The Relationship Between School Dropout and Pregnancy Among Adolescent Girls and Young Women in South Africa: A HPTN 068 Analysis

Abstract

Background. Prevention of both school dropout and teen pregnancy represent clear public health priorities for South Africa, yet their complex and potentially cyclical relationship has not been fully explored. Objective. To further understand how this relationship operates, we analyzed data from a randomized trial of young women aged 13 to 20 years enrolled in school in rural South Africa to estimate the association between pregnancy and subsequent dropout and between dropout and subsequent pregnancy. Method. We examined inverse probability (IP) of exposure-weighted survival curves for school dropout by pregnancy and for pregnancy by school dropout. We used weighted curves to calculate 1-, 2-, and 3-year risk differences and risk ratios. Additionally, we used an IP-weighted marginal structural cox model to estimate a hazard ratio (HR) for each relationship. Results. Dropout from school was associated with subsequent pregnancy (HR 3.58; 95% confidence interval [CI] [2.04, 6.28]) and pregnancy was associated with subsequent school dropout (HR 2.36; 95% CI [1.29, 4.31]). Young women who attended school but attended fewer days had a higher hazard of pregnancy than those who attended more school (HR 3.64; 95% CI [2.27, 5.84]). Conclusion. Pregnancy is both a cause and a consequence of school dropout. Consideration of school attendance and academic performance could ultimately enhance pregnancy prevention efforts in this population. Programs should be tailored differently for (1) girls who have dropped out of school, (2) those who are in school and at risk for pregnancy, and (3) those who are in school and become pregnant.

Keywords

adolescent girls and young women pregnancy school dropout South Africa

In South Africa, pregnancy among adolescent girls and young women is extremely common. A third of young women have had their first child before the age of 20 years, and roughly two thirds of births among teenage mothers are reported as unplanned or unwanted (MacPhail, Pettifor, Pascoe, & Rees, 2007; National Department of Health, Medical Research Council, & OrcMacro, 2007; Rosenberg et al., 2015). Early age at first birth has been associated with medical complications for the mother and the infant as well as adverse outcomes later in life such as reduced educational achievement and economic instability (Gibbs, Wendt, Peters, & Hogue, 2012; Panday, Makiwane, Ranchod, & Letsoalo, 2009). Teen pregnancy has been shown to lead to school dropout (Ardington, Menendez, & Mutevedzi, 2015; Clark & Mathur, 2012; M. J. Grant & Hallman, 2008; Marteleto, Lam, & Ranchhod, 2008; Timæus & Moultrie, 2015), and, conversely, school dropout among nonpregnant teens has been shown to lead to subsequent pregnancy (Manlove, 1998; Rosenberg et al., 2015).

School dropout rates in South Africa are low (<5%) between the ages of 6 and 15 years as per South African law mandating school attendance (Timæus, Simelane, & Letsoalo, 2012), but the rates begin to rise (15% to >50%) after the age of 16 years. Dropout rates are highest in poor, majority Black schools (Timæus et al., 2012) and may be exacerbated by poor quality of education characterized by low literacy and numeracy levels, resource shortages, underqualified teachers, and high failure rates (Modisaotsile, 2012). School dropout has also been associated with several negative health outcomes among young women beyond pregnancy, including increased sexual risk behaviors (Hargreaves et al., 2008; Stroeken et al., 2012) and almost three times higher risk for HIV infection (Stoner, Pettifor, et al., 2017).

Prevention of both school dropout and teen pregnancy represent clear public health priorities for South Africa, yet their complex and potentially cyclical relationship has not been fully explored. Understanding the extent to which pregnancy affects school dropout and to which school dropout affects pregnancy could guide targeted interventions or programs to prevent one or both of these outcomes. Here, we estimated the association between pregnancy and subsequent dropout and between dropout and subsequent pregnancy in a cohort of adolescent girls and young women in rural South Africa.

Method

Study Sample and Procedures

We analyzed data from the HIV Prevention Trials Network (HPTN) 068 study, a randomized trial to determine if providing a conditional cash transfer intervention would reduce young women’s risk for HIV acquisition (Pettifor, MacPhail, Hughes, et al., 2016; Pettifor, MacPhail, Selin, et al., 2016). The intervention provided a transfer of cash to young women and their households, conditional on young women attending 80% of school days. The study enrolled 2,533 young women (all Black South African) aged 13 to 20 years who were not pregnant or married at baseline and were attending high school grades 8 to 11 in the rural Bushbuckridge subdistrict of Mpumalanga province, South Africa. All young women were enrolled in school at study enrollment but were retained in the cohort if they dropped out of school during the study. Our analysis excluded young women without at least one follow-up visit after enrollment and those for whom pregnancy information was not available.

