|Home | About | Journals | Submit | Contact Us | Français|
Drug overdose is a common cause of non-AIDS death among people with HIV and the leading cause of death for people who inject drugs. People with HIV are often exposed to opioid medications during their HIV care experience; others may continue to use illicit opioids despite their disease status. In either situation, there may be a heightened risk for nonfatal or fatal overdose. The potential mechanisms for this elevated risk remain controversial. We systematically reviewed the literature on the HIV–overdose association, meta-analyzed results, and investigated sources of heterogeneity, including study characteristics related to hypothesize biological, behavioral, and structural mechanisms of the association. Forty-six studies were reviewed, 24 of which measured HIV status serologically and provided data quantifying an association. Meta-analysis results showed that HIV seropositivity was associated with an increased risk of overdose mortality (pooled risk ratio 1.74, 95% confidence interval 1.45, 2.09), although the effect was heterogeneous (Q = 80.3, P <0.01, I2 = 71%). The wide variability in study designs and aims limited our ability to detect potentially important sources of heterogeneity. Causal mechanisms considered in the literature focused primarily on biological and behavioral factors, although evidence suggests structural or environmental factors may help explain the greater risk of overdose among HIV-infected drug users. Gaps in the literature for future research and prevention efforts as well as recommendations that follow from these findings are discussed.
Drug overdose is a common cause of non-AIDS death among people with HIV and the leading cause of death for people who inject drugs [1–6]. People with HIV are often exposed to opioid medications during their HIV care experience; others may continue to use illicit opioids despite their disease status. Either scenario may present a heightened risk for fatal and nonfatal opioid overdose. Recently, both the US President’s Emergency Plan for AIDS Relief and the Global Fund to Fight AIDS, Tuberculosis and Malaria (http://www.pepfar.gov/) issued guidance that they will support overdose prevention activities, acknowledging the growing toll that overdose takes on people with HIV [7,8].
Opiates and synthetic and semisynthetic opioid drugs (hereafter referred to collectively as opioids) are important medications for the treatment of pain and end-stage disease. However, this class of drugs also has the potential for the development of physical dependence, abuse, and addiction. Opioids include oxycodone, hydrocodone, hydromorphone, fentanyl, morphine, and methadone. Indeed, heroin, an illegal opiate, is one of the most common drugs of abuse worldwide. The WHO reports that approximately 13.5 million people use opioids, including 9.2 million heroin users . Many opioid users have a history of injection; similarly, many injectors report past opioid use. The global number of IDUs is estimated at 11–22 million people .
Injection drug use represents a major route of infection for HIV, accounting for up to 80% of new HIV infections in eastern Europe and central Asia . Additionally, HIV-infected IDUs can act as a bridge population for transmission of HIV to the larger population. Injection-related HIV-epidemics exist in countries throughout the world, with incidence ranging from 1.5% in Australia and New Zealand to 17% in east/south-east Asia, 27% in eastern Europe, and 29% in Latin America .
Although the association between HIV infection and injection drug use has been well documented, the potential association between HIV and overdose has received less attention. In addition to the high rates of opioid mortality among IDUs [2,12,13], nonfatal overdose events are also common. Lifetime nonfatal overdose rates among heroin users range from 29 to 68% [14–17], and among illicit users of prescription opioids in the USA ranges from 28  to 38% . Nonfatal overdose is associated with significant morbidity, including pneumonia, renal failure, mental impairment, and cardiac arrhythmia . The strongest predictor of experiencing an overdose is prior overdose [20,21], with most users reporting multiple overdose experiences in their lifetime [18,21,22].
Over the past two decades, investigators have considered a biological connection between HIV seropositivity and an elevated risk of overdose, whereas others dismiss detected associations as uncontrolled residual confounding . Proposed biological mechanisms include underlying disorder (due to HIV infection) [3,24], abnormal liver function or pulmonary problems [13,25], poor physical health , medical complications from injecting , and reduction in CD4+ cell counts . Additionally, some authors suggest a behavioral connection between HIV seropositivity and an elevated riskof overdose, namely drug use and risk-taking behaviors [1,25,29,30]. There may also be other important structural and environmental influences that have not been thoroughly considered.
