Skip to main content
American Journal of Epidemiology logoLink to American Journal of Epidemiology
. 2019 Oct 29;189(5):422–432. doi: 10.1093/aje/kwz238

Depression and Incident HIV in Adolescent Girls and Young Women in HIV Prevention Trials Network 068: Targets for Prevention and Mediating Factors

Dana E Goin 1, Rebecca M Pearson 2,3, Michelle G Craske 4,5, Alan Stein 6,7, Audrey Pettifor 6,8,9, Sheri A Lippman 6,10, Kathleen Kahn 6, Torsten B Neilands 10, Erica L Hamilton 11, Amanda Selin 6,12, Catherine MacPhail 6,9,13, Ryan G Wagner 6, F Xavier Gomez-Olive 6, Rhian Twine 6, James P Hughes 14, Yaw Agyei 15, Oliver Laeyendecker 16,17,18, Stephen Tollman 6, Jennifer Ahern 1,
PMCID: PMC7306677  PMID: 31667490

Abstract

The human immunodeficiency virus (HIV) epidemic among adolescent girls and young women (AGYW) in sub-Saharan Africa is a critical public health problem. We assessed whether depressive symptoms in AGYW were longitudinally associated with incident HIV, and identified potential social and behavioral mediators. Data came from a randomized trial of a cash transfer conditional on school attendance among AGYW (ages 13–21 years) in rural Mpumalanga Province, South Africa, during 2011–2017. We estimated the relationship between depressive symptoms and cumulative HIV incidence using a linear probability model, and we assessed mediation using inverse odds ratio weighting. Inference was calculated using the nonparametric bootstrap. AGYW with depressive symptoms had higher cumulative incidence of HIV compared with those without (risk difference = 3.5, 95% confidence interval (CI): 0.1, 7.0). The strongest individual mediators of this association were parental monitoring and involvement (indirect effect = 1.6, 95% CI: 0.0, 3.3) and reporting a partner would hit her if she asked him to wear a condom (indirect effect = 1.5, 95% CI: –0.3, 3.3). All mediators jointly explained two-thirds (indirect effect = 2.4, 95% CI: 0.2, 4.5) of the association between depressive symptoms and HIV incidence. Interventions addressing mental health might reduce risk of acquiring HIV among AGYW.

Keywords: adolescent health, depression, HIV/AIDS, South Africa

Abbreviations

AGYW

adolescent girls and young women

HIV

human immunodeficiency virus

HPTN

HIV Prevention Trials Network

Every hour, 30 adolescents are infected with the human immunodeficiency virus (HIV) (1). In 2016, the global number of adolescents aged 10–19 years living with HIV rose to 2.1 million, a 30% increase since 2005 (1). Of new HIV infections among adolescents, 3 in 4 occur in sub-Saharan Africa (1), and adolescent girls are especially burdened. The ratio of adolescent incident HIV infection in sub-Saharan Africa was almost 3 girls to each boy in 2017 (2). Understanding risk factors during adolescence before infection is thus crucial to inform prevention strategies.

The most common method of HIV transmission worldwide, at almost 70% of infections, is unprotected heterosexual intercourse (3). Therefore, age at sexual debut, choice of partner, and risky sexual behavior are key targets for prevention (4). However, given limited success in reducing HIV incidence, there is increasing recognition of the critical need to address upstream factors that shape risky behaviors (5–8).

Long-standing evidence demonstrates a link between depression and HIV in sub-Saharan Africa (9, 10). Depression is known to influence treatment adherence and mortality among HIV-infected individuals (11, 12). Furthermore, a high prevalence of depressive disorders has been reported at the time of HIV testing (13) and among HIV-infected individuals in South Africa (14–16). It is understood that a diagnosis of HIV can contribute to depression (17), given that it is a life-threatening illness requiring lifelong management and medication. The stigma of HIV likely contributes further to depression (18), and concerns about losing a job or partnership or, in the case of pregnancy, the prospect of the infant acquiring HIV, add more stress (19).

However, the possibility of depression as a risk factor for HIV incidence in sub-Saharan Africa has not been examined to date. Depressive symptoms have been linked to increased risk of HIV incidence among men who have sex with men in the United States (20, 21), but the relationship has not previously been tested among adolescent girls and young women (AGYW) in sub-Saharan Africa, although a number of plausible pathways exist. Depressive symptoms among young South African women are associated with increased alcohol use and risky sexual behaviors (22–25), both considered to be risk factors for HIV (3, 26, 27). The association between depression and risky sexual behaviors might be related to sexual abuse as a child or sexual assault as an adolescent/adult (28). Furthermore, depression could lead to use of sex as a coping mechanism to regulate negative emotions, increase a young woman’s desire to please her partners (4, 29), and contribute to a lack of efficacy or power in sexual relationships (30–32). Depressive symptoms are also associated with reduced educational attainment (33), another known risk factor for HIV in this population (26, 27, 34), and limited involvement or engagement in community life (35). Finally, depression in adolescence can be associated with poor relationships with parents and can reduce parental closeness or involvement (36–38), which can in turn put girls at greater risk due to lower parental monitoring and advice.

Existing evidence linking depression to risky sexual behavior and HIV infection in sub-Saharan Africa relies on cross-sectional studies that leave the direction of causality unclear (i.e., is depression a cause or consequence of risky sexual behavior/HIV?) (9, 10, 14, 17, 39–41). In the present study, we used unique longitudinal data to test the hypothesis that depressive symptoms in early adolescence are associated with incident HIV infection. Furthermore, we investigated whether a range of social and behavioral factors mediated the relationship between depressive symptoms and incident HIV. This study was conducted using the HIV Prevention Trials Network (HPTN) 068 cohort, a longitudinal study of AGYW in South Africa. We hypothesized that depressive symptoms in adolescence would increase HIV incidence and that this would be explained, at least in part, by behavioral mediators. Because depression is common and potentially modifiable in low and middle-income countries (42), it represents a promising prevention target.

