Illicit Drug Use and HIV-1 Disease Progression: A Longitudinal Study in the Era of Highly Active Antiretroviral Therapy

Abstract

This study assessed the association between longitudinal patterns of illicit drug use and clinical progression of human immunodeficiency virus (HIV) disease. Confidential computer-based interviews, which addressed illicit drug use and other factors, were completed by HIV-infected participants in Baltimore, Maryland, at 6-month intervals from 1998 onward. To assess this association, the authors used a random-effects model in which clinically defined opportunistic conditions were linked to self-reported periods of drug use, enabling four categories of drug use to be distinguished: nonusers, intermittent users during abstinent periods, intermittent users during active periods, and persistent users. Included in the analysis were 1,851 participants who completed ≥1 survey. For participants who used drugs intermittently over time, the risk of developing new opportunistic conditions during periods of abstinence was similar to that for those who never used drugs (odds ratio = 1.2, 95% confidence interval: 0.9, 1.7). In contrast, compared with that for nonusers, the risk of opportunistic infection was significantly higher for intermittent drug users during periods of active use (odds ratio = 2.2, 95% confidence interval: 1.4, 2.9) and for persistent drug users (odds ratio = 1.9, 95% confidence interval: 1.2, 2.8). Active drug use is temporally linked to HIV disease progression and mortality. Effectively targeting and treating active substance abuse in HIV treatment settings may provide a mechanism to improve clinical outcomes.

Illicit drug use is an important cofactor in the human immunodeficiency virus (HIV)-1 pandemic. In the United States, injection drug use was an acquisition risk factor in 31 percent of the cases of acquired immunodeficiency syndrome reported to the Centers for Disease Control and Prevention through 2002 (1), and, in eastern Europe and Russia, injection drug use is fueling the most rapidly growing epidemic in the world (2).

A history of injection drug use has been associated with delayed use of highly active antiretroviral therapy (HAART) in developed countries (3–5). In the HAART era, a history of injection drug use has been variably reported to be associated with (6, 7) and not associated with (8, 9) clinical disease progression and death in HIV-1-infected individuals. However, such studies have two limitations. First, drug use was assessed as a fixed covariate, based on HIV-1 acquisition risk behavior recorded at cohort enrollment. However, drug use is a dynamic and chronic process. For example, a recent report from a cohort of injection drug users in Baltimore, Maryland, found that approximately only 30 percent of individuals persistently used drugs during an average follow-up of 8 years; the remainder had more complex patterns of drug use (10). Second, noninjection use of heroin and cocaine is common, and individuals using drugs via other routes are misclassified when only injection use is considered.

Active drug use has been reported to be associated with nonadherence to HAART and poorer virologic response to therapy (11–13). Others have found that changes in drug use activity were temporally associated with use of and adherence to HAART and with changes in CD4 cell counts and HIV-1 RNA levels (14). The goal of our study was to assess the longitudinal association of drug use with HIV-1 clinical disease progression in a university-associated clinical cohort.

MATERIALS AND METHODS

Setting and data collection

An observational cohort was established in an urban HIV-1 clinic in 1990, and data from over 5,000 HIV-1-infected patients have been gathered (15). Trained technicians review and abstract information from clinical records onto structured data collection forms; the information is then entered into an automated database. Clinic medical records, the main hospital medical record, and various institutional automated databases (e.g., laboratory, radiology, pathology, discharge summaries) are abstracted. Comprehensive demographic, clinical, laboratory, pharmaceutical, and psychosocial data are collected when patients enroll in the clinic and at 6-month intervals thereafter. Nonexclusive risk factors for HIV-1 acquisition (history of injection drug use, men who have sex with men, known heterosexual exposure, etc.) are identified by clinicians when patients enroll in the cohort. Clinical diagnoses of opportunistic conditions associated with acquired immunodeficiency syndrome (16), other selected conditions, and dates for these events are routinely abstracted. Information on death is obtained from a death registry maintained by the clinic. In addition, death records of the Maryland Bureau of Vital Records and the national Social Security death index are searched regularly.

