Depressive symptoms at HIV testing and two-year all-cause mortality among men who inject drugs in Vietnam

Abstract

People who inject drugs (PWID) with HIV experience an elevated risk of death. A potentially important determinant of survival is the high burden of depression. This study examined the relationship of depressive symptoms at HIV testing with two-year all-cause mortality among newly diagnosed HIV-positive PWID in Vietnam. At HIV testing, 141 PWID (42%) experienced severe depressive symptoms, and over the two years following diagnosis, 82 PWID (24%) died. Controlling for potential confounders, the two-year risk of death among those with depressive symptoms was 9.7% (95% CI: −1.2%-20.6%) higher than the risk among those without depressive symptoms. This increased risk of mortality for PWID with depressive symptoms was relatively consistent throughout the two-year period: at 6, 12, and 18 months, the risk difference was 12.6% (5.5%-19.7%), 13.9% (4.6%-23.2%), and 11.0% (0.9%-21.1%), respectively. HIV diagnosis may provide an important opportunity for depression screening and treatment, subsequently improving survival in this key population.

INTRODUCTION

Infection with human immunodeficiency virus (HIV) contributes to substantial morbidity and mortality worldwide (1,2), and there remain significant global disparities in controlling the epidemic. A particularly vulnerable group are people who inject drugs (PWID) (3–5). Parenteral exposure to infected blood is one of the most efficient means of HIV transmission (6–8), which has resulted in rapid and uncontrolled HIV epidemics among PWID (5). PWID with HIV experience poorer outcomes than their non-injecting peers at all stages of the HIV care continuum, resulting in persistently high rates of mortality (9–12). Injection drug use is the key driver of the HIV epidemic in Asia and Eastern Europe (3,4), and in Vietnam, HIV prevalence among PWID is above 30% in some provinces (13). There is an urgent need to improve care engagement and treatment adherence in this vulnerable population (5,13,14). However, achieving widespread use of antiretroviral therapy (ART) and improving survival among PWID present significant challenges (10,15–17).

One factor that may contribute to poor HIV outcomes among PWID is the high burden of depression. The prevalence of depressive symptoms is as high as 50% among PWID with HIV in Vietnam (18). A large body of research has linked depression to HIV disease progression and mortality (19–23). Compared to patients without depression, depressed HIV patients have lower rates of engaging in HIV care (24–26), adhering to ART (27,28), and achieving viral suppression (29), resulting in a higher risk of mortality (30–32). However, few studies on depression and HIV outcomes focus on PWID (33,34). It is unknown whether identifying and treating depression could improve HIV care engagement and treatment adherence among PWID. Given the high mortality experienced by this group, it is critical to understand the role of depression in worsening outcomes (5,10,14,35–37).

Due to our limited understanding of depression among PWID with HIV, coupled with their disproportionately poor outcomes, the objective of our study was to assess the relationship of depressive symptoms at the time of HIV testing with two-year all-cause mortality among PWID in Vietnam. We hypothesized that PWID with depressive symptoms at the time of HIV testing would have a higher risk of mortality at 6, 12, 18, and 24 months. We focused on the assessment of depressive symptoms at a clinically actionable decision point (i.e., time of HIV testing prior to new diagnosis). We could then examine how hypothetical interventions to reduce depressive symptoms at that time might subsequently impact mortality over the next 24 months.

METHODS

Study population

Our analysis used longitudinal data from a randomized controlled trial of an intervention to reduce HIV- and injection drug use-related stigma and high-risk injecting and sexual behaviors among PWID with HIV in Thai Nguyen from 2009-2013 (38). Thai Nguyen is a northeastern province in Vietnam with increases in injection drug use since the 1990s. At the time of the study, over 6,000 PWID lived in the province (39). Out of the 180 communes in Thai Nguyen, the trial enrolled participants in the 32 communes with the highest number of PWID. Participant recruitment was conducted using snowball sampling, in which recruiters who were former and current drug users approached members of drug networks in private places to discuss study enrollment. The trial enrolled 455 participants who met the following eligibility criteria: 1) HIV-positive diagnosis confirmed through study testing, 2) male (given that 97% of PWID in Thai Nguyen are male), 3) at least 18 years old, 4) had sex in the past 6 months, 5) injected drugs in the previous six months, and 6) planned to live in Thai Nguyen for the next 24 months. Of those 455 enrolled participants, 336 had no history of HIV testing or had not previously received a positive test result. These 336 participants were newly diagnosed with HIV at the trial’s baseline visit and are the focus of this analysis. We focused on new diagnoses to understand how depressive symptoms at the time of HIV testing were associated with mortality over the next two years following diagnosis.