METHODS

Participants

The HPTN 068 cohort of AGYW was established in 2012 as part of a National Institutes of Health–funded randomized controlled trial (HPTN 068) to estimate the effects of cash transfers conditional on school attendance on HIV incidence among adolescent girls in rural Mpumalanga Province, South Africa (26). Participants included 2,533 girls and women in grades 8–11 (aged 13–21 years) at baseline. Eligibility criteria required the enrollees to be unmarried, not pregnant at enrollment, and able to read. They and their parents/guardians were also required to be able to open a bank account. All households with eligible AGYW in the study area were recruited; only 7% did not participate, due to either parent/guardian or participant refusal (43). The study did not exclude girls who tested positive for HIV infection at baseline in order to avoid undesired disclosure of HIV status. Additional details on the study recruitment and enrollment have been previously published (26, 43). Participants were followed annually for up to 4 years, with additional graduation and postintervention follow-up visits that resulted in follow-up duration of up to 6 years. Retention during the study period was approximately 91% (26). Participants completed detailed surveys about their economic and health behaviors, attitudes regarding social norms, and life experiences, and they provided blood samples to test for HIV each year. Surveys about household composition and wealth were also undertaken with the participant’s head of household, providing rich longitudinal cohort data with HIV incidence at multiple time points throughout adolescence.

Institutional review board approval for this study was obtained from the University of North Carolina at Chapel Hill, the University of the Witwatersrand Human Research Ethics Committee, and the Mpumalanga Province Health Research and Ethics Committee. Approval for analytical work was also obtained by the University of California, Berkeley, and the University of California, San Francisco.

Exposure definition

To assess depressive symptoms, we used a 10-item short form of the Children’s Depression Inventory scale (44), administered at the baseline visit. The reduced scale has comparable results with the full scale and has previously been used in South Africa (45–47); the full scale has been validated in sub-Saharan Africa (48). Each question had 3 levels, and following standard practice, we coded the most positive response as 0, the middle response as 1, and the most negative response as 2. We summed the items over the 10 questions and we dichotomized the index at 7, which is the prorated equivalent of the recommended cutoff, at 19, from the full 27-item Children’s Depression Inventory, which is used to identify adolescents with depressive disorders in nonclinical samples (44, 45). In a validation study of the short-form Children’s Depression Inventory in Rwanda, the sensitivity and specificity were 0.64 and 0.97, respectively, for the cutoff value of 7 (48).

Outcome definition

HIV status was determined at each follow-up visit by administering 2 parallel rapid HIV tests (the Determine HIV-1/2 test (Alere Medical Co., Matsudo-shi, Chiba, Japan) and the US Food and Drug Administration–cleared Uni-gold Recombigen HIV test (Trinity Biotech, Bray, County Wicklow, Ireland)). If one or both tests were reactive or positive, confirmatory HIV testing was done with the Food and Drug Administration–cleared GS HIV-1 Western Blot Kit (Bio-Rad Laboratories Inc., Redmond, Washington). A detailed description of study visit and laboratory procedures is published elsewhere (26). Cumulative incidence of HIV was evaluated from the second through the fifth follow-up visits, which spanned a period of up to 6 years.

Mediator definitions

We examined the following possible mediators, assessed at the first follow-up visit, which was approximately a year after the measurement of depressive symptoms: health and sexual behaviors, indicators of relationship power, and family or social involvement. Health and sexual behaviors included alcohol use more than once a month, number of partners in the previous 12 months, unprotected sex in the previous 3 months, and purchased or received birth control or condoms in the previous month. Indicators of relationship power included a recent sexual partner ≥5 years older, having engaged in transactional sex in the past 12 months, a partner who would hit her if she asked him to use a condom, a partner who would be angry if she asked him to use a condom, a partner who would suspect she was cheating if she asked him to use a condom, low power according to the Sexual Relationship Power Scale (49), and intimate partner violence in the previous 12 months. Family and social involvement mediators included having an average of >5 days of missed school per month during the previous year, having attended >80% of school days between follow-up visits, parental monitoring and involvement in girl’s life (a detailed description of this measure is provided in Web Appendix 1, available at https://academic.oup.com/aje), and the number of community organizations of which the girl was a member.

We chose these mediators based on literature that had identified these factors as behaviors that could be affected by depressive symptoms, could modify HIV risk, and could plausibly be targeted by interventions.

Analytical plan

To establish temporal ordering between depressive symptoms, mediators, and HIV, we excluded AGYW who were HIV-infected at the baseline or first follow-up visit (n = 113, 4.5%). We considered mediators from the first annual follow-up visit to ensure temporality between depression at the baseline visit and mediators (Figure 1). HIV incidence was included from the second to the fifth annual follow-up visits, which spanned a period of up to 6 years.

Figure 1.

Figure 1

Timeline of depressive symptoms measurement, mediator assessment, and human immunodeficiency virus (HIV) testing over the course of the study, HIV Prevention Trials Network 068 Cohort, Mpumalanga, South Africa, 2011–2017.

We controlled for baseline covariates that might have affected depressive symptoms, HIV incidence, and the mediators, based on theory and prior literature. These included household food insecurity, per capita household consumption, percentage of household members receiving a government grant, adverse and positive life events, orphan status, maternal and paternal educational attainment, pregnancy history, age, and trial arm. Detailed definitions of the covariates are available in Web Appendix 2.

We estimated the effect of depressive symptoms on cumulative HIV incidence using a linear probability model to estimate parameters on the additive (risk difference) scale (50, 51). For the mediation analyses, we estimated natural direct and indirect effects using inverse odds-ratio weighting and considered multiple mediators (52, 53). This method uses the invariance property of the odds ratio to create weights that block the association between depression and the mediator(s), thus isolating the natural direct effect of depression (i.e., the effect that does not go through the mediator(s)). Some strengths of this approach, in contrast to other available mediation methods (54, 55), are that it allows multiple mediators to be considered simultaneously and allows for interactions between depression and the mediators. The natural indirect effect, the association of depression with HIV via the mediator(s), is the difference between the total effect and the natural direct effect.