An ongoing series of patient surveys was initiated in the clinic in November 1998. Surveys were offered longitudinally to participants at 6-month intervals. Each day, the clinic schedule was reviewed, and patients attending appointments were approached and asked to participate in the survey if they had not participated previously or if 6 months had elapsed since their last survey. The 15-minute surveys were integrated into normal clinic wait time, participants were reimbursed for their time, and the participation rate was more than 98 percent among those approached. The surveys were administered by audio, computer-assisted self-interview. This survey method has been found to enhance the reporting of potentially stigmatizing behaviors (17, 18). Participants were informed that their responses to questions were confidential and would be unavailable to clinicians and staff in the clinic. Participants were asked in detail about their use of alcohol, marijuana, heroin, and cocaine and of other drugs in the 6 months preceding the survey by using questions adapted from an instrument developed and validated for the AIDS Link to Johns Hopkins Intravenous Experience (ALIVE) cohort study (19, 20). The University Institutional Review Board approved this study, and study participants provided written informed consent.

Definitions and outcomes

All participants who completed at least one survey were included in the analysis. The explanatory variable of interest was heroin or cocaine use (taken by any route), which we previously found to be associated with underuse of HAART, nonadherence, and poorer HIV-1 RNA and CD4 cell count responses to treatment (11, 14). At-risk alcohol use was defined as a report of 14 or more drinks per week for men or seven or more drinks per week for women (21). Peak HIV-1 RNA was defined as the highest HIV-1 RNA measured prior to the first interview, and nadir CD4 cell count was defined as the lowest CD4 count measured prior to the first interview. Antiretroviral use and adherence was categorized into three groups at each survey, according to self-report: 1) using HAART without any missed doses in the previous 2 weeks, 2) using HAART with one or more missed doses in the previous 2 weeks, and 3) and not using HAART.

The primary outcome for this analysis was the development of new HIV-1–associated opportunistic conditions (16). This outcome was selected because drug users are at higher risk of death from non-HIV-1–related causes (e.g., overdose, homicide, and suicide) than nonusers (10), whereas development of opportunistic conditions is a specific indication of HIV-1 disease progression. In the analysis, recurrent episodes of Pneumocystis jiroveci (formerly P. carinii) pneumonia, candidal esophagitis, and bacterial pneumonia were counted as separate and distinct clinical events as long as there was clinical, microbiologic, or radiologic evidence of resolution of the primary episode prior to a secondary episode. In contrast, other opportunistic conditions were coded as events at their initial manifestation only, because secondary events were thought to represent recurrence of a primary condition in most cases: cytomegalovirus retinitis, cytomegalovirus involvement of the gastrointestinal tract, toxoplasmosis, Cryptosporidium parvum diarrhea, cryptococcal meningitis, disseminated Mycobacterium avium complex, Mycobacterium tuberculosis, progressive multifocal leukoencephalopathy, non-Hodgkin's lymphoma, Kaposi's sarcoma, cervical cancer, HIV-1 dementia/myelopathy complex, and HIV-1–associated wasting (loss of >10 percent of baseline weight, without other identified etiology).

Statistical analysis

We assessed the relation between drug use and outcomes by using both patient-level and observation-level data. For the patient-level analysis, participants were categorized as nonusers, intermittent users, or persistent users according to their longitudinal pattern of heroin or cocaine use over the entire duration of their follow-up. Subjects who completed only a single survey were categorized on the basis of both self-reported drug use in the survey and their history of injection drug use, which was recorded by clinicians at enrollment in the cohort (enrollment preceded the first survey by a median of 2 years). For example, a patient who completed only a single survey in which he or she denied heroin or cocaine use in the prior 6 months, but for whom injection drug use had been recorded as an HIV-1 risk factor at cohort enrollment, was categorized as an intermittent user. Follow-up for each participant was closed at death or was censored at the date of last data abstraction or the date of last contact if lost to follow-up. Baseline differences between nonusers, intermittent drug users, and persistent drug users were assessed with the chi-square and Wilcoxon rank-sum tests for discrete and continuous variables, respectively. Time to death from the first survey was compared in the three groups with the Kaplan-Meier method. Cox proportional hazards modeling was used to adjust for baseline covariates.