Institutional review board approval for this study was obtained from the University of North Carolina at Chapel Hill and the University of the Witwatersrand Human Research Ethics Committee as well as the Provincial Department of Health’s Research Ethics Committee.

Up to four annual assessments of the young women were conducted from 2011 to 2015. Assessments occurred at enrollment and roughly every 12 months thereafter. Each annual study visit included an Audio Computer-Assisted Self-Interview with self-reported information on pregnancy, sexual behaviors, and demographics. Additionally, information on school attendance and dropout was collected directly from high school attendance registers, which were closely monitored by both school staff and study staff to ensure accuracy. School attendance data were collected in the months of February, May, and August because these months were most representative of normal attendance due to lack of holidays/exams.

We examined the relationship between pregnancy and school dropout in both directions (with each as an exposure and the other as the outcome) using time-to-event methods to account for the different lengths of time that the girls contributed to the study.

Pregnancy and Subsequent School Dropout

To estimate the association between pregnancy and subsequent school dropout, we constructed an analytic cohort in which each young woman was followed from enrollment in continuous months until date of school dropout, or date of censoring if she moved, died, was lost to follow-up, graduated, or reached the end of the study time period. We used continuous time until the actual month of dropout between visits. A woman was considered to never have been pregnant until the first survey, including baseline, at which she reported being pregnant within the past 12 months (yes/no). After this time, she was considered “exposed” to pregnancy throughout the remainder of follow-up. We used exposure information only from the visits prior to outcome ascertainment to ensure that exposure occurred before the outcome. The outcome of school dropout was defined as the first report of school dropout in the school attendance registers for any reason.

We used inverse probability (IP) of exposure weights to estimate absolute 3-year risks as well as the 3-year risk difference, risk ratio, and hazard ratio (HR) comparing risk of dropout had all women been “exposed” to pregnancy throughout the study (e.g., had been pregnant at enrollment) with the risk if all women had never become pregnant during the study. Risks, risk differences, and risk ratios show risk at specific time points over the study period, while the HR is a measure of the effect over the entire time period.

School Dropout and Subsequent Pregnancy

To estimate the association between school dropout and pregnancy, we constructed an analytic cohort in which each young woman was followed from enrollment in yearly intervals until the follow-up visit with a reported pregnancy, or censoring if she moved, was lost to follow up, graduated, or reached the end of the study period. We used yearly intervals as pregnancy was only reported at each annual survey. A young woman was considered to be in school until the first visit after which she was recorded dropping out of school. She was then defined as having dropped out for the remainder of the study period. We also examined low attendance in school (<80% of school days vs. 80% or more) at each survey as another measure of schooling, as defined in previous studies using these data (Pettifor, MacPhail, Hughes, et al., 2016; Stoner, Pettifor, et al., 2017). The outcome was defined as the first self-reported pregnancy in the past 12 months during the study. We used time to first pregnancy as very few girls had more than one child by the end of the study (N = 29, 1.23%). Again, we used exposure information only from the visits prior to outcome ascertainment to ensure that exposure occurred before the outcome. In this case, dropout was assessed using school attendance status at two follow-up visits before the outcome to ensure no overlap. We estimated absolute 3-year risks as well as the 3-year risk difference, risk ratio, and HR comparing risk for pregnancy had all women been “exposed” to dropout throughout the study with the risk if all women had never dropped out during the study.

We did not exclude young women who had a pregnancy prior to the study but did include prior pregnancy in our weights to control for confounding. An additional sensitivity analysis was done excluding young women who had a previous pregnancy at baseline (N = 197) and, for analyses using two time periods prior, those pregnant at baseline or the first visit (N = 260). However, results remained similar, and therefore, young women with a prior pregnancy at baseline were retained in the analysis cohort (Supplemental Table S1, available in the online version of this article).

Statistical Methods

Risks under each exposure were estimated using the complement of the extended Nelson-Aalen estimator of the survival function weighted to account for time-varying confounders of each exposure-outcome relationship (Cole & Hernán, 2004; Westreich et al., 2010; Xie & Liu, 2005). These risk functions were used to calculate risk ratios and risk differences comparing exposures. Confidence intervals around weighted risks, risk differences, and risk ratios were calculated using a nonparametric bootstrap calculated from 200 full samples with replacement from the observed data.

To examine the relationship between pregnancy and school dropout, we additionally estimated an IP-weighted HR using weighted Cox proportional hazards regression with robust standard errors. To estimate the association between dropout and subsequent pregnancy (in yearly intervals), we estimated an IP-weighted HR using pooled logistic regression. Confidence intervals around the HRs were created using the robust sandwich variance estimator (Cole, Hernán, Anastos, Jamieson, & Robins, 2007).