To account for covariate missingness we used multiple imputation with 30 imputed data sets (56). The nonparametric bootstrap was used to calculate 95% Wald-type confidence intervals. We estimated direct and indirect effects for each mediator separately and all mediators jointly. We conducted a sensitivity analysis using inverse probability of censoring weights to assess whether loss to follow-up affected our results. We also conducted a quantitative bias analysis using results on the risk ratio scale to assess how strong mediator-outcome confounding would need to be to change our results (57). A full description of this procedure is provided in Web Appendix 3.

All analyses were performed using R, version 3.4 (R Foundation for Statistical Computing, Vienna, Austria) (58); the code is available in Web Appendix 4.

RESULTS

Main effects

There were 2,533 AGYW included in the study; of these, 118 were excluded as a result of having no information on HIV status (n = 5, 0.2%) or were HIV-positive at baseline or the first follow-up visit (n = 113, 4.5%). This resulted in a sample size of 2,415. The prevalence of depressive symptoms and incidence of HIV varied across the baseline covariates (Table 1), but overall, 18.2% of AGYW were classified as having depressive symptoms at baseline, and the cumulative incidence of HIV after the first follow-up visit was 0.072 (Table 2). AGYW with depressive symptoms had 0.107 cumulative HIV incidence over the years of follow-up, compared with 0.065 cumulative incidence among those without depressive symptoms (Table 2). After adjustment for covariates, depressive symptoms at baseline were associated with higher cumulative incidence of HIV (risk difference = 3.5, 95% confidence interval: 0.1, 7.0) (Table 3).

Table 1.

Descriptive Statistics for Baseline Covariates, Severe Depressive Symptoms, and Human Immunodeficiency Virus Incidence for the HIV Prevention Trials Network 068 Cohort, Mpumalanga, South Africa, 2011–2017

Covariate No. Depressive Symptoms, % HIV-Positive, %
Orphaning statusa
 1 parent deceased 567 19.5 8.1
 2 parents deceased 105 20.8 10.2
Household economic status
 Food insecureb
  No 2,022 17.9 7.7
  Yes 393 19.5 10.1
 Decile of total household per capita expendituresb
  1 242 16.4 9.6
  2 235 17.0 8.6
  3 240 21.5 7.6
  4 244 15.5 6.6
  5 246 20.9 7.3
  6 242 19.0 8.8
  7 241 20.7 12.6
  8 239 17.4 8.4
  9 240 18.0 7.6
  10 246 15.4 3.7
 >25% of household receiving a government grantb
  No 560 14.8 7.0
  Yes 1,855 19.2 8.4
 Maternal educationc
  No school 443 18.7 8.0
  Some primary 480 21.8 9.4
  Completed primary 111 23.0 7.1
  Some high school 716 17.9 9.9
  Completed high school 570 16.0 5.6
  University or vocational 94 7.1 3.0
 Paternal education
  No school 499 17.9 9.4
  Some primary 433 21.3 9.2
  Completed primary 125 20.7 8.5
  Some high school 553 17.6 8.4
  Completed high school 687 17.3 7.0
  University or vocational 118 13.2 2.1
 Any negative life eventb
  No 1,772 17.8 8.2
  Yes 643 19.3 7.8
 Any positive life eventb
  No 1,965 18.4 7.8
  Yes   450 17.1 9.3
Table continues
Demographic factors
 Ever been pregnant
  No 2,211 16.6 7.5
  Yes 204 35.4 15.0
 Age at baseline, years
  13 269 15.2 6.3
  14 482 13.5 5.5
  15 527 18.1 7.5
  16 505 16.1 8.7
  17 358 18.6 10.0
  18 163 34.7 11.4
  19 79 31.4 11.9
  ≥20 32 28.2 19.2
Trial arm
 Control 1,212 17.3 8.3
 Intervention 1,203 19.1 7.9

Abbreviation: HIV, human immunodeficiency virus.

a This category does not sum to the total because it does not include participants for whom both parents are still living.

b Details on measurement and definitions of these variables are available in Web Appendix 2.

c The total for this category does not sum to the total due to rounding across multiply imputed data sets.

Table 2.

Human Immunodeficiency Virus Incidence According to Exposure Status and Year of Follow-Up for the HIV Prevention Trials Network 068 Cohort, Mpumalanga, South Africa, 2011–2017

No. of Years of Follow-up Overall No Depressive Symptoms Depressive Symptoms
No. of HIV Cases Observed Person-Years at Riska Cases/Person-Years Cumulative Incidenceb No. of HIV Cases Observed Person-Years at Risk Cases/Person-Years Cumulative Incidenceb No. of HIV Cases Observed Person-Years at Risk Cases/Person-Years Cumulative Incidenceb
2 22 2,329 0.009 0.009 16 1,906 0.008 0.008 6 423 0.014 0.014
3 43 2,136 0.020 0.027 27 1,747 0.015 0.022 16 389 0.041 0.050
4 42 1,913 0.022 0.044 34 1,570 0.022 0.039 8 343 0.023 0.068
5 36 1,336 0.027 0.059 26 1,093 0.024 0.052 10 243 0.041 0.091
6 32 671 0.048 0.072 25 547 0.046 0.065 7 124 0.056 0.107
Total 175 8,385 0.021 0.072 128 6,863 0.019 0.065 47 1,522 0.031 0.107

Abbreviation: HIV, human immunodeficiency virus.

a The total observed person-years at risk at the second year of follow-up is n = 2,329 rather than the total n = 2,415 because 86 were lost to follow-up before the second year.

b Cumulative incidence includes all cases up through the relevant year of follow-up.

Table 3.