We used observation-level data to assess the association between drug use and development of new opportunistic conditions, using longitudinal data analysis techniques to account for repeated observations. Participants were queried about heroin or cocaine use in the 6 months preceding each survey. The 6-month period prior to the first survey was the first observation period for all subjects. When fewer than 6 months separated two consecutive surveys, only that time segment between the surveys was considered for the observation period of the latter survey. When more than 6 months separated consecutive surveys, the observation period for the latter survey was permitted to extend up to 12 months. When consecutive surveys were separated by more than 12 months, the time segment that extended from the first survey in the pair to the date 12 months before the second survey in the pair was excluded from the analysis because it was thought that drug use activity could not be reliably inferred. Opportunistic conditions that met the criteria described above were assigned to individual observation periods based on the occurrence date. On the rare occasions when participants developed more than one opportunistic condition in a single observation period, the binary outcome was coded the same as if a single event had occurred.

For each observation for intermittent drug users, a time-dependent variable was used to distinguish whether drugs were being used actively. Doing so created four categories of drug-use activity for the observation-level analysis: 1) nonusers; 2) intermittent users, abstinent; 3) intermittent users, active; and 4) persistent users. We next fit a random-effects model, which models intrasubject associations as a latent Gaussian process that represents a particular subject's underlying propensity to develop an opportunistic condition. We found that both drug use and opportunistic infections were strongly associated with survival. Thus, survival was likely to be a confounder in the analysis because self-reported drug-use data for a given observation period could be obtained only if the study participant survived long enough to complete the survey corresponding to that time period.

To address this issue, we used a parametric joint longitudinal-survival model (20, 22) that linked the random-effects model and a parametric survival model via the latent Gaussian variable. A Weibull function fit the survival data well and was selected for the parametric survival model. Associations between variables of interest and opportunistic conditions are represented by odds ratios from the joint longitudinal-survival model, which adjusts both for other covariates included in the model and for survival differences associated with those variables.

RESULTS

A total of 1,851 participants met the inclusion criteria and were included in the analysis: 1,028 (56 percent) were categorized as nonusers, 588 (32 percent) as intermittent users, and 235 (13 percent) as persistent users (table 1). Participants contributed a total of 4,719 person-years of follow-up to the study: 2,725 (58 percent) by nonusers, 1,497 (32 percent) by intermittent users, and 497 (10 percent) by persistent users. There were no significant differences among the groups according to sex or nadir CD4 cell count. Peak HIV-1 RNA levels were slightly lower in the intermittent users than in the nonusers. As anticipated from the categorization scheme, there were significant differences among the groups regarding HIV-1 acquisition risk factors. At-risk alcohol use was significantly more frequent in persistent and intermittent drug users than in nonusers. Compared with nonusers, intermittent users and persistent users were slightly older, had been diagnosed with HIV-1 for slightly longer, were less likely to be White, had lower CD4 cell counts at the first survey, had higher HIV-1 RNA levels at the first survey, and were less likely to have used HAART prior to the first survey. Additionally, persistent users were significantly less likely than intermittent users to have used HAART prior to the first survey. Persistent drug users completed significantly fewer surveys, had a later first interview date, and had less follow-up time than either nonusers or intermittent users.

A total of 5,486 surveys, or individual observations, were included in the longitudinal analysis of drug use and development of opportunistic conditions: 3,099 (56 percent) for nonusers, 1,093 (20 percent) for intermittent users during abstinent periods, 789 (14 percent) for intermittent users during periods of active use, and 508 (9 percent) for persistent users. Of the 1,851 participants who completed surveys, 645 (35 percent) completed one, 356 (19 percent) completed two, 246 (13 percent) completed three, 192 (10 percent) completed four, 130 (7 percent) completed five, 106 (6 percent) completed six, 94 (5 percent) completed seven, 54 (3 percent) completed eight, 24 (1 percent) completed nine, and three (<1 percent) completed 10. Persistent users reported using drugs a median of 27 days in the prior 6 months (interquartile range, 6–150), and intermittent users (when active) reported using drugs a median of 14 days in the prior 6 months (interquartile range, 3–90) (p < 0.001).