Inverse Probability Weights for Both Relationships and Covariate Definitions

We used a causal directed acyclic graph (DAG) to identify a minimally sufficient adjustment set of covariates and identified the same time-fixed and time-varying covariates for both outcomes. Time-fixed covariates included age at baseline (categorized as 13-14, 15-16, 17-18, and 19+ years), prior pregnancy, depression, and cash transfer intervention arm (Pettifor, MacPhail, Hughes, et al., 2016). Depression was defined as a baseline children’s depression inventory score equal to or above seven (Cluver, Gardner, & Operario, 2007; Kovacs, 1985). Time-varying covariates included alcohol use more than once per month, older partner (≥5 years), wealth in quartiles according to household assets, and single or double orphan status. Based on the literature, we used ≥5 years as the uniform cutoff for all age-disparate relationships (Harling et al., 2014; Schaefer et al., 2017; Stoner et al., 2019). We used wealth quartiles rather than parental educational attainment as they were highly correlated and parental education had more missing information.

Other covariates that were descriptively examined at baseline but were not included in the minimally sufficient adjustment set in our DAG were physical violence by a partner, grade repetition, hope for the future score (Abler et al., 2017), risky sexual behaviors (any unprotected sex, transactional sex, number of partners, contraception use), any report of violence from teachers or students at school, parent/guardian beliefs that education is more important for boys than girls, and revised children’s manifest anxiety score (Boyes & Cluver, 2013).

We accounted for confounders set using stabilized IP of exposure weights (Cole & Hernán, 2008). Weights had the form

\prod_{k \leq t} \frac{P {A (k) = a (k)}}{P {A (k) = a (k) | Z k = z (k), \bar{A} (k) = \bar{a} (k)}},

where $A (t)$ represents exposure at time t, $\bar{A} (t)$ represents exposure history up to time t, and $Z (t)$ is a vector of confounders at time t. Numerators and denominators of the weights were estimated using pooled logistic regression (Cole et al., 2003; Cole & Hernán, 2008). To adjust for potentially informative censoring, we multiplied our IP exposure weights by time varying inverse probability-of-censoring weights estimated using pooled logistic regression models (Cole et al., 2003).

Results

Of the 2,533 young women included in the original randomized trial, we excluded 163 young women with no follow-up Audio Computer-Assisted Self-Interview survey after baseline and 5 who were missing pregnancy information for the duration of the study. Of the 2,365 young women in our final analytic cohort, 136 (5.7%) dropped out of school, and almost 16% (N = 372, 15.7%) had an incident pregnancy between enrollment and the end of the study period.

Young women who reported ever being pregnant at baseline (N = 197, 8.4%) were older and were more likely to be depressed, anxious, an orphan, have a lower hope for the future score, and/or ever have experienced physical violence from a partner or reported violence from teachers or students at school (Table 1). They also had worse educational outcomes—they were more likely to have ever repeated a grade and less likely to have high attendance in school. We did not find differences by randomization arm, alcohol use, or having an older partner or a belief that education was more important for boys than girls.

Table 1.

Baseline Characteristics of Young Women Aged 13 to 20 Years by Ever Been Pregnant in Agincourt, South Africa, from March 2011 to December 2012 (N = 2365).^a

Baseline characteristic	Ever been pregnant at baseline
	No (n = 2146, 91.6%)		Yes (n = 197, 8.4%)		Total (N = 2,343)
	n (%)	Mdn (IQR)	n (%)	Mdn (IQR)	n (%)	Mdn (IQR)
Young women’s age at baseline (years)		15 (14, 16)		17 (16, 18)		15 (14, 17)
13-14	744 (34.7)		5 (2.5)		749 (32.0)
15-16	950 (44.3)		46 (23.4)		996 (42.5)
17-18	388 (18.1)		106 (53.8)		494 (21.1)
19-20	64 (3.0)		40 (20.3)		104 (4.4)
Household wealth (assets)
Low	538 (25.1)		64 (32.5)		602 (25.7)
Middle to low	570 (26.6)		54 (27.4)		634 (26.7)
Middle	519 (24.2)		45 (22.8)		564 (24.1)
High	515 (24.0)		34 (17.3)		549 (23.5)
Randomization arm	1127 (52.5)		104 (52.8)		1231 (52.5)
Alcohol use more than once a month	42 (2.0)		5 (2.5)		47 (2.0)
Children’s depression index score		2 (1, 5)		4 (2, 8)		2 (1, 5)
Children’s depression index score ≥7	352 (17.3)		64 (32.6)		416 (18.6)
Children’s manifest anxiety score		4 (1, 7)		6 (3, 10)		4 (1, 7)
Children’s manifest anxiety score ≥7	547 (25.8)		95 (49.0)		642 (27.8)
Single or double orphan <18	415 (19.5)		56 (28.7)		471 (20.3)
Ever repeated a grade	665 (31.0)		139 (70.6)		804 (34.3)
High attendance in school (≥80%)	2057 (96.3)		180 (91.8)		2237 (95.9)
Any experience of physical violence	293 (14.0)		93 (47.2)		386 (16.8)
Violence from teachers or students at school	211 (9.8)		33 (16.8)		244 (10.4)
Belief that boys need to be educated more than girls (girl)	158 (7.4)		21 (10.7)		179 (7.6)
Belief that boys need to be educated more than girls (parent)	294 (13.7)		26 (13.2)		320 (13.7)
Hope for the future score		32 (26, 38)		26 (26, 37)		32 (26, 38)
Ever had sex	417 (19.5)		180 (91.8)		597 (25.5)
Among sexually active (N = 597)
Modern contraception use or condoms	145 (35.2)		141 (78.8)		286 (48.4)
Modern contraception use—no condoms	99 (24.0)		110 (61.4)		209 (35.4)
Any unprotected sex in the past 3 months	108 (26.5)		70 (39.8)		178 (30.5)
Transactional sex	51 (14.7)		35 (21.9)		86 (16.9)
Older partner (≥5 years)	78 (19.3)		41 (22.9)		119 (20.4)