Estimated Total Effectsa of Depressive Symptoms on Incidence of Human Immunodeficiency Virus and Estimated Natural Direct and Indirect Effects of Several Mediators in the HIV Prevention Trials Network 068 Cohort, Mpumalanga, South Africa, 2011–2017b

Mediator Natural Direct Effect 95% CI Natural Indirect Effect 95% CI
Health and school behaviors
 Alcohol use in the past month 2.40 −1.0, 5.8 1.10 −0.4, 2.7
 More than 5 days of missed school on average per month in the previous year 2.10 −1.3, 5.4 1.40 −0.2, 3.1
 High school attendance between rounds of >80% 2.30 −1.1, 5.7 1.20 −0.4, 2.8
 Parental monitoring and involvement 1.90 −1.3, 5.0 1.60 0.0, 3.3
 Community engagement (number of community organizations involved in) 2.60 −0.9, 6.1 0.90 −0.7, 2.5
Sexual behaviors and self-efficacy
 Had a recent sexual partner ≥5 years older than oneself 2.50 −0.9, 5.9 1.00 −0.5, 2.6
 No. of partners in the previous 12 months (0, 1, 2, or more) 2.30 −1.1, 5.7 1.20 −0.4, 2.9
 Had unprotected sex in the previous 3 months 2.20 −1.2, 5.5 1.40 −0.3, 3.0
 Engaged in transactional sex in the previous 12 months 2.50 −1.0, 6.0 1.00 −0.6, 2.6
 Sexual relationship power: partner would hit me if I requested he use a condom 2.00 −1.3, 5.2 1.50 −0.3, 3.3
 Sexual relationship power: partner would be angry if I requested he use a condom 2.20 −1.1, 5.6 1.30 −0.4, 3.0
 Sexual relationship power: partner would assume I’m cheating if I ask him to use a condom 2.20 −1.2, 5.5 1.40 −0.3, 3.0
 Low power according to sexual relationship power scale 2.30 −1.0, 5.6 1.20 −0.4, 2.9
 Experienced intimate partner violence in the previous 12 months 2.20 −1.1, 5.5 1.30 −0.3, 2.8
 Purchased birth control or condoms 2.40 −1.0, 5.7 1.20 −0.4, 2.7
All mediators 1.10 −2.1, 4.4 2.40 0.2, 4.5

Abbreviations: CI, confidence interval; HIV, human immunodeficiency virus.

a The estimated total effect risk difference was 3.5 (95% CI: 0.1, 7.0).

b Estimates were adjusted for household consumption, food insecurity, any negative or positive events, percent of the household receiving a government grant, orphaning status, maternal and paternal educational attainment, pregnancy history, age at baseline, and arm of the trial.

Mediation

Estimates of mediation parameters between depressive symptoms and HIV are presented in Table 3. The estimated natural direct effect on the risk difference scale of depressive symptoms on incident HIV, considering all mediators jointly, was 1.1 (95% confidence interval: −2.1, 4.4), while the estimated natural indirect effect considering all mediators was 2.4 (95% confidence interval: 0.2, 4.5), suggesting that the mediators explain about two-thirds of the absolute association between depressive symptoms and HIV when considered jointly. The remaining estimate of the natural direct effect suggests that depressive symptoms might also increase the risk of HIV through biological or other behavioral mechanisms not examined here.

The mediators that individually explained the largest portion of the association between depressive symptoms and HIV incidence were parental monitoring and involvement and reporting that a partner would hit her if she asked him to wear a condom. The next strongest mediators were missing ≥5 days of school on average per month in the past year, having unprotected sex in the past 3 months, reporting her partner would assume she was cheating or become angry if she asked him to use a condom, and experiencing intimate partner violence in the past 12 months. Having a partner ≥5 years older, engaging in transactional sex, and the number of community organizations AGYW were involved with mediated the smallest portions of the overall association. The results did not meaningfully change when we used inverse probability of censoring weights to control for differential loss to follow-up (Web Table 1).

We conducted a quantitative bias analysis on the risk ratio scale for the amount of unobserved mediator-outcome confounding that could change the results, focusing on the mediation analysis that considered all mediators jointly (Web Table 2). We found that unobserved confounding would need to be strong in order to alter our findings (Web Tables 3 and 4).

DISCUSSION

This study examined the association of depressive symptoms with HIV incidence among AGYW in South Africa, a population bearing a large burden of HIV/AIDS and the highest number of incident cases in the region hardest hit by the epidemic. We found prevalence of depressive symptoms of 18.2% in this young adolescent population, which was almost twice the prevalence recorded in a nationally representative survey of adults in South Africa (59). Previous evidence in sub-Saharan Africa showed that HIV infection increased risk of depression (15, 18, 40), and depression reduced treatment adherence and survival among HIV-positive individuals (10, 11, 41). Cross-sectional studies also suggested a co-occurrence of depression and HIV (14, 39). The findings of this study suggest that having depressive symptoms in adolescence might also increase risk of HIV infection. To our knowledge, this is the first demonstration that depression is longitudinally related to HIV acquisition among AGYW in sub-Saharan Africa. The results are especially important given the high burden of depression among adolescents and point to areas for prevention strategies.

Because depression has been linked to lower condom use (30), and unprotected sex is the main route of HIV transmission in South Africa (60), we considered mediators that might connect depressive symptoms to behaviors that could influence risk of HIV. Mediators considered in this study—including alcohol use, school attendance, parental monitoring and involvement, community participation, sexual behaviors, and sexual relationship power—jointly explained about two-thirds of this association. Parental monitoring and involvement and reporting fear that a partner would hit her if she requested that he use a condom were responsible for the largest estimated individual mediation effects. However, we acknowledge that these mediators represent complex behaviors whose development could precede and contribute to the manifestation of depressive symptoms. In this study, we were not able to disentangle all the interconnections between these complicated emotional and social factors over time, but by measuring them at distinct visits we attempted to capture meaningful temporal connections between them and their relationships to HIV incidence.