Baseline depression and HIV testing and diagnosis

At the baseline visit, participants received HIV testing, and prior to receipt of HIV results, they were administered a face-to-face interview using a structured questionnaire. The questionnaire collected information on demographics, injecting behaviors and other substance use, quality of life, social support, and HIV and injection drug use-related stigma. The questionnaire included the Center for Epidemiologic Studies Depression Scale (CES-D), which is a 20-item screening tool and widely used measure of depressive symptoms experienced over the past week (40). Previous research has demonstrated that the CES-D is a valid and reliable measure of depressive symptoms in Vietnam (41). Importantly, this depression assessment took place before participants learned the positive result of their HIV test and therefore reflects pre-existing depressive symptoms, not a reaction to the diagnosis. HIV test results were provided within one week of the baseline visit, and all participants received post-test counseling in accordance with the WHO/CDC protocol for HIV testing and counseling.

Follow-up assessments and ascertainment of mortality

Follow-up visits for the trial took place at 6, 12, 18, and 24 months after HIV diagnosis. Participants were asked to provide a blood specimen to assess CD4 cell count and responded to the same questionnaire administered at baseline. Over follow-up, there were no study withdrawals, and for participants who missed a visit, study outreach workers attempted to contact participants using the contact information collected at baseline. During tracing procedures, family members of participants informed the study outreach worker if a participant had died. Therefore, vital status was known for all participants at all visits, through either their completion of the visit or their family member report of vital status at the visit time. Although mortality was ascertained at each 6-month visit, the exact time of death during the 6-month interval since the last visit was not known. For analysis, the time of death was defined as the first scheduled follow-up visit when mortality was ascertained (i.e., 6, 12, 18, or 24 months).

Statistical analysis

The prevalence of depressive symptoms at HIV diagnosis was assessed using the CES-D score measured at the baseline visit, with scores of 23 or greater classified as severe depressive symptoms (18,40,42). We examined the association between depression at HIV diagnosis and mortality over the following two years by using Kaplan-Meier cumulative risk curves to estimate the crude risk difference (RD) in mortality at 6, 12, 18, and 24 months.

We then used inverse probability weights to control for baseline levels of confounders at HIV testing (43). We used a propensity score model to predict the probability of baseline depressive symptoms as a function of potential confounders. Then, each participant was weighted by the inverse of the predicted probability of their observed depression status. In this weighted population, there is no longer an association between baseline depressive symptoms and potential confounding variables. Confounders were determined a priori based on variables that could affect both depression and mortality but did not mediate their relationship (44,45). Based on these criteria, confounders controlled for in the weighted analysis were age, marital status, education, employment, history of drug overdose, experience of HIV and injection drug use-related stigma, perceived social support, self-rated health, CD4 cell count, and trial intervention arm. Current frequency of injection drug use was assessed at baseline but was not included as a confounder in our primary analysis; we hypothesized that baseline depressive symptoms may affect frequency of injection drug use, with drug use then mediating the relationship between depression and subsequent overdose-related mortality. We performed a sensitivity analysis that included frequency of injection drug use as a confounder in the weighted model to determine how much influence this modeling decision had on our study results. Given that the parent trial’s intervention was previously shown to increase 24-month survival (46), we also assessed potential modification of the relationship between depression and mortality by including interaction terms in the model between depression and intervention arms.

Finally, we used the weighted risk estimates for 24-month mortality to calculate the population attributable risk difference (PARD) and the number needed to treat (NNT). The PARD is calculated by subtracting the risk of death for participants without baseline depression (R₀) from the risk of death in the total population (R_T): (R_T − R₀). It indicates the reduction in the 24-month risk of death in our study population if a hypothetical intervention had eliminated depression (without changing other baseline risk factors for mortality). The NNT is the reciprocal of the absolute value of the risk difference (1/|RD|) and corresponds to the number of people we would have to treat with a hypothetical intervention eliminating depression in order to reduce the expected number of deaths by 1 over the 24 months following HIV diagnosis.