Note. Mdn = median; IQR = interquartile range.

Missing at baseline: Ever been pregnant n = 22; school attendance n = 10; attendance n = 10; wealth n = 4; alcohol use n = 3; depression n = 107; anxiety n = 33; ever had sex n = 3; unprotected sex n = 15; single or double orphan n = 20; parental monitoring n = 40; mother’s educational level n = 191; exchange sex n = 89; intimate partner violence n = 48, older partner n = 15. Boys more educated 1; modern contraception 7.

Young women who reported having a pregnancy at baseline or during follow-up were more likely to drop out of school compared with young women who never had a pregnancy (Figure 1). The risk differences and risk ratios comparing those who were ever pregnant versus those who were never pregnant increased over the study period for a 3-year risk difference of 2.3% (95% CI [−1.5%, 6.0%]) (Table 2). The weighted hazard of school dropout for young women who ever had a pregnancy was 2.36 (95% CI [1.29, 4.31]) times the hazard in young women who never had a pregnancy (Table 3).

Figure 1.

Cumulative incidence curves for the association between pregnancy in the past 12 months and incident school dropout for 2,360 young women aged 13 to 23 years participating in HPTN 068 (2011-2015). (A) Unweighted. (B) Weighted. Treatment weighted curves accounted for the following covariates: age at baseline, used alcohol in the past 12 months, baseline depression, having any partner 5 years or older, wealth, intervention assignment, orphan status, and pregnancy at baseline.

Table 2.

Weighted Risk Differences and Risk Ratios at 1, 2, and 3 Years for the Effect of Pregnancy in Past 12 Months on Incident Dropout Among Young Women Aged 13 to 23 years, Enrolled in HPTN 068 Between 2011 and 2015 at Different Time Intervals.^a

Comparison	n	Risk	Risk difference	95% CI	Risk ratio	95% CI
Dropout
1 year
Ever pregnant	42	0.014	0.005	[−0.012, 0.021]	1.50	[0.42, 5.36]
Never pregnant	8	0.009	0	0	1	1
2 years
Ever pregnant	25	0.019	0.011	[−0.012, 0.035]	1.59	[0.58, 4.40]
Never pregnant	79	0.03	0	0	1
3 years
Ever pregnant	42	0.055	0.023	[−0.015, 0.060]	1.70	[0.82, 3.52]
Never pregnant	109	0.032	0	0	1	1
Pregnancy
1 year
Ever dropout	—	—	—	—	—	—
Never dropout	—	—	—	—	—	—
2 years
Ever dropout	13	0.116	0.052	[−0.036, 0.139]	1.80	[0.70, 4.64]
Never dropout	158	0.064	0	0	1	1
3 years
Ever dropout	32	0.372	0.217	[−0.065, 0.499]	2.41	[1.20, 4.81]
Never dropout	236	0.154	0	0	1	1

Note. HPTN = HIV Prevention Trials Network; CI = bootstrap confidence interval.

Weighted models conditioned on the following covariates: age at baseline, alcohol use, baseline depression, having any partner 5 years or older, wealth, intervention assignment, orphan status, prior pregnancy at baseline.

Table 3.

Hazard Ratios for the Effect of Pregnancy on Incident School Dropout and Dropout on Incident Pregnancy Among Young Women Aged 13 to 23 Years, Enrolled in HPTN 068 from 2011-2015.^a

	Unweighted		Weighted
Comparison	HR	95% CI	HR	95% CI
Pregnancy on dropout
Total
Ever pregnant during study	3.07	[1.91, 4.91]	2.36	[1.29, 4.31]
Never pregnant during study	1	1	1	1
Dropout on pregnancy (2 time periods later)
Total
Dropout	8.60	[5.21, 14.19]	3.58	[2.04, 6.28]
No dropout	1	1	1	1
Low attendance	5.64	[3.73, 8.53]	3.64	[2.27, 5.84]
High attendance	1	1	1	1

Note. HPTN = HIV Prevention Trials Network; CI = robust confidence intervals; HR = hazard ratio.