Evidence indicates that parental monitoring and involvement can influence HIV risk. Parental involvement has been linked to reduced sexual activity and risky sexual behavior among adolescents in sub-Saharan Africa (61, 62), and parental communication around sexuality and sexual risk-taking with adolescent children has been tied to improved self-efficacy and interpersonal communication about sex, especially among AGYW (36). Positive parenting has been identified as an important contributor to adolescent physical and mental health in sub-Saharan Africa (63–65), but the directional relationship between parental involvement and child mental health and behavior is less well-studied. However, evidence from the United States indicates that the relationship between parental monitoring and child behavior is bidirectional (37) and that the direction of the relationship changes as children age into adolescence (38). For example, increased parental monitoring has been associated with lower childhood depression (66), but during adolescence, child behavior can also affect parental monitoring and parenting behaviors (38). Our results suggest that lack of parental monitoring and involvement plays an important role in mediating the association of adolescent depressive symptoms with HIV incidence, such that depression leads to less involvement from parents, which then increases risk of HIV. However, it is also possible that parental monitoring and involvement earlier in a child’s life influences later parental behavior, in addition to depressive symptoms and risky sexual behaviors.

Having gender-based power imbalances in sexual relationships has been tied to increased risk of acquiring HIV and other sexually transmitted infections in South Africa (67, 68), and it has also been associated with inconsistent condom use (31). Previous work has shown that South African men who were physically violent toward a partner were more likely to be HIV-positive, increasing HIV risk for those partners (69). Prior work has found that young HIV-infected women with low sexual relationship power are at increased risk of depression in sub-Saharan Africa (32), and the results of this study indicate that AGYW with depressive symptoms might be more vulnerable to relationships in which they have less power or control over decision-making. One of the largest estimated mediation effects was observed for AGYW reporting that a partner would hit her if she requested that he use a condom. Having a partner who would assume infidelity or become angry if she requested that he use a condom during sex and experiencing intimate partner violence in the past 12 months also had some of the largest estimated mediation effects. These results indicate that depressive symptoms might reduce the ability of AGYW to negotiate condom use during sex and could increase their risk of HIV. Depression might also be associated with choosing sexual partners who are less open to negotiation or who are more likely to commit intimate partner violence.

In South Africa, children’s mental health has been linked to school attendance and other educational outcomes (70). Furthermore, remaining in school and educational attainment during adolescence are protective against HIV incidence among AGYW in South Africa (34). School attendance was inversely related to both depressive symptoms and HIV incidence in this cohort, and missing more than 5 days of school on average per month was among the largest estimated individual mediation effects.

There were some limitations to this work. Although we were able to control for a large set of social, economic, and demographic covariates, residual confounding is always a concern in observational studies. Our total effect estimates assume there is no unmeasured confounding between the exposure and outcome. Additional assumptions for causal inference are positivity, consistency, and the stable unit treatment value assumption. Furthermore, our measurements of depressive symptoms and social and behavioral factors were self-reported. This could introduce same-source bias, in which measurement error is correlated, because, for example, depressive symptoms could affect a young woman’s perception of parental monitoring or recall of past behavior. This is a limitation for the measurement of our exposure and mediators but not the outcome: HIV status was determined by blood test. In addition, while our sample included a large cohort of AGYW, we nonetheless had limited power to detect natural direct and indirect effects, particularly for a rare outcome. Detection of mediation effects tends to be underpowered with sample sizes designed for main effects only (71). Furthermore, when estimating natural direct and natural indirect effects there are several assumptions that need to be considered beyond those in a total effect analysis. In particular, in addition to concern about confounding between the exposure and outcome, there must also be no confounding of the exposure and the mediator and between the mediator and the outcome. We conducted a quantitative bias analysis to assess how strong unmeasured mediator-outcome confounding would need to be to invalidate our results and found that strong confounding would be necessary, increasing our confidence in the findings.

Our results suggest that interventions that improve mental health among AGYW might also improve HIV-prevention efforts. They suggest the importance of incorporating family support and educational or empowerment initiatives around condom use, negotiating condom use with respect to gender-based sexual relationship power, and addressing male norms around condom use in prevention programming. While depression is a modifiable public health problem, access to mental health-care services in South Africa is currently quite poor. Nevertheless, relatively low-cost interventions, delivered by lay workers in the community, have been shown to be effective in low- and middle-income countries, such as the rural South African context (72–74). Such interventions in adolescence, targeting depression as well as parental relationships and addressing social and structural factors that encourage women to participate in sexual relationships with gender-based power dynamics, could help prevent young women from acquiring HIV.

Supplementary Material

kwz238_Goin_Web_Material_Final

ACKNOWLEDGMENTS

Author affiliations: Division of Epidemiology & Biostatistics, School of Public Health, University of California, Berkeley, Berkeley, California (Dana E. Goin, Jennifer Ahern); Centre for Academic Mental Health, Population Health Sciences, Bristol Medical School, Bristol University, Bristol, United Kingdom (Rebecca M. Pearson); Bristol Biomedical Research Centre, National Institute for Health Research, Bristol, United Kingdom (Rebecca M. Pearson); Department of Psychology, University of California, Los Angeles, Los Angeles, California (Michelle G. Craske); Department of Psychiatry and Biobehavioral Sciences, University of California, Los Angeles, Los Angeles, California (Michelle G. Craske); Medical Research Council/Wits University Rural Public Health and Health Transitions Research Unit (Agincourt), School of Public Health, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa (Alan Stein, Audrey Pettifor, Sheri A. Lippman, Kathleen Kahn, Amanda Selin, Catherine MacPhail, Ryan G. Wagner, F. Xavier Gomez-Olive, Rhian Twine, Stephen Tollman); Department of Psychiatry, University of Oxford, Oxford, United Kingdom (Alan Stein); Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina, Chapel Hill, Chapel Hill, North Carolina (Audrey Pettifor); Wits Reproductive Health and HIV Institute, Faculty of Health Sciences, School of Clinical Medicine, University of the Witwatersrand, Johannesburg, South Africa (Audrey Pettifor, Catherine MacPhail); Division of Prevention Science, Department of Medicine, University of California, San Francisco, San Francisco, California (Sheri A. Lippman, Torsten B. Neilands); HIV Prevention Trials Network Leadership and Operations Center, Science Facilitation Department, FHI 360, Durham, North Carolina (Erica L. Hamilton); Carolina Population Center, University of North Carolina, Chapel Hill, Chapel Hill, North Carolina (Amanda Selin); School of Health and Society, University of Wollongong, Wollongong, New South Wales, Australia (Catherine MacPhail); Department of Biostatistics, School of Public Health, University of Washington, Seattle, Washington (James P. Hughes); Department of Pathology, School of Medicine, Johns Hopkins University, Baltimore, Maryland (Yaw Agyei); Laboratory of Immunoregulation, National Institute of Allergy and Infectious Diseases, National Institutes of Health, Bethesda, Maryland (Oliver Laeyendecker); Division of Infectious Diseases, School of Medicine, Johns Hopkins University, Baltimore, Maryland (Oliver Laeyendecker); and Department of Epidemiology, School of Public Health, Johns Hopkins University, Baltimore, Maryland (Oliver Laeyendecker).