Recent research pertaining to HIV [31–34], drug use [34,35], and overdose [21,36] describe elements of the ‘risk environment’ that can affect these health outcomes. For example, an opioid overdose is more likely to be fatal if the user is alone rather than in the presence of others. Injury epidemiology draws upon the Haddon matrix  to guide risk factor identification, prevention, and control that relate not only to host (patient), agent (the opioid), and environment (setting/context) factors but also to the timing of the injury: pre-event, during event, and postevent. Although few investigators have critically examined the shared biological, behavioral, and structural risks associated with HIV infection and overdose, a comprehensive understanding of this nexus could hold enormous promise for public health interventions and could also provide potentially lifesaving information to healthcare professionals, who prescribe opioid medications to patients with HIV or who interact with HIV-infected opioid users.
This study aimed to systematically review the literature on the putative association of HIV infection with overdose, meta-analyze results, explore sources of heterogeneity, and investigate biological, behavioral, and structural causal mechanisms of association.
We performed a search of the PubMed database, conducted from July to September 2010, using combinations of the terms: HIV, drug, overdose, heroin, naloxone, narcan, buprenorphine, methadone, opiate/opioid, fatal, fatality, death, mortality, morbidity, and nonfatal. We identified 251 unique articles that fitted our search criteria. All abstracts were read to identify potentially suitable manuscripts; 69 were identified as possibly relevant and were read completely. Of these, 23 were identified as irrelevant, and data were extracted from the remaining 46 studies using a standardized format. Inclusion criteria for articles were as follows: those presenting original research, being written in English, and being published in a peer-reviewed journal. There were no limitations on publication date or geography.
Extracted data were entered in a database with 28 distinct fields corresponding broadly to study details [design, period, population, HIV status (serological evidence, self-report)]; exposure and outcome definition and measurement; quantification of an association between HIV infection and overdose (or lack thereof); discussed causal mechanisms (biological, behavioral, environmental, none); and study limitations. Double data extraction was performed, with consensus on the extracted content reached by discussion, and final review of material by the first author.
To be eligible for the meta-analysis, studies must have reported measures of association [odds ratio (OR), relative risk (RR), hazard ratio] or sufficient data to calculate an unadjusted RR (i.e., overdose incidence by HIV status in a cohort design; OR and prevalence of overdose among HIV-uninfected in a cross-sectional or case–control design) with upper and lower confidence interval (CI) limits or standard errors. Adjusted effect measures were used in the analysis when included in the studies. Studies were pooled for analysis; ORs were transformed into RRs following standard methods [38–40].
When detected, we explored sources of variability extracted from the systematic review, including crude vs. adjusted analysis; cross-sectional vs. cohort design; US study location vs. outside of the USA; clinic-based population (yes/no); IDU only vs. mixed/unknown IDU status population; pre-HAART, post-HAART, and both pre-HAART and post-HAART era; year of study; calculated vs. reported measure of association; and fatal vs. nonfatal/both overdose outcomes. The Q-statistic and I2-statistic evaluated heterogeneity of effects . Categorical data on potential sources of heterogeneity were assessed with Q-statistics comparing categories; continuous data were assessed using mixed-effects meta-regression. We used Comprehensive Meta-Analysis version 2 software (Biostat, Englewood, New Jersey, USA, 2005)  to generate summary estimates of the effects of HIV infection on overdose risk. Random effects models were employed to more accurately account for differences in sampling, methods, and aims among the reviewed studies . To assess for publication bias, we checked for asymmetry in funnel plots of effects against standard errors and applied Egger’s test . We first present the meta-analysis findings, and then summarize causal mechanisms of the HIV–overdose association from the systematic review.
Of the 46 studies reviewed, six were case series or chart reviews, seven were cross sectional, and 33 were prospective or retrospective cohort studies. Studies spanned 15 countries, were published from 1988 to 2010, and reported data collected retrospectively or prospectively over a 29-year period (1977–2006). Table 1 [1–4,13,23–30,42–74] details reviewed studies.