Young women who dropped out of school during the study period had a higher weighted risk for incident pregnancy compared with young women who never dropped out (Figure 2). The risk difference comparing those who ever dropped out versus those who never dropped out at 3 years was 21.7% (−6.5%, 49.9%) (Table 2). The weighted HR for the association between dropout and incident pregnancy was 3.58 (95% CI [2.04, 6.28]) (Table 3). Low attendance in school (<80% of school days) was also associated with a higher weighted hazard of pregnancy compared with high attendance (≥80%) in school (HR 3.64; 95% CI [2.27, 5.84]). The number of young women who became pregnant and dropped out of school was low over the study period, leading to large confidence intervals around all estimates (limited precision).

Figure 2.

Cumulative incidence curves for the association between dropout and incident pregnancy two time periods later for young women aged 13 to 23 years participating in HPTN 068 (2011-2015). (A) Unweighted. (B) Weighted. Treatment weighted curves accounted for the following covariates: time-varying age, used alcohol in the past 12 months, time-varying depression, having a partner 5 years or older, Wealth, intervention assignment, orphan status, previous pregnancy at baseline.

Discussion

In this bidirectional assessment of pregnancy and school dropout among adolescent girls and young women in rural South Africa, dropout and pregnancy were highly associated. Girls who were pregnant either before or during the study period were more likely to drop out than girls who did not become pregnant (HR 2.36; 95% CI [1.29, 4.31]), and girls who dropped out of school were more likely to experience a pregnancy compared with girls who were retained in school (HR 3.58; 95% CI [2.04, 6.28]). Even among young women who remained in school, those who attended fewer days had a higher hazard of pregnancy than those with greater attendance (HR 3.64; 95% CI [2.27, 5.84]). Together, our findings suggest that pregnancy is both a driver and a consequence of school dropout among South African adolescent girls.

For young women who are in school, risk factors for pregnancy are complex and are likely to vary by context and population (Christofides et al., 2014; M. J. Grant & Hallman, 2008; Jewkes, Morrell, & Christofides, 2009; Panday et al., 2009; Stroeken et al., 2012; Willian, 2013). In our study, we found that girls who became pregnant were more likely to drop out of school and that violence, limited future aspirations, and poor educational performance (school attendance and grade repetition) were all associated with ever having been pregnant at baseline. Gender inequity is a major driver of teen pregnancy, and girls who have experienced physical violence are more likely to become pregnant and to have an unwanted pregnancy (Christofides et al., 2014; Jewkes et al., 2009; Jewkes, Vundule, Maforah, & Jordaan, 2001). Young women who exhibit poor scholastic performance and those who have repeated grades are also more likely to get pregnant and are more likely to drop out following pregnancy (M. J. Grant & Hallman, 2008; Marteleto et al., 2008; Timæus & Moultrie, 2013).

Inconsistent and incorrect use of effective contraception likely accounts for a nontrivial proportion of teen pregnancy in South Africa. More than half of sexually active young women in South Africa report that they do not use contraception to prevent pregnancy (MacPhail et al., 2007). In our study, girls who experienced a prior pregnancy were significantly more likely to be using modern contraception at baseline compared with girls who had never been pregnant. Poor contraceptive uptake is associated with lack of adolescent friendly health services and purported stigma and discrimination by health care workers, friends, families, and other institutions (MacPhail et al., 2007; Panday et al., 2009; Zwang & Garenne, 2008). Furthermore, some studies have shown that along with this stigma, societal norms preclude girls from engaging with sexual health and contraceptive services until after they experience their first pregnancy (MacPhail et al., 2007). Therefore, interventions to prevent pregnancy in girls who are in school should include components to address gender inequity (i.e., gender-based violence) and provide comprehensive sexual health education and contraception services.

Young women who become pregnant and leave school infrequently return to school following childbirth (Jewkes et al., 2009; Willian, 2013), and interventions are needed to promote continued enrollment or reentry after pregnancy. Some studies have shown that young women who have performed well academically are more likely to return to school, further highlighting the need to intervene early in order to improve school performance before young women drop out (Ardington et al., 2015; M. J. Grant & Hallman, 2008). In our study population, only 14% of young women who dropped out for any reason eventually returned to school (Stoner, Edwards, et al., 2017). South African law mandates that pregnant girls be allowed to stay in school during pregnancy and return following childbirth. However, these policies are often not implemented due to attitudes of school staff and peers. Furthermore, limited resources among families, including lack of childcare, may prevent girls from returning to school (Jewkes et al., 2009; Panday et al., 2009; Willian, 2013). At present, South Africa does not provide childcare services for young women with children returning to school following pregnancy. Subsequently, efforts that encourage girls to return and remain in school following childbirth are critical. Future interventions or programs should focus on improving indicators of schooling before dropout occurs; building support among school staff, families, and peers; and providing services to support the parenting needs of girls returning to school.