This study was funded by the National Institutes of Health (grant R01 MH110186) and by the National Institute of Allergy and Infectious Diseases, the National Institute of Mental Health, and the National Institute on Drug Abuse (awards UM1 AI068619 (HPTN Leadership and Operations Center), UM1AI068617 (HIV Prevention Trials Network Statistical and Data Management Center), and UM1AI068613 (HIV Prevention Trials Network Laboratory Center)). This work was also supported by the National Institute of Mental Health (grant R01MH087118) and the Carolina Population Center (grant P2C HD050924). Additional support was provided by the Division of Intramural Research, National Institute of Allergy and Infectious Diseases. The MRC/Wits Rural Public Health and Health Transitions Research Unit and Agincourt Health and Socio-Demographic Surveillance System, a node of the South African Population Research Infrastructure Network (SAPRIN), have been supported by the National Department of Science and Innovation, the University of the Witwatersrand, and the Medical Research Council, South Africa, and by the Wellcome Trust, United Kingdom (grants 058893/Z/99/A, 069683/Z/02/Z, 085477/Z/08/Z, and 085477/B/08/Z).

We thank the HIV Prevention Trials Network 068 study team and all trial participants.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conflict of interest: none declared.

REFERENCES

  • 1. UNICEF Children and AIDS: statistical update. https://data.unicef.org/wp-content/uploads/2017/11/HIVAIDS-Statistical-Update-2017.pdf. Accessed August 25, 2018.
  • 2. UNICEF Adolescent HIV prevention. https://data.unicef.org/topic/hivaids/adolescents-young-people/. Accessed August 25, 2018.
  • 3. Shaw GM, Hunter E. HIV transmission. Cold Spring Harb Perspect Med. 2012;2(11):a006965. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4. Lara LAS, Abdo CHN. Age at time of initial sexual intercourse and health of adolescent girls. J Pediatr Adolesc Gynecol. 2016;29(5):417–423. [DOI] [PubMed] [Google Scholar]
  • 5. Harrison A, Colvin CJ, Kuo C, et al. . Sustained high HIV incidence in young women in southern Africa: social, behavioral, and structural factors and emerging intervention approaches. Curr HIV/AIDS Rep. 2015;12(2):207–215. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Ramjee G, Daniels B. Women and HIV in sub-Saharan Africa. AIDS Res Ther. 2013;10(1):30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. UNAIDS On the fast-track to end AIDS by 2030: focus on location and population. http://www.unaids.org/en/resources/documents/2015/FocusLocationPopulation. Accessed August 25, 2018.
  • 8. UNAIDS Prevention gap report. http://www.unaids.org/sites/default/files/media_asset/2016-prevention-gap-report_en.pdf. Accessed August 25, 2018.
  • 9. Bernard C, Dabis F, de Rekeneire N. Prevalence and factors associated with depression in people living with HIV in sub-Saharan Africa: a systematic review and meta-analysis. PLoS One. 2017;12(8):e0181960. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Abas M, Ali G-C, Nakimuli-Mpungu E, et al. . Depression in people living with HIV in sub-Saharan Africa: time to act. Trop Med Int Health. 2014;19(12):1392–1396. [DOI] [PubMed] [Google Scholar]
  • 11. Nakimuli-Mpungu E, Bass JK, Alexandre P, et al. . Depression, alcohol use and adherence to antiretroviral therapy in sub-Saharan Africa: a systematic review. AIDS Behav. 2012;16(8):2101–2118. [DOI] [PubMed] [Google Scholar]
  • 12. Antelman G, Kaaya S, Wei R, et al. . Depressive symptoms increase risk of HIV disease progression and mortality among women in Tanzania. J Acquir Immune Defic Syndr. 2007;44(4):470–477. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Kagee A, Saal W, De Villiers L, et al. . The prevalence of common mental disorders among South Africans seeking HIV testing. AIDS Behav. 2017;21(6):1511–1517. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Myer L, Smit J, Roux LL, et al. . Common mental disorders among HIV-infected individuals in South Africa: prevalence, predictors, and validation of brief psychiatric rating scales. AIDS Patient Care STDS. 2008;22(2):147–158. [DOI] [PubMed] [Google Scholar]
  • 15. Olley BO, Gxamza F, Seedat S, et al. . Psychopathology and coping in recently diagnosed HIV/AIDS patients—the role of gender. S Afr Med J. 2003;93(12):928–931. [PubMed] [Google Scholar]
  • 16. Shisana O, Rehle T, Simbayi LC, et al. South African National HIV prevalence, incidence and behaviour survey, 2012. http://repository.hsrc.ac.za/handle/20.500.11910/2490. Accessed September 7, 2018. [DOI] [PubMed]
  • 17. Simoni JM, Safren SA, Manhart LE, et al. . Challenges in addressing depression in HIV research: assessment, cultural context, and methods. AIDS Behav. 2011;15(2):376–388. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Simbayi LC, Kalichman S, Strebel A, et al. . Internalized stigma, discrimination, and depression among men and women living with HIV/AIDS in Cape Town, South Africa. Soc Sci Med. 2007;64(9):1823–1831. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Brittain K, Mellins CA, Phillips T, et al. . Social support, stigma and antenatal depression among HIV-infected pregnant women in South Africa. AIDS Behav. 2017;21(1):274–282. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Koblin BA, Husnik MJ, Colfax G, et al. . Risk factors for HIV infection among men who have sex with men. AIDS. 2006;20(5):731–739. [DOI] [PubMed] [Google Scholar]