For all estimates, our interpretation focuses on the strength of association and precision, rather than statistical significance. All analyses were conducted using R Version 3.4.3 (R Foundation for Statistical Computing, Vienna, Austria).

Ethics

The parent trial and this analysis were approved by the ethical review committees at all participating institutions. Written informed consent was obtained from all participants.

RESULTS

The characteristics of the 336 PWID enrolled in the trial and newly diagnosed with HIV at baseline are shown in Table 1. Based on the trial inclusion criteria, all participants were male. The average age of participants was 35 years, and half were married or cohabitating (49%). One-third had at least a high school education (33%), and the majority were employed full-time (70%). Most participants had a CD4 cell count between 200 and 500 cells/μL (47%) or less than 200 cells/μL (40%). The majority of participants (72%) rated their general health as fair or good, and the remaining 28% of participants rated their general health as poor. All participants reported injecting heroin in the past three months, half reported injecting heroin daily (51%), and a minority reported a history of overdose (18%). Most participants reported current alcohol use (72%) and cigarette smoking (94%). All 336 participants did not know their HIV-positive status at the time of HIV testing, and 141 participants (42%) reported severe depressive symptoms at testing (prior to diagnosis). Over the following 24 months, 82 participants (24%) died.

Table 1.

Baseline characteristics of 336 PWID newly diagnosed with HIV in Thai Nguyen, Vietnam 2009-2013.

Characteristic	N (%) or Mean (SD)
	Baseline depressive symptoms (n=141)	No baseline depressive symptoms (n=195)
Age in years (range 19-60)	35 (6)	35 (6)
Education
Primary school or less	15 (11)	15 (8)
Secondary school	76 (54)	118 (61)
High school	43 (31)	51 (26)
University/college	7 (5)	11 (6)
Employment status
Full-time	97 (69)	138 (71)
Part-time	30 (21)	35 (18)
Unemployed	14 (10)	22 (11)
Marital status
Single	59 (42)	66 (34)
Widowed, divorced, or separated	56 (40)	107 (55)
Married or cohabitating	26 (18)	22 (11)
Self-rated health
Poor	66 (47)	28 (14)
Fair or Good	75 (53)	167 (86)
CD4 cell count (cells/μL)a
≥500	16 (12)	24 (13)
200-499	67 (49)	89 (47)
<200	55 (40)	78 (41)
Current frequency of injection drug usea
1 time/month	4 (3)	12 (6)
2-3 times/month	9 (6)	20 (11)
1 time/week	7 (5)	20 (11)
2-3 times/week	14 (10)	25 (13)
4-6 times/week	18 (13)	34 (18)
Everyday	89 (63)	80 (42)
History of overdose	34 (24)	27 (14)
Current alcohol use	81 (57)	161 (83)
Current cigarette smoking	133 (94)	182 (93)
HIV stigma score (range 14-44)b	30 (5)	30 (4)
Injection drug use stigma score (range 9-28)b	19 (3)	18 (3)
Social support score (range 0-400)c	256 (83)	287 (84)
Trial intervention arm
Control	24 (17)	33 (17)
Community	40 (28)	65 (33)
Individual	30 (21)	37 (19)
Combined (Community+Individual)	47 (33)	60 (31)
24-month mortality	44 (31)	38 (20)

^aCD4 cell count was missing for 7 participants (2%), and frequency of injection drug use was missing for 4 participants (1%).

^bStigma scores were calculated with stigma scales on perceived, internalized, and experienced stigma previously developed for the study population.

^cSocial support was calculated using a modified version of the MOS social support scale.