Few studies have examined the relationship between dropout and subsequent pregnancy or considered interventions to prevent pregnancy in adolescents who have dropped out of school (Manlove, 1998; Rosenberg et al., 2015). Our results are similar to another study in the same region of South Africa, which found that girls who were enrolled in school had a lower hazard of pregnancy than girls who had dropped out, and that pregnancy rates were lower during school term than summer holiday (Rosenberg et al., 2015). School attendance may prevent pregnancy by providing periods of structure and supervision in young adults’ lives, which reduce opportunities for sexual activity (Barnes, Hoffman, Welte, Farrell, & Dintcheff, 2007; Rosenberg et al., 2015; Stoner, Edwards, et al., 2017). Similarly, young women out of school in settings where unemployment and poverty are high may be ambivalent about preventing pregnancy or encouraged to have children in order to increase their social status (K. Grant et al., 2002; Jewkes et al., 2009). Therefore, interventions to prevent pregnancy among adolescents out of school should help girls return to school and/or provide an environment where they are engaged in meaningful activities and supported by peers.

There are some limitations to this analysis. First, information on pregnancy in our study was self-reported and may be underreported. Young women who misreport pregnancy may also be more likely to misreport other sensitive information such as gender-based violence or depression. However, our assessment of school dropout was derived using school attendance registers and is likely unrelated to potential mismeasurement of other variables in this analysis. We expect our estimates to be biased toward the null and thus attenuated. Furthermore, we did not have information about the outcome of the pregnancy and would expect the association to be stronger in girls who have a child rather than those who do not carry their pregnancy to term. Third, there may be common causes for both school dropout and pregnancy or the potential for unmeasured confounding in our estimates. However, we did include the most important confounders identified by our DAG, including prior pregnancy, and explored models with other covariates. Estimates were generally similar for models with alternative adjustment sets. Fourth, we used depression information from baseline only as a different measure of depression was collected over study follow-up, and depression experience may have changed over time. Last, we did not capture the exact date of pregnancy, and we were unable to distinguish timing of pregnancies and school dropouts that occurred close together. However, our analytic framework ensures temporality for most events by defining our exposure in the time period before outcome ascertainment.

Our study is unique in that we use longitudinal school dropout and attendance information collected directly from high schools rather than from national census data, as reported in other studies. Our information is likely more accurate as it was monitored and collected from schools during a randomized trial and we are better able to finely capture characteristics that vary over time. Additionally, we were able to incorporate temporality in our analysis and control for both individual- and household-level characteristics. We contribute stronger causal evidence for the bidirectional relationship between pregnancy and dropout.

Overall, our findings provide important insights into the relationship between pregnancy and school dropout among young women in South Africa. Furthermore, our results posit that while pregnancy is both a driver and a consequence of school dropout, consideration of school attendance and academic performance could ultimately enhance pregnancy prevention efforts in this population. Programs should be tailored differently for (1) girls who have dropped out of school, (2) those who are in school and at risk for pregnancy, and (3) those who are in school and do become pregnant. Programs to prevent pregnancy among adolescents who have dropped out of school should encourage girls to return to school and/or focus on creating supportive and structured environments. For girls in school at risk for pregnancy, prevention programs should address gender inequity, provide comprehensive sexual health education, and improve access to contraception services. For young women in school who have recently had a child, programs should build support from school staff and families or focus on improving indicators of schooling before dropout occurs.

Supplemental Material

HEB831755_Supplemental_Material – Supplemental material for The Relationship Between School Dropout and Pregnancy Among Adolescent Girls and Young Women in South Africa: A HPTN 068 Analysis

Supplemental material, HEB831755_Supplemental_Material for The Relationship Between School Dropout and Pregnancy Among Adolescent Girls and Young Women in South Africa: A HPTN 068 Analysis by Marie C. D. Stoner, Katherine B. Rucinski, Jessie K. Edwards, Amanda Selin, James P. Hughes, Jing Wang, Yaw Agyei, F. Xavier Gomez-Olive, Catherine MacPhail, Kathleen Kahn and Audrey Pettifor in Health Education & Behavior