  • 21. Mimiaga MJ, O’Cleirigh C, Biello KB, et al. . The effect of psychosocial syndemic production on 4-year HIV incidence and risk behavior in a large cohort of sexually active men who have sex with men. J Acquir Immune Defic Syndr. 2015;68(3):329–336. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Abler LA, Sikkema KJ, Watt MH, et al. . Longitudinal cohort study of depression, post-traumatic stress, and alcohol use in South African women who attend alcohol serving venues. BMC Psychiatry. 2014;14:224. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Davis EC, Rotheram-Borus MJ, Weichle TW, et al. . Patterns of alcohol abuse, depression, and intimate partner violence among township mothers in South Africa over 5 years. AIDS Behav. 2017;21(2):174–182. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Onah MN, Field S, van Heyningen T, et al. . Predictors of alcohol and other drug use among pregnant women in a peri-urban South African setting. Int J Ment Health Syst. 2016;10:38. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25. Parry CDH, Tomlinson M, Bryant K, et al. . Fresh perspectives on the alcohol and HIV nexus: a call for action in an era of increased opportunities and challenges. AIDS Behav. 2017;21(suppl 2):121–125. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Pettifor A, MacPhail C, Hughes JP, et al. . The effect of a conditional cash transfer on HIV incidence in young women in rural South Africa (HPTN 068): a phase 3, randomised controlled trial. Lancet Glob Health. 2016;4(12):e978–e988. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. McKinnon LR, Karim QA. Factors driving the HIV epidemic in southern Africa. Curr HIV/AIDS Rep. 2016;13(3):158–169. [DOI] [PubMed] [Google Scholar]
  • 28. Tyler KA. Social and emotional outcomes of childhood sexual abuse: a review of recent research. Aggress Violent Behav. 2002;7(6):567–589. [Google Scholar]
  • 29. Littleton HL, Grills AE, Drum KB. Predicting risky sexual behavior in emerging adulthood: examination of a moderated mediation model among child sexual abuse and adult sexual assault victims. Violence Vict. 2014;29(6):981–998. [DOI] [PubMed] [Google Scholar]
  • 30. Nduna M, Jewkes RK, Dunkle KL, et al. . Associations between depressive symptoms, sexual behaviour and relationship characteristics: a prospective cohort study of young women and men in the Eastern Cape, South Africa. J Int AIDS Soc. 2010;13:44. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31. Pettifor AE, Measham DM, Rees HV, et al. . Sexual power and HIV risk, South Africa. Emerg Infect Dis. 2004;10(11):1996–2004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32. Hatcher AM, Tsai AC, Kumbakumba E, et al. . Sexual relationship power and depression among HIV-infected women in rural Uganda. PLoS One. 2012;7(12):e49821. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33. Hamad R, Fernald LCH, Karlan DS, et al. . Social and economic correlates of depressive symptoms and perceived stress in South African adults. J Epidemiol Community Health. 2008;62(6):538–544. [DOI] [PubMed] [Google Scholar]
  • 34. Pettifor AE, Levandowski BA, MacPhail C, et al. . Keep them in school: the importance of education as a protective factor against HIV infection among young South African women. Int J Epidemiol. 2008;37(6):1266–1273. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35. Tomita A, Burns JK. A multilevel analysis of association between neighborhood social capital and depression: evidence from the first South African National Income Dynamics Study. J Affect Disord. 2013;144(1):101–105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36. Bastien S, Kajula L, Muhwezi W. A review of studies of parent-child communication about sexuality and HIV/AIDS in sub-Saharan Africa. Reprod Health. 2011;8:25. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37. Pardini DA. Novel insights into longstanding theories of bidirectional parent–child influences: introduction to the special section. J Abnorm Child Psychol. 2008;36(5):627–631. [DOI] [PubMed] [Google Scholar]
  • 38. Bradley RH, Corwyn R. From parent to child to parent … : paths in and out of problem behavior. J Abnorm Child Psychol. 2013;41(4):515–529. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39. Spies G, Seedat S. Depression and resilience in women with HIV and early life stress: does trauma play a mediating role? A cross-sectional study. BMJ Open. 2014;4(2):e004200. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40. Olley BO, Seedat S, Stein DJ. Persistence of psychiatric disorders in a cohort of HIV/AIDS patients in South Africa: a 6-month follow-up study. J Psychosom Res. 2006;61(4):479–484. [DOI] [PubMed] [Google Scholar]
  • 41. Rabkin JG. HIV and depression: 2008 review and update. Curr HIV/AIDS Rep. 2008;5(4):163–171. [DOI] [PubMed] [Google Scholar]
  • 42. Cuijpers P, Karyotaki E, Reijnders M, et al. . Psychotherapies for depression in low- and middle-income countries: a meta-analysis. World Psychiatry. 2018;17(1):90–101. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Pettifor A, MacPhail C, Selin A, et al. . HPTN 068: a randomized control trial of a conditional cash transfer to reduce HIV infection in young women in South Africa—study design and baseline results. AIDS Behav. 2016;20(9):1863–1882. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Kovacs M. The Children’s Depression Inventory (CDI). Psychopharmacol Bull. 1985;21(4):995–998. [PubMed] [Google Scholar]
  • 45. Cluver L, Gardner F, Operario D. Psychological distress amongst AIDS-orphaned children in urban South Africa. J Child Psychol Psychiatry. 2007;48(8):755–763. [DOI] [PubMed] [Google Scholar]
  • 46. Sharer M, Cluver L, Shields J. Mental health of youth orphaned due to AIDS in South Africa: biological and supportive links to caregivers. Vulnerable Child Youth Stud. 2015;10(2):141–152. [Google Scholar]
  • 47. Cluver LD, Orkin M, Gardner F, et al. . Persisting mental health problems among AIDS-orphaned children in South Africa. J Child Psychol Psychiatry. 2012;53(4):363–370. [DOI] [PubMed] [Google Scholar]