A total of 27 studies were eligible for inclusion in the meta-analysis [1,2,13,25,29,30,42–60,75]. Of these, 22 were prospective or retrospective cohort designs, five were cross-sectional designs; 17 were conducted outside of the USA; and 16 used clinic-based samples. Participants were mostly IDUs (n = 16), or a mixture of participants of injecting and noninjecting or of unknown injecting status (n = 11). Overdose outcomes tended to pertain to fatal (n = 21) rather than nonfatal (n = 6) events. All but three studies [26,44,52] employed HIV serological testing to verify HIV status. Examination of funnel plots of effects showed no evidence of publication bias (data not shown); an Egger’s test was not significant (t = 1.07, P = 0.29).
The pooled RR across the 27 studies was 1.60 (CI 1.16, 2.21); the Q-statistic was statistically significant and the I2-statistic was 94%. Figure 1a reports the forest plot. Covariates tested for sources of variation showed that only one factor was significant at the P value less than 0.10 level: study population type (IDU only vs. mixed or unknown IDU status).
Three large studies did not collect HIV status biologically [26,44,52] and may have fundamentally underestimated the HIV–overdose association by ‘missing’ a large number of truly HIV-infected people who may not have known their status or underreported their HIV infection status. Pooling effects across the 24 studies reporting HIV status biologically returned an overall RR of 1.74 (CI 1.45, 2.09); the Q-statistic was statistically significant (Q = 80.3, P <0.01) and the I2-statistic was 71%. Figure 1b reports the forest plot. Testing for heterogeneity showed a statistically significant source was study population type (Q = 4.57, P = 0.03). Studies with IDU only population types (n = 14) had pooled RR of 1.48 (CI 1.17, 1.86), but heterogeneity remained high (Q-statistic significant, I2 = 73%). Studies with a mixed (IDU, non-IDU) or unknown IDU status population (n = 10) had pooled RR of 2.18 (CI 1.66, 2.87); the Q-statistic was not statistically significant and the I2-statistic was 40%. A trend indicated study design as an additional source of heterogeneity across the 24 studies (Q = 3.71, P = 0.05) in which cross-sectional designs (n = 4) had a pooled RR of 2.41 (CI 1.65, 3.51); the Q-statistic was not significant and the I2-statistic was 44%. Cohort studies (n = 20) had a pooled RR of 1.59 (CI 1.32, 1.92); the Q-statistic was significant and the I2-statistic was 65%. Other potential sources of heterogeneity were not significant by Q-statistic or meta-regression.
Proposed causal mechanisms for observed associations between HIV infection and overdose risk encompassed biological and behavioral factors and considered HIVas a potential confounder as well as a mediator of the association. Still, some authors dismissed a detected association as speculative. Biological explanations considered how aspects of HIV itself could elevate this risk through clinical status, immunosuppression, opportunistic infections, and poorer physical health [1,24–26,30, 46,59]. Some authors named more specific mechanisms of action, focusing on pulmonary conditions or infections more common in people with HIV that are likely to exacerbate the respiratory depression that causes death from overdose [13,25]. Several studies posited that conditions that affect the body’s ability to metabolize [e.g. liver disorder, co-infection with hepatitis B virus or hepatitis C virus (HCV)], which are also more common in people with HIV, could explain the observed association between HIV and overdose [13,24,25,46,58]. Wang et al.  systematically tested the mediating impact of some of these biological factors on the observed HIV–overdose association, determining that HIV immunosuppression and associated multisystem disorder (respiratory/pulmonary) accounted for a 20% reduction in effect. Abnormal liver functioning reduced the association by 12–35% .
Few studies examined patient HIV treatment status or HIV treatment adherence, although treatment was raised as a mediator of overdose risk. Assessing this association was impossible because investigators used population-level mortality data that indicated HIV status, but not treatment history. Oftentimes, even if HIV treatment status and adherence were assessed, direct comparisons between these variables and rates of overdose were not considered. In some instances, CD4 cell counts, which may serve as a proxy for HIV disease state, were assessed. Among this limited number of studies, findings varied, with some suggesting greater risk [24,61,76] and others no association [28,50] between lower CD4 cell counts and heightened overdose risk.