Footnotes

Acknowledgements

We thank the HPTN 068 study team and all trial participants.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by the National Institutes of Health (R01 MH110186, 2T32AI102623-06) and by Award Numbers UM1 AI068619 (HPTN Leadership and Operations Center), UM1AI068617 (HPTN Statistical and Data Management Center), and UM1AI068613 (HPTN Laboratory Center) from the National Institute of Allergy and Infectious Diseases, the National Institute of Mental Health, and the National Institute on Drug Abuse of the National Institutes of Health. This work was also supported by NIMH R01 (R01MH087118) and the Carolina Population Center and its NIH Center (Grant P2C HD050924). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The MRC/Wits Rural Public Health and Health Transitions Research Unit and Agincourt Health and Socio-Demographic Surveillance System have been supported by the University of the Witwatersrand, the Medical Research Council, South Africa, and the Wellcome Trust, UK (Grants 058893/Z/99/A; 069683/Z/02/Z; 085477/Z/08/Z; 085477/B/08/Z).

Supplemental Material

Supplemental Table S1 is available in the online version of this article at .

References

Abler

Hill

Maman

DeVellis

Twine

Kahn

. . . Pettifor

(2017). Hope matters: Developing and validating a measure of future expectations among young women in a high HIV prevalence setting in rural South Africa (HPTN 068). AIDS and Behavior, 21, 2156-2166. doi:10.1007/s10461-016-1523-6

Ardington

Menendez

Mutevedzi

(2015). Early childbearing, human capital attainment, and mortality risk: Evidence from a longitudinal demographic surveillance area in rural KwaZulu-Natal, South Africa. Economic Development and Cultural Change, 63, 281-317.

Barnes

G. M.

Hoffman

J. H.

Welte

J. W.

Farrell

M. P.

Dintcheff

B. A.

(2007). Adolescents’ time use: Effects on substance use, delinquency and sexual activity. Journal of Youth and Adolescence, 36, 695-710. doi:10.1007/s10964-006-9075-0

Boyes

M. E.

Cluver

L. D.

(2013). Performance of the Revised Children’s Manifest Anxiety Scale in a sample of children and adolescents from poor urban communities in Cape Town. European Journal of Psychological Assessment, 29, 113-120. doi:10.1027/1015-5759/a000134

Christofides

N. J.

Jewkes

R. K.

Dunkle

K. L.

McCarty

Jama Shai

Nduna

Sterk

(2014). Risk factors for unplanned and unwanted teenage pregnancies occurring over two years of follow-up among a cohort of young South African women. Global Health Action, 7, 23719. doi:10.3402/gha.v7.23719

Clark

Mathur

(2012). Dating, sex, and schooling in Urban Kenya. Studies in Family Planning, 43, 161-174. doi:10.1111/j.1728-4465.2012.00315.x

Cluver

Gardner

Operario

(2007). Psychological distress amongst AIDS-orphaned children in urban South Africa. Journal of Child Psychology and Psychiatry, 48, 755-763. doi:10.1111/j.1469-7610.2007.01757.x

Cole

S. R.

Hernán

M. A.

(2004). Adjusted survival curves with inverse probability weights. Computer Methods and Programs in Biomedicine, 75, 45-49. doi:10.1016/j.cmpb.2003.10.004

Cole

S. R.

Hernán

M. A.

(2008). Constructing inverse probability weights for marginal structural models. American Journal of Epidemiology, 168, 656-664. doi:10.1093/aje/kwn164

10.

Cole

S. R.

Hernán

M. A.

Anastos

Jamieson

Robins

(2007). Determining the effect of highly active antiretroviral therapy on changes in human immunodeficiency virus type 1 RNA viral load using a marginal structural left-censored mean model. American Journal of Epidemiology, 166, 219-227. doi:10.1093/aje/kwm047

11.

Cole

S. R.

Hernán

M. A.

Robins

J. M.

Anastos

Chmiel

Detels

. . . Munoz

(2003). Effect of highly active antiretroviral therapy on time to acquired immunodeficiency syndrome or death using marginal structural models. American Journal of Epidemiology, 158, 687-694. doi:10.1093/aje/kwg206

12.

Gibbs

C. M.

Wendt

Peters

Hogue

C. J.

(2012). The impact of early age at first childbirth on maternal and infant health. Paediatric and Perinatal Epidemiology, 26(Suppl. 1), 259-284. doi:10.1111/j.1365-3016.2012.01290.x

13.

Grant

Grace

Trujillo

Halpert

Kessler-Cordeiro

Razzino

Davis

(2002). Predicting desire for a child among low-income urban adolescent girls: Interpersonal processes in the context of poverty. Journal of Primary Prevention, 22, 341-359. doi:10.1023/A:1015223522236

14.

Grant

M. J.

Hallman

K. K.

(2008). Pregnancy-related school dropout and prior school performance in KwaZulu-Natal, South Africa. Studies in Family Planning, 39, 369-382. doi:10.1111/j.1728-4465.2008.00181.x

15.

