95% CI, .33–.91]). However, those receiving ART but having detectable VL did not have a significantly lower hazard of death. A longer duration of injection drug use (AHR, 1.03 [95% CI, 1.01–1.05]) independently predicted mortality among women only, as did at least daily prescription opioid use (AHR, 1.93 [95% CI, 1.24–3.00]) and white ethnicity/ancestry (AHR, 1.56 [95% CI, 1.13–2.16]) among men only.

Table 2.

Univariable and Multivariable Cox Regression Analyses of Factors Associated With All-Cause Mortality

Characteristic	Females (n = 353)		Males (n = 608)
	Unadjusted HR (95% CI)	Adjusted HR (95% CI)	Unadjusted HR (95% CI)	Adjusted HR (95% CI)
Calendar year of cohort enrollment (per year later)	0.93 (.89–.97)		0.92 (.89–.95)
Agea (per 10 years older)	1.14 (.94–1.39)		1.05 (.88–1.25)
Ethnicity/ancestry (white vs other)	1.06 (.74–1.51)		1.38 (1.02–1.89)	1.56 (1.13–2.16)
Years since first injection at baseline (per year longer)	1.02 (1.00–1.03)	1.03 (1.01–1.05)	1.01 (.99–1.02)
Heroin injectiona,b (≥daily vs <daily)	1.06 (.69–1.63)		1.17 (.79–1.73)
Cocaine injectiona,b (≥daily vs <daily)	1.18 (.76–1.82)		1.33 (.92–1.92)
Methamphetamine injectiona,b (≥daily vs <daily)	0.77 (.11–5.59)		0.22 (.03–1.54)
Prescription opioids injection or noninjection usea,b (≥daily vs <daily)	1.24 (.62–2.48)		2.17 (1.39–3.38)	1.93 (1.24–3.00)
Crack cocaine smokinga,b (≥daily vs <daily)	0.98 (.68–1.42)		1.01 (.71–1.43)
Cannabis usea,b (yes vs no)	1.09 (.75–1.57)		0.90 (.68–1.21)
Alcohol usea,b (≥daily vs <daily)	1.13 (.65–1.95)		1.51 (1.00–2.27)
Unstable housinga,b (yes vs no)	1.12 (.78–1.61)		1.13 (.84–1.54)
Sex worka,b (yes vs no)	0.92 (.61–1.40)		1.10 (.50–2.41)
Cumulative incarceration eventsa
1–2 times vs never	1.14 (.74–1.78)		1.39 (.95–2.02)
3–5 times vs never	0.85 (.50–1.42)		1.44 (.94–2.20)
>5 times vs never	1.16 (.59–2.30)		1.02 (.60–1.74)
Enrolled in OATa,b (yes vs no)	0.78 (.54–1.11)		0.72 (.54–.98)
Enrolled in addiction treatment other than OATa,b (yes vs no)	0.75 (.52–1.06)		0.85 (.64–1.14)
ART access and plasma HIV-1 RNA VLa,b
≥1 day dispensed ART and VL of <50 copies/mL	0.50 (.31–.82)	0.55 (.33–.91)	0.37 (.26–.54)	0.58 (.39–.88)
≥1 day dispensed ART and VL of ≥50 copies/mL	1.13 (.75–1.69)	1.23 (.81–1.87)	0.65 (.46–.93)	0.75 (.52–1.06)
0 day dispensed ART	ref	ref	ref	ref
Years since first record in the registrya, c (per year longer)	0.85 (.75–.97)	0.86 (.77–.97)	0.83 (.77–.89)	0.85 (.78–.92)
Years since HAART initiationa, d (per year longer)	0.93 (.89–.98)		0.90 (.87–.94)
HCV serostatusa (positive vs negative)	1.27 (.33–4.88)		1.55 (.65–3.72)

Abbreviations: ART, antiretroviral therapy; CI, confidence interval; HAART, highly active antiretroviral therapy; HCV, hepatitis C virus; HIV, human immunodeficiency virus; HR, hazard ratio; MMT, methadone maintenance therapy; OAT, opioid agonist therapy; PWID, people who inject drugs; VL, viral load.

^aDenotes time-varying variables.