Behavioral factors that could explain the observed association of HIV infection and overdose risk were also raised. Authors discussed how high-risk lifestyle and psychiatric comorbidities could simultaneously explain overdose events and HIV status [1,26,29,30,48,59,62]. Although several studies raised whether HIV-infected drug users have more suicidal tendencies and, therefore, may engage in riskier drug use [30,58], research has not found such evidence . Indeed, residual confounding from high-risk behaviors of HIV-infected drug users, frequent injectors, shooting gallery use, syringe sharing, or other ‘risk-taking personality’ traits appeared to produce small (15%) reductions in observed HIV–overdose effects . Other authors suggest that drug users tend to reduce injecting behaviors after HIV diagnosis and also as HIV infection advances . One study  found that people who knew their HIV status were more likely to reduce opioid consumption, rather than engage in other overdose preventive behaviors.
Seven studies measured structural and environmental risks for overdose. Factors considered in these studies included access to medication-assisted therapy (MAT) for opioid dependence, homelessness, neighborhood poverty and socioeconomic status, incarceration, and isolation or using drugs alone. Of the studies reviewed, several found that access to and enrollment in methadone treatment greatly reduced HIV-infected IDUs’ risk of overdose [42,51,57]. Research has shown homeless and poor drug users to be at increased risk of overdose and that receiving government welfare payments were associated with lower overdose risk [43,45,57]. Tardiff et al.  found that high neighborhood poverty was associated with an increased risk of being HIV-positive, and that opioid overdose was more likely to occur among HIV-positive than HIV-negative decedents in New York City. Prison time, including lifetime incarceration and recent prison release, was another significant structural contributor to overdose risk [4,43]. Seaman et al.  observed that the risk of death from an overdose was eight times higher within 2 weeks of being released in an HIV-infected IDU cohort . A case series of HIV-infected hospital patients, who died a non-AIDS death, found that 38 of 64 deaths were drug-involved intoxications (59%), 15 of which (39%) had recently been released from prison . Surprisingly, the authors do not remark on history of incarceration among the decedents, although a contemporaneous analysis on the same population mentions alterations to tolerance due to abstinence, such as recent return from prison, as an explanation for overdose risk in the predominantly injection-transmitted HIV cohort . An intriguing study by Neira-León et al.  showed that HIV-infected IDUs were no more likely than HIV-uninfected IDUs to use drugs alone, and that this environmental overdose risk factor was responsive to intervention, either medical care or overdose prevention education.
This article is the first to systematically review and meta-analyze the literature on the putative association between HIV infection and overdose risk. Among 46 studies extracted and reviewed in this undertaking, 24 reported data sufficient for inclusion in a meta-analysis and tested for HIV-infection biologically. Despite the heterogeneous pool of studies, the meta-analysis results suggest that people who use drugs have a 74% greater risk of overdose if they are HIV-infected compared with their counterparts who are not HIV-infected.
Causal mechanisms discussed in the literature to explain the increased risk tended to consider biological and behavioral factors, but other factors may also influence this association, including environmental and structural factors. Only seven studies (15%) systematically reviewed and considered environmental or structural factors in the analysis of overdose risk factors. Data from these studies suggest that environmental and structural risk factors shown previously to increase overdose risk also affect HIV-infected populations and, to the extent that they are more pronounced in people with HIV, could help explain some of the higher overdose risk associated with HIV status. The limited number of studies indicates a need for future research to consider the role these factors may play in overdose risk for HIV-infected and uninfected persons.
Similarities in risk and protective factors for HIV and overdose suggest an opportunity to reduce HIV transmission and overdose mortality by scaling-up established prevention interventions and extending existing legal and policy tools, most notably HAART, MAT, and prescribed naloxone.
Too few studies examined HIV treatment status or adherence to quantify its mediation of overdose risk. Wang et al.  provided the most compelling evidence of a biological component, particularly immunosuppression and multisystem disorder, to the HIV–overdose association. It is, therefore, biologically plausible that HAART’s benefits for HIV-infected patients could also extend to protection against overdose for people who use drugs. Yet, even when HIV treatment is received, potential medication interactions with continued street drug use may contribute to persistent overdose risk. For instance, several antiretroviral medications are known to increase or decrease methadone and other street drug blood levels or have known interactions with benzodiazepines and marijuana [77,78], both of which may be used therapeutically and abused. As people with a substance use disorder are often excluded from randomized clinical trials, it may be challenging to anticipate many of the side-effects and consequences of new antiretroviral medications in this population. Several published reviews offer basic clinical guidance on these topics [77–80]. Further research is needed to better understand specific HIV medication interactions and how to reduce risk of overdose in substance using patients.