Table 2 shows the crude and weighted risk differences in mortality at 6, 12, 18, and 24 months following HIV diagnosis for participants with and without severe depressive symptoms at baseline. In the crude analysis, those with depressive symptoms faced a higher absolute risk of death at 24 months (RD = 11.7%, 95% CI: 2.3%, 21.2%). Using a weighted model to control for potential confounders attenuated the 24-month risk difference from 11.7% to 9.7% (95% CI: −1.2%, 20.6%). This increased risk of mortality for PWID with depressive symptoms was fairly consistent in magnitude throughout the two-year period: at 6, 12, and 18 months after HIV diagnosis, the weighted RD was 12.6% (95% CI: 5.5%, 19.7%), 13.9% (95% CI: 4.6%, 23.2%), and 11.0% (95% CI: 0.9%, 21.1%), respectively (Fig. 1). Table 2 also shows results from our sensitivity analysis where we included frequency of injection drug use in the weighted model. All estimates in the sensitivity analysis were slightly attenuated but remained largely consistent in magnitude and direction. Across all analyses and time points, there was an increased risk of mortality for PWID with depressive symptoms between 8 and 14 percentage points (Fig. 2). We did not find evidence of modification of the depression-mortality relationship by trial intervention arm (likelihood ratio test p-value > 0.5).

Table 2.

Association of depressive symptoms at HIV testing with cumulative risk of all-cause mortality at 6, 12, 18, and 24 months among 336 PWID newly diagnosed with HIV in Vietnam.

Months since HIV diagnosis	Cumulative events		Crude analysis	Weighted analysisa	Sensitivity analysisb
	N (%)		RD (95% CI)	RD (95% CI)	RD (95% CI)
6	Depression	25 (18)	14.1 (7.3, 21.0)	12.6 (5.5, 19.7)	12.5 (3.7, 21.2)
	No depression	7 (4)	Ref	Ref	Ref
12	Depression	35 (25)	14.1 (5.7, 22.4)	13.9 (4.6, 23.2)	12.1 (2.1, 22.0)
	No depression	21 (11)	Ref	Ref	Ref
18	Depression	38 (27)	10.5 (1.6, 19.5)	11.0 (0.9, 21.1)	9.2 (−1.8, 20.2)
	No depression	32 (16)	Ref	Ref	Ref
24	Depression	44 (31)	11.7 (2.3, 21.2)	9.7 (−1.2, 20.6)	8.3 (−3.3, 20.0)
	No depression	38 (19)	Ref	Ref	Ref

^aThe weighted model in the main analysis controlled for baseline confounding by age, marital status, education, employment, history of drug overdose, experience of HIV and injection drug use-related stigma, perceived social support, self-rated health, CD4 cell count, and trial intervention arm.

^bThe weighted model in the sensitivity analysis included frequency of injection drug use, in addition to the variables listed above.

Figure 1.

Weighted cumulative risk of death (with 95% CI) by baseline depression (at HIV testing) at 6, 12, 18, and 24 months after HIV diagnosis for 336 PWID in Vietnam.

Figure 2.

Crude and weighted cumulative RDs (with 95% CI) for mortality at 6, 12, 18, and 24 months after HIV diagnosis, comparing PWID with and without depressive symptoms at HIV testing.

Based on the weighted risk estimates for 24-month mortality from our main analysis, we calculated the PARD and NNT to understand how eliminating depression could impact survival. We estimated the PARD to be 4.0% (95% CI: 3.0%, 5.1%), indicating that the two-year risk of death in our study population would have been approximately 4 percentage points lower if a hypothetical intervention had eliminated depression (without changing other baseline risk factors for mortality). The NNT is 10.3, indicating that for every 10 people treated with a hypothetical intervention to eliminate depression, we would reduce the number of deaths by 1 over the 24 months following HIV diagnosis.

DISCUSSION

Our study found that depressive symptoms at the time of HIV testing were strongly associated with all-cause mortality at 6, 12, 18, and 24 months following HIV diagnosis in a sample of 336 PWID in Thai Nguyen, Vietnam. The differences in risk of death across time points were fairly consistent in magnitude, but tended to decrease over time from diagnosis at baseline. Our estimates attenuated slightly when we controlled for confounding using a weighted model and when we included frequency of injection drug use as a confounder in a sensitivity analysis. There was a loss of precision at later time points (18 and 24 months) due to higher mortality; the increase in the underlying risks of death resulted in larger variance. Overall, PWID with depressive symptoms experienced a risk of death between 8 and 14 percentage points higher than the risk among PWID without depressive symptoms throughout the 24 months following HIV diagnosis. Consistent with prior work (30–32), this study shows that people living with HIV and depressive symptoms face a markedly higher risk of mortality than those without depression, and it is the first to demonstrate this relationship among PWID with HIV in the two years following diagnosis. Of note, the overall mortality over two years (24%) was very high among participants. Similar to other provinces in Vietnam, the supply of ART was limited in Thai Nguyen during the study period (2009-2013) and was generally not provided for patients with CD4 cell count above 250 cells/μL (51).