^bDenotes activities/events in the past 6 months.

^cThe starting time point is the first date when a participant accessed HIV care according to the provincial Drug Treatment Program database.

^dThe starting time point is the first date when a participant was dispensed HAART medications according to the provincial Drug Treatment Program database.

DISCUSSION

Our results suggest that the scale-up of “seek-and-treat” HIV treatment interventions beginning in 2010 [6] have served to significantly reduce HIV-related mortality among our community-recruited cohort of 961 PWID living with HIV in this setting. As expected, in multivariable survival analyses, those who were receiving ART and maintaining their VL to undetectable levels had almost half the hazard of death compared to those who were not receiving ART; however, the mortality benefit of ART was not confirmed when VL was ≥50 copies/mL. Therefore, helping HIV-infected PWID maintain their VL at undetectable levels is an important area to direct future efforts to further reduce mortality.

The finding that HIV-related mortality has declined among both men and women is particularly encouraging, as previous studies have reported poorer access and adherence to ART and persistently high HIV-related mortality among female PWID in this setting [4, 7]. However, sex-based differences remain in that there were no reductions in overdose or other nonaccidental mortality among women, unlike among men. On the other hand, men had higher liver-related mortality rates than women. Moreover, the finding that the self-identified white ethnicity/ancestry was an independent predictor of death among men was somewhat counterintuitive in this setting where Indigenous populations have had lower access to ART until recently [8]. Future research should further examine the potential benefits of expanded access to comprehensive HIV care beyond HIV-related morbidity and mortality and identify the gap in care among PWID living with HIV.

Of note, high-intensity prescription opioid use independently predicted mortality among men. Recent US studies reported high rates of prescription opioid use among individuals living with HIV [9, 10], and long-term prescription opioid use has been shown to increase the mortality risk among this population [11]. Investigating how different patterns of prescription opioid use increase the mortality risk remains a topic for future research.

This study has several limitations. First, ACCESS participants were not randomly recruited, and therefore generalizability may be limited. Second, the self-reported data may be affected by reporting biases. However, we note that this type of data has been commonly utilized in observational studies involving PWID and has been found to be valid [12]. We also note that several measures of interest, including mortality, HIV/AIDS clinical monitoring, and ART dispensation, were acquired through comprehensive administrative databases in a setting of no-cost universal medical care. Third, as with all observational studies, the relationships between the explanatory variables and outcome assessed may be under the influence of unmeasured confounding. While we sought to address this bias with multivariable adjustment of the key predictors of survival, there may be residual confounding.

To conclude, our results demonstrate that HIV-related mortality rates have declined significantly since 2010, coincident with the scale-up of a community-wide “seek-and-treat” campaign to engage members of traditionally hard-to-treat groups in HIV treatment. High-intensity prescription opioid use predicted mortality among men, indicating a need to address factors shaping harms from prescription opioid use among this population.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Presented in part: Eighth International AIDS Society Conference on HIV Pathogenesis, Treatment and Prevention, Vancouver, Canada, 20 July 2015.

Notes

Acknowledgments. The authors thank the study participants for their contribution to the research, as well as current and past researchers and staff.

Disclaimer.  The funding agencies had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.

Financial support. This work was supported by the National Institute on Drug Abuse of the US National Institutes of Health (grant number R01DA021525); the Canada Research Chairs program through a Tier 1 Canada Research Chair in Inner City Medicine (to E. W.); the Canadian Institutes of Health Research New Investigator Award (to M. J. M. and grant number MSH-141971 to K. H.); the Michael Smith Foundation for Health Research Scholar Award (to M. J. M. and K. H.); the British Columbia Ministry of Health (grant paid to J. M.’s institution); the US National Institutes of Health (grant number R01DA036307 to J. M.’s institution); and the Canadian Institutes of Health Research Foundation (grant number FDN-148476 to T. K.).

Potential conflicts of interest. M. J. M.’s institution has received an unrestricted gift to support his research from NG Biomed, Ltd. J. M. has received limited unrestricted funding, paid to his institution, from AbbVie, Bristol-Myers Squibb, Gilead Sciences, Janssen, Merck, and ViiV Healthcare.

All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.