Access to HAART medications, prescribed by providers prepared to prevent and manage potential interactions between antiretrovirals and drugs with abuse potential, may be viewed as a protective factor against overdose for HIV-infected drug users. Efforts to ensure that all HIV-infected drug users have adequate and stable access to HAART have the potential to reduce mortality due to AIDS and to overdose.
Numerous studies including several examined in this systematic review [30,51,57] have established receipt of MAT, particularly methadone and buprenorphine therapies, as protective against fatal overdose. MAT reduces illicit drug use, decreases HIV risk behavior, and decreases drug-associated crime [81–86]. Access to adequate MAT also increases compliance for antiretroviral therapy among HIV-infected patients [87–90]. On a community level, providing both MAT and HAART can reduce the transmission of HIV from opioid users to others, helping to reduce the overall incidence and prevalence of HIV [81,85,86,90,91]. The Joint United Nations Programme on HIV/AIDS estimates that there are only eight people receiving MAT for every 100 people who inject drugs globally . We note that studies have found increased overdose risk immediately following dropout or cessation across a variety of treatment types, although the risk is lowest with MAT , so relapse and overdose prevention are important while in treatment.
One public health intervention to decrease overdose-associated morbidity and mortality is the distribution of naloxone to at-risk drug users, their families, and friends. Naloxone is a prescription medication with no abuse potential that reverses an opioid overdose and is part of the standard emergency medical response to an opioid overdose [94,95]. Such programs train people in overdose prevention, response, and naloxone administration. Between 1996 and 2010, US programs, at more than 155 sites located in 16 states, have documented over 10 000 overdose reversals with naloxone by over 50 000 trained bystanders [96–101], with mounting evidence linking program enrollment to reductions in community opioid overdose mortality [97,102,103].
Targeting naloxone provision to HIV-infected opioid users has the potential to reduce fatal overdose in this population. Physicians who care for HIV-infected opioid users could prescribe naloxone to at-risk patients, including patients who are opioid injectors, who illicitly use prescription opioids, who are prescribed long-acting opioids, or who are diagnosed with other conditions or illnesses known to increase overdose risk. Programs that provide care and support to HIV-infected people could also distribute naloxone, given sufficient medical oversight of the prescribing practices. Syringe exchange programs are a common mechanism for distributing naloxone and, in places with high HIV prevalence among injectors, provide a means of reaching people at high overdose risk. Countries (or states) with formularies for HIV-related medications could consider adding naloxone to these lists; it is already on WHO’s Model List of Essential Medicines .
We detected important methodological limitations in the reviewed studies, which warrant mention. First, studies varied immensely in their definition of ‘overdose’, more often failing to define it. In some cases, intentionality was not addressed, which may have increased outcome measurement variability. Fatal drug overdose as cause of death was sometimes defined by a committee of study investigators; others employed standard disease classification . Nonfatal overdose may have been ascertained by self-report – typically without defining or describing symptoms of overdose – or by administrative record of the victim’s admittance to a hospital or response by emergency medical services. Both approaches may be severely flawed. Self-reported overdose may be subject to response bias: if HIV-infected persons are more likely to report nonfatal overdoses because of heightened awareness of their health status, associations could be biased away from the null. Furthermore, for studies conducted in places where police also attend emergency calls and may arrest those at the injury scene on drug-related charges, relying upon administratively defined nonfatal overdoses may severely underestimate the total number of events and introduce selection bias if certain types of drug users are less likely to call for help. It is also likely that nondifferential error in self-reported overdose occurred, but would likely have resulted in a bias toward the null hypothesis for these studies. Another challenge to this review was verification of drug treatment status and HIV adherence, as they were poorly cataloged. Employing more systematic definitions of overdose, expanding the use of computer-assisted technology to improve validity of responses, better tracking treatment attendance and exposure; also, cataloging disease progression and adherence to HIV medication regimens would greatly facilitate comparisons across studies for future meta-analyses and could better inform readers of these important details.