This study focused on the total effect of depressive symptoms at time of HIV testing on mortality over 24 months. There are several possible mechanisms through which depressive symptoms may impact mortality among PWID with HIV. Depression could hinder health-seeking behaviors and interfere with HIV treatment, leading to AIDS-related mortality. Prior studies have shown that depressed HIV patients have lower rates of initiation and retention in HIV care (24–26), ART adherence (27,28), and viral suppression (29). Depression may also increase injection drug use and result in overdose-related mortality, although the time-ordering of the relationship between depression and injection drug use is not clear. There is evidence that depression is a risk factor for injection drug use, preceding its initiation (47,48), and also that depressive symptoms can be substance-induced (49) or a consequence of stigma experienced by PWID (50). In our study data, both depressive symptoms and frequency of injection drug use were assessed at the same time, preventing us from disentangling whether injection drug use led to or resulted from depressive symptoms. As shown in the sensitivity analysis, the RD estimates were slightly reduced when frequency of injection drug use was included in the weighted model. This could be due to removing confounding (if injection drug use resulted in depression) or due to inappropriately adjusting away a mediating path from depression to mortality (if depression resulted in injection drug use). Although we cannot determine the directionality from these data alone, there remains a clear overall association between depression and mortality in this population.

Based on these results, HIV testing and diagnosis may present an important opportunity to screen and treat depressive symptoms. For PWID diagnosed with HIV, post-test counseling and ART clinic visits could include depression screening and referral to care. In Vietnam, mental health services are a national health priority, and there is growing attention and funding for increasing local services and availability of treatment (52). Hypothetically, if depression were successfully treated, this study’s estimates of the PARD and NNT indicate the possible impact on survival in this population. Identifying depressive symptoms at the time of HIV diagnosis and treating those symptoms as part of HIV care has the potential to improve outcomes for this vulnerable population.

An important limitation of our study is that we focus on baseline depressive symptoms. Depression is known to be episodic, with variability in symptom severity over time. We chose to consider only the baseline measure because the timing of this assessment corresponded to a clinically actionable decision point: the time of HIV testing for patients who are newly diagnosed with HIV. Despite not accounting for subsequent changes in depression status, we observed a persistent relationship between baseline depression and mortality over 24 months. Even though the RD in mortality decreased in magnitude at 18 and 24 months, baseline depressive symptoms remained relevant as a determinant of mortality. Another limitation of the depression assessment is that the CES-D corresponds to probable depression, not a clinical diagnosis. However, validation studies have found that the CES-D has high reliability and validity when compared with clinical assessments made by psychiatrists (40,41).

Our study conclusions are specific to this sample and should not be inferred as representative of the population. Due to the snowball sampling approach, participants were not randomly sampled for this study and may differ from other PWID with HIV in Vietnam who were not recruited. In addition, our findings may not be applicable to other groups, such as women or settings where the HIV epidemic is not concentrated among PWID. However, male PWID in Vietnam are a critical population to study. They are the driver of the HIV epidemic in Vietnam, and one-third of PWID in Thai Nguyen are HIV-positive. Our findings may apply to other Asian and Eastern European countries where the HIV epidemic is concentrated among similar groups of men who inject drugs. There is also increasing relevance to the United States, as the current opioid epidemic has the potential to result in explosive HIV outbreaks among PWID.

In conclusion, our study found that PWID with depressive symptoms at time of HIV testing faced a markedly higher risk of death over the next two years. This is the first study to examine the relationship between depression and mortality following HIV diagnosis among PWID. The time of HIV diagnosis offers an important opportunity to screen and treat depressive symptoms and could subsequently improve survival in this key population.