There were 278 deaths during follow-up (15 percent): 118 (11 percent) among nonusers, 109 (19 percent) among intermittent users, and 51 (22 percent) among persistent users. Time to death was assessed in the three groups with a Kaplan-Meier function (figure 1). The estimated survival rates at 36 months were 87 percent for nonusers, 80 percent for intermittent users (log-rank test p < 0.0001), and 68 percent for persistent users (log-rank test p < 0.0001 compared with nonusers, and p = 0.015 compared with intermittent users). Compared with that for nonusers, the risk of death was higher for intermittent users (hazard ratio = 1.9, 95 percent confidence interval (CI): 1.4, 2.4) and for persistent users (hazard ratio = 2.9, 95 percent CI: 2.1, 4.1), adjusting for age, race, sex, nadir CD4 cell count, and peak HIV-1 RNA level in a Cox proportional hazards model.

Of 5,486 discrete observations for study participants, a new opportunistic condition was recorded in 537 (9 percent): 9 percent of surveys of nonusers, 9 percent of surveys of intermittent users during abstinent periods, 13 percent of surveys of intermittent users during active periods, and 13 percent of surveys of persistent users (table 2). Table 3 shows bivariate and multivariate associations between selected factors and development of new opportunistic conditions when the joint longitudinal-survival modeling technique described above was used. During periods of abstinence, intermittent drug users had a risk of developing new opportunistic conditions that was similar to that for nonusers (odds ratio (OR) = 1.4, 95 percent CI: 1.0, 1.9). In contrast, compared with that for nonusers, the risk of opportunistic infection was statistically significantly higher for intermittent drug users during periods of active drug use (OR = 2.3, 95 percent CI: 1.5, 3.0) and for persistent drug users (OR = 2.1, 95 percent CI: 1.4, 3.1). In the multivariate model, higher peak HIV-1 RNA and lower nadir CD4 cell counts were significantly associated with clinical disease progression, whereas sex, race, and at-risk alcohol use were not.

To assess the effect of a potential mediating factor on the association between drug use and disease progression, we added the antiretroviral use and adherence time-dependent variable to the final adjusted model. Compared with those not using HAART, those using HAART and who missed one or more doses in the prior 2 weeks (OR = 0.8, 95 percent CI: 0.6, 0.1), and those using HAART and who missed no doses (OR = 0.7, 95 percent CI: 0.6, 1.0), were at decreased risk of disease progression. After adjustment for HAART use and adherence, the associations between drug use and HIV-1 disease progression changed only slightly. Compared with that for nonusers, the risk of disease progression was similar for intermittent drug users during abstinent periods (OR = 1.3, 95 percent CI: 0.9, 1.8) and was higher for intermittent users during active periods (OR = 1.6, 95 percent CI: 1.2, 2.3) and for persistent users (OR = 1.9, 95 percent CI: 1.2, 2.8). No statistically significant interactions were detected among the covariates. Similar results were obtained when information for participants who completed fewer than two surveys (n = 1,205; 35 percent) was excluded from the analysis (data not shown).

To address whether the frequency of drug use was associated with HIV-1 disease progression in users, we conducted a supplementary analysis in which self-reported drug use at each survey was categorized as none, use for 20 or fewer days in the past 6 months, or use for more than 20 days in the past 6 months (20 days was the median reported frequency of use among active users). We constructed a random-effects model to assess the risk of opportunistic infection according to the frequency of self-reported drug use. Compared with nonusers of drugs, those who reported drug use for 20 or fewer days in the prior 6 months were at increased risk of opportunistic infection (OR = 1.8, 95 percent CI: 1.3, 2.4), adjusting for sex, race, age, nadir CD4 count, and peak HIV RNA level. This risk was similar in magnitude to that experienced by those reporting use for more than 20 days (OR = 1.9, 95 percent CI: 1.3, 2.6).