This study has several strengths. First, we comprehensively searched the available literature, including a broad range of study types, populations, and outcomes, and pooled available evidence to investigate and quantify the association of HIV with overdose risk. Second, this study applied theories relevant to the exposure and outcome from infectious disease epidemiology and injury epidemiology. In doing so, our review extends previous discussions in single studies that limited the putative causal association between HIV and overdose risk to only biological factors. As with many findings from the HIV field [106–108], considering the effects of extra-medical interventions and influences can be revealing, clinically significant, and can point the way toward approaches that may have larger public health impact.
This study has some potential limitations that suggest avenues of future research. First, the meta-analyzed results reflected a high degree of variability across studies, but we were unable to determine many statistically significant sources of heterogeneity from the tested covariates. The relatively small number of studies, especially those providing covariates such as MAT, HCV, and HIV treatment status and adherence, limited our ability to test heterogeneity sources. To accommodate heterogeneity levels, we report pooled effects from random effects models, as recommended . Second, we were unable to adequately review studies on structural and environmental risk factors that may differentially augment the risk of or constitute causal mechanisms of overdose in people with HIV. The topic of structural risk is an active area of inquiry in overdose [21,36], other injury health outcomes such as motor vehicle accidents [109,110] as well as in HIV transmission [31,33]. A third limitation was the sparse literature on behavioral factors, especially one’s social network composition, size, and supportiveness that may exert causal influences on overdose risk that differ by HIV status. Fourth, few reviewed articles were identified in resource-poor countries, or in locations with current injection-driven epidemics. To address all of these significant limitations, there is a clear indication for future research. Longitudinal studies of health outcomes for participants with HIV should consider including nonfatal overdose as a sentinel health event and exploring fatal overdose as a primary outcome, especially for observational and treatment intervention studies with people using opioids. In places where opioids are increasingly being prescribed to treat chronic pain (e.g. USA, Canada), the population at risk, and that should be considered in longitudinal studies of overdose, may extend beyond those with a history of substance use disorder or infected by injection drug use. Finally, the reviewed articles included those published only in English and studies conducted in humans. Although English-language articles represent the majority of research published on this subject, roughly 10% of the 317 database search results were not in English, making it possible that we overlooked important non-English language contributions.
This systematic review and meta-analysis found evidence of a positive association between HIV status and overdose risk. There may be biological, behavioral, and structural mechanisms influencing the greater risk of overdose for people with HIV. Future research to explore these mechanisms is indicated. Expanded access to existing effective interventions to reduce overdose risk is warranted for drug users infected with or at risk of HIV. Healthcare providers who treat HIV-infected patients with a history of substance use disorder and/or who prescribe opioid medications should consider counseling patients on how to reduce their risk of overdose. Healthcare providers may also consider prescribing naloxone to patients, discussing the option of initiating buprenorphine therapy, or offering a referral to another MAT, as appropriate.
T.C.G. conceived of the study, reviewed articles, performed analysis, and participated in manuscript writing. S.K.M. and M.A.Y. reviewed articles, assisted with analysis, and participated in manuscript writing. E.R.P. performed the meta-analysis and participated in manuscript writing. J.D.R. provided feedback on the study approach, interpretation of data, and reviewed manuscript drafts.
The authors are grateful to Nickolas Zaller, Sarah Bowman, Roxanne Saucier, Nabarun Dasgupta, and R. Douglas Bruce for their thoughtful comments and edits to earlier drafts of this manuscript.
This work was supported in part by a grant from the Centers for Disease Control and Prevention (CDC) (R21CE001846-01 to T.C.G.); award number K24DA022112 from the National Institutes of Health, National Institute on Drug Abuse (NIH/NIDA); and by grant number P30-AI-42853 from the National Institutes of Health, Center for AIDS Research (NIH/CFAR).
This work was also made possible in part through the support of training grant number 5T32DA013911 from the National Institutes of Health, National Institute on Drug Abuse (NIH/NIDA).
The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of NIH, NIDA, CDC, or CFAR.
Conflicts of interest
The authors declare no conflicts of interest.