  • 48. Binagwaho A, Fawzi MCS, Agbonyitor M, et al. . Validating the Children’s Depression Inventory in the context of Rwanda. BMC Pediatr. 2016;16:29. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49. Pulerwitz J, Gortmaker SL, DeJong W. Measuring sexual relationship power in HIV/STD research. Sex Roles. 2000;42(7–8):637–660. [Google Scholar]
  • 50. Robins J. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect. Math Model. 1986;7(9):1393–1512. [Google Scholar]
  • 51. van der Laan MJ, Robins JM. Unified Methods for Censored Longitudinal Data and Causality. New York, NY: Springer Science & Business Media; 2003. [Google Scholar]
  • 52. Tchetgen Tchetgen EJ. Inverse odds ratio-weighted estimation for causal mediation analysis. Stat Med. 2013;32(26):4567–4580. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53. Nguyen QC, Osypuk TL, Schmidt NM, et al. . Practical guidance for conducting mediation analysis with multiple mediators using inverse odds ratio weighting. Am J Epidemiol. 2015;181(5):349–356. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54. Valeri L, VanderWeele TJ. Mediation analysis allowing for exposure–mediator interactions and causal interpretation: theoretical assumptions and implementation with SAS and SPSS macros. Psychol Methods. 2013;18(2):137–150. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55. Petersen ML, Sinisi SE, van der Laan MJ. Estimation of direct causal effects. Epidemiology. 2006;17(3):276. [DOI] [PubMed] [Google Scholar]
  • 56. Rubin DB. Multiple Imputation for Nonresponse in Surveys. Hoboken, NJ: John Wiley & Sons; 2004. [Google Scholar]
  • 57. Ding P, VanderWeele TJ. Sharp sensitivity bounds for mediation under unmeasured mediator-outcome confounding. Biometrika. 2016;103(2):483–490. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58. R Core Team R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2017. [Google Scholar]
  • 59. Tomlinson M, Grimsrud AT, Stein DJ, et al. . The epidemiology of major depression in South Africa: results from the South African Stress and Health study. South Afr Med J. 2009;99(5):367–373. [PMC free article] [PubMed] [Google Scholar]
  • 60. Sarkar DNN. Barriers to condom use. Eur J Contracept Reprod Health Care. 2008;13(2):114–122. [DOI] [PubMed] [Google Scholar]
  • 61. Biddlecom A, Awusabo-Asare K, Bankole A. Role of parents in adolescent sexual activity and contraceptive use in four African countries. Int Perspect Sex Reprod Health. 2009;35(2):72–81. [DOI] [PubMed] [Google Scholar]
  • 62. Puffer ES, Green EP, Sikkema KJ, et al. . A church-based intervention for families to promote mental health and prevent HIV among adolescents in rural Kenya: results of a randomized trial. J Consult Clin Psychol. 2016;84(6):511–525. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 63. Lachman JM, Cluver LD, Boyes ME, et al. . Positive parenting for positive parents: HIV/AIDS, poverty, caregiver depression, child behavior and parenting in South Africa. AIDS Care. 2014;26(3):304–313. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 64. Meinck F, Cluver LD, Orkin FM, et al. . Pathways from family disadvantage via abusive parenting and caregiver mental health to adolescent health risks in South Africa. J Adolesc Health. 2017;60(1):57–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 65. Sherr L, Macedo A, Cluver LD, et al. . Parenting, the other oldest profession in the world—a cross-sectional study of parenting and child outcomes in South Africa and Malawi. Health Psychol Behav Med. 2017;5(1):145–165. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 66. Murphy DA, Marelich WD, Herbeck DM, et al. . Family routines and parental monitoring as protective factors among early and middle adolescents affected by maternal HIV/AIDS. Child Dev. 2009;80(6):1676–1691. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 67. Jewkes RK, Dunkle K, Nduna M, et al. . Intimate partner violence, relationship power inequity, and incidence of HIV infection in young women in South Africa: a cohort study. Lancet. 2010;376(9734):41–48. [DOI] [PubMed] [Google Scholar]
  • 68. Dunkle KL, Jewkes RK, Brown HC, et al. . Gender-based violence, relationship power, and risk of HIV infection in women attending antenatal clinics in South Africa. The Lancet. 2004;363(9419):1415–1421. [DOI] [PubMed] [Google Scholar]
  • 69. Jewkes R, Sikweyiya Y, Morrell R, et al. . The relationship between intimate partner violence, rape and HIV amongst south African men: a cross-sectional study. PLoS One. 2011;6(9):e24256. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 70. Orkin M, Boyes ME, Cluver LD, et al. . Pathways to poor educational outcomes for HIV/AIDS-affected youth in South Africa. AIDS Care. 2014;26(3):343–350. [DOI] [PubMed] [Google Scholar]
  • 71. Kenny DA, Judd CM. Power anomalies in testing mediation. Psychol Sci. 2014;25(2):334–339. [DOI] [PubMed] [Google Scholar]
  • 72. Hamdani SU, Ahmed Z, Sijbrandij M, et al. . Problem management Plus (PM+) in the management of common mental disorders in a specialized mental healthcare facility in Pakistan; study protocol for a randomized controlled trial. Int J Ment Health Syst. 2017;11(1):40, e5840. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 73. Patel V, Weiss HA, Chowdhary N, et al. . Lay health worker led intervention for depressive and anxiety disorders in India: impact on clinical and disability outcomes over 12 months. Br J Psychiatry. 2011;199(6):459–466. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 74. Turner EL, Sikander S, Bangash O, et al. . The effectiveness of the peer-delivered Thinking Healthy PLUS (THPP+) Program for maternal depression and child socioemotional development in Pakistan: study protocol for a randomized controlled trial. Trials. 2016;17(1):442. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

kwz238_Goin_Web_Material_Final

Articles from American Journal of Epidemiology are provided here courtesy of Oxford University Press

RESOURCES