DISCUSSION

Injection drug use is a major risk factor for HIV-1 acquisition, and substance abuse disorders are highly prevalent in HIV-1–infected individuals in the United States (23). Previous reports from our cohort demonstrated that active drug use was associated with underuse of and nonadherence with HAART (11) and that transitions in drug use, either achieving abstinence or relapsing, during follow-up were temporally associated with concordant changes in viral load and CD4 cell count, indicators of treatment effectiveness (14). In the current study, we found that, compared with nonusers, intermittent drug users had a twofold increased risk of death and persistent drug users had a threefold increased risk of death. However, active heroin use and cocaine use confer an increased risk of death irrespective of HIV-1 infection (10), and our primary interest was to assess the association between drug use and HIV-1 disease progression. In an adjusted analysis, we found that persistent drug use, compared with nonuse, was associated with a twofold risk of developing new opportunistic conditions. Remarkably, during periods of abstinence, intermittent drug users had a risk of HIV-1 disease progression similar to that for nonusers. However, during periods of active use, intermittent users had a risk of disease progression that was similar to that for persistent users.

Two hypotheses may explain the association we observed between drug use and HIV-1 disease progression. First, drug use may have direct or indirect biologic effects that hasten the progression of HIV-1 infection. Studies have found opioid exposure in vitro to be associated with increased replication of HIV-1 (24) and impaired lymphocyte function (25). However, in pre-HAART natural history cohorts, HIV-infected injection drug users were not found to experience more rapid lymphocyte decline or progression to acquired immunodeficiency syndrome than HIV-infected men who have sex with men (26, 27). Hepatitis C virus coinfection, which is highly prevalent in HIV-infected injection drug users, has been reported by one group to be associated with an impaired CD4 cell response to HAART compared with individuals not infected with hepatitis C virus, despite a similar HIV-1 RNA response (28). However, another group found no evidence that hepatitis C coinfection was associated with immunologic response to HAART or with HIV-1 disease progression (29).

The second hypothesis is that the association between active drug use and HIV-1 disease progression is mediated primarily through detrimental effects of the former on access to care and adherence to therapy. In support of this hypothesis, Moore et al. (30) reported that HIV-1–infected drug users had a risk of disease progression similar to that for non–drug users in the pre-HAART era but experienced disease progression at a significantly higher rate than non–drug users did during the HAART era. Relative to other HIV-1 acquisition risk groups, injection drug users have poorer access to care (31, 32) and underuse HAART (3–5, 32, 33).

In several studies (12, 34–36), drug and alcohol abuse have been linked to poor adherence to HAART. For example, Arnsten et al. (12) measured adherence to HAART in 85 HIV-infected former and active drug users by using electronic bottle-cap recorders, and they found that average adherence was 68 percent among those not using cocaine compared with just 27 percent among those using cocaine. A constellation of factors is likely to contribute to the association between drug use and use of HIV care (37), including subjugation of health concerns to the needs of addiction, competing medical conditions (38), unstable living conditions, alienation from family and social support structures, mistrust of the medical system (39), and clinician discomfort with treating drug abusers (40).

In our study, we attempted to adjust for the potential confounding effects of HAART use and adherence by adding a time-dependent variable to our multivariate model. The estimates for the associations between drug use and HIV-1 disease progression changed only slightly in this model. This finding may support the hypothesis of a direct biologic effect of drug use on disease progression. However, self-reported adherence is a relatively poor surrogate of actual adherence; therefore, substantial residual confounding may remain.

Notably, most of the studies assessing disease progression or mortality have categorized drug users on the basis of HIV-1 risk behaviors identified at enrollment to the cohort and not according to drug-use activity over time (7, 30, 41), which is likely to underestimate the impact of ongoing drug use on HIV-1 treatment outcomes and disease progression. In our study, we found that persistent drug users were at significantly increased risk of developing new opportunistic conditions and had the highest risk of death, while nonusers had the lowest risk of death. In contrast, the risk of intermittent drug users developing new opportunistic conditions was temporally tied to drug-use activity, and, as a group, their mortality rate was midway between that of persistent drug users and that of non–drug users. Drug abuse is a dynamic behavior. A recent report from a longitudinal cohort of injection drug users in Baltimore, in which participants were followed up for an average of nearly 8 years, identified three broad patterns of drug use: persistent use in approximately 30 percent, long-term cessation in approximately 20 percent, and more complicated relapsing and remitting patterns in the remaining 50 percent (10).

Our study has several limitations. First, it was conducted in a single university-associated HIV-1 clinic, and our findings may not generalize to centers where demographic and behavioral factors are dissimilar to those of our cohort. Data were collected from a convenience sample, which may be a biased subset of the clinic population if individuals who were more ill were more likely to attend clinic appointments than those who felt well. Second, our assessment of drug-use activity relied on participant self-report. To maximize participant candor, surveys were conducted by using an audio, computer-assisted self-interview, a technique that has been found to increase reporting of potentially stigmatizing behaviors (17, 18). Additionally, subjects were informed that interview procedures were separate from clinical care and that their responses to interview items would not be available to clinicians, nurses, or other members of the clinical care team. HIV-1 disease progression outcomes were abstracted from clinical records in a prospective data collection process that was separate from participant interviews. Of note, the misclassification that would occur if individuals who were actively using drugs tended to minimize or deny drug use would be expected to bias the observed association between drug use and HIV-1 disease progression toward the null. Thus, the magnitude of association between drug use and disease progression that we observed is likely to be an underestimate.

Third, depression, which has been associated with adherence and HIV treatment outcomes (42), was not assessed in participant surveys. Fourth, reverse causality, where subjects relapsed to drug use because they had become ill with opportunistic conditions, cannot be excluded in our study design. Fifth, cause of death was not available in our cohort database, which precluded analyzing mortality according to causes related and unrelated to HIV-1.

The results of this study and earlier reports from our cohort (11, 14, 31) have implications for the treatment of HIV-1–infected drug users. Periods of active drug use in HIV-1–infected individuals pose dual risks. First, persistent or relapsed drug use directly affects adherence to both HAART (12) and clinic appointments (31), which contributes to the clinical disease progression and higher mortality that we observed in this study. Second, active drug use is an efficient vehicle for HIV-1 transmission to other individuals, through shared injecting paraphernalia and other associated behaviors (43). The risk of transmitting drug-resistant HIV-1 strains may be greater when relapsed drug use in HAART-treated individuals is accompanied by waning adherence and a rebounding viral load. The Centers for Disease Control and Prevention has advocated that clinicians who treat HIV-1–infected individuals identify and address transmission risk behaviors in their patients as a routine part of clinical care (44).

HIV-1 infection and drug abuse are chronic conditions that interact with one another over time. Treatment paradigms that integrate HIV-1 treatment, substance abuse treatment, and psychiatric care should be emphasized (45, 46). In the case of opioid dependence, highly effective pharmacologic treatment modalities exist (47, 48), and strategies that link such treatment with HIV-1 primary care should be strongly encouraged (49). Recent legislation and the US Food and Drug Administration's approval of sublingual buprenorphine for the management of opioid dependence in primary-care settings (47) are particularly apt for HIV-1 treatment clinics.

In conclusion, compared with mortality in HIV-1-infected non–drug users in this longitudinal study, mortality was approximately twofold higher in intermittent drug users and threefold higher in persistent users. The risk of developing new opportunistic conditions was significantly higher for persistent users than for nonusers. For intermittent users, periods of active drug use posed a risk of HIV-1 disease progression that was similar to that for persistent users, while abstinent periods were associated with a risk of disease progression similar to that for non–drug users. Effectively targeting and treating active drug use may lead to improved HIV-1 clinical outcomes.

This work was supported by the National Institute on Drug Abuse (grants DA15616, DA11602, and DA00432).

Dr. Chaisson is a consultant to Bristol-Myers Squibb Company (New York, New York) and Tibotec, Inc. (Yardley, Pennsylvania).

References