Hargreaves

J. R.

Morison

L. A.

Kim

J. C.

Bonell

C. P.

Porter

J. D. H.

Watts

. . . Pronyk

P. M.

(2008). The association between school attendance, HIV infection and sexual behaviour among young people in rural South Africa. Journal of Epidemiology and Community Health, 62, 113-119. doi:10.1136/jech.2006.053827

16.

Harling

Newell

Tanser

Kawachi

Subramanian

Bärnighausen

(2014). Do age-disparate relationships drive HIV incidence in young women? Evidence from a population ohort in rural KwaZulu-Natal, South Africa. Journal of Acquired Immune Deficiency Syndrome, 66, 443-451. doi:10.1097/QAI.0000000000000198

17.

Jewkes

Morrell

Christofides

(2009). Empowering teenagers to prevent pregnancy: Lessons from South Africa. Culture, Health and Sexuality, 11, 675-688. doi:10.1080/13691050902846452

18.

Jewkes

Vundule

Maforah

Jordaan

(2001). Relationship dynamics and teenage pregnancy in South Africa. Social Science & Medicine, 52, 733-744. doi:10.1016/S0277-9536(00)00177-5

19.

Kovacs

(1985). The children’s depression inventory (CDI). Psychopharmacology Bulletin, 21, 995-998. doi:10.3724/SP.J.1041.2015.01004

20.

MacPhail

Pettifor

A. E.

Pascoe

Rees

H. V.

(2007). Contraception use and pregnancy among 15-24 year old South African women: A nationally representative cross-sectional survey. BMC Medicine, 5, 31. doi:10.1186/1741-7015-5-31

21.

Manlove

(1998). The influence of high school dropout and school disengagement on the risk of school-age pregnancy. Journal of Research on Adolescence, 8, 187-220. doi:10.1207/s15327795jra0802_2

22.

Marteleto

Lam

Ranchhod

(2008). Sexual behavior, pregnancy, and schooling among young people in Urban South Africa. Studies in Family Planning, 39, 351-368. doi:10.1111/j.1728-4465.2008.00180.x

23.

Modisaotsile

B. M.

(2012, March). The failing standard of basic education in South Africa (Policy Brief No. 72). Retrieved from http://www.ai.org.za/wp-content/uploads/downloads/2012/03/No.-72.The-Failing-Standard-of-Basic-Education-in-South-Africa1.pdf

24.

National Department of Health, Medical Research Council, & OrcMacro. (2007). South Africa Demographic and Health Survey 2003. Pretoria, South Africa: Author. Retrieved from https://dhsprogram.com/pubs/pdf/FR206/FR206.pdf

25.

Panday

Makiwane

Ranchod

Letsoalo

(2009). Teenage pregnancy in South Africa: With a specific focus on school-going learners. Pretoria, South Africa: Department of BASIC education.

26.

Pettifor

MacPhail

Hughes

J. P.

Selin

Wang

Gómez-Olivé

F. X.

. . . Kahn

(2016). 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 Global Health, 4, e978-e988. doi:10.1016/S2214-109X(16)30253-4

27.

Pettifor

MacPhail

Selin

Gómez-Olivé

F. X.

Rosenberg

Wagner

R. G.

, . . . HPTN 068 protocol team. (2016). 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 and Behavior, 20, 1863-1882. doi:10.1007/s10461-015-1270-0

28.

Rosenberg

Pettifor

Miller

W. C.

Thirumurthy

Emch

Afolabi

S. A.

. . . Tollman

(2015). Relationship between school dropout and teen pregnancy among rural South African young women. International Journal of Epidemiology, 44, 928-936. doi:10.1093/ije/dyv007

29.

Schaefer

Gregson

Eaton

J. W.

Mugurungi

Rhead

Takaruza

. . . Nyamukapa

(2017). Age-disparate relationships and HIV incidence in adolescent girls and young women: Evidence from a general-population cohort in Zimbabwe. AIDS, 31, 1461-1470. doi:10.1097/QAD.0000000000001506

30.

Stoner

M. C. D.

Edwards

J. K.

Miller

W. C.

Aiello

A. E.

Halpern

C. T.

Julien

. . . Pettifor

(2017). Effect of schooling on age-disparate relationships and number of sexual partners among young women in rural South Africa enrolled in HPTN 068. Journal of Acquired Immune Deficiency Syndromes (1999), 76, e107-e114. doi:10.1097/QAI.0000000000001544

31.

Stoner

M. C. D.

Nguyen

Kilburn

Gomez-Olive

F. X.

Edwards

J. K.

Selin

. . . Pettifor

(2019). Age-disparate partnerships and incident HIV infection in adolescent girls and young women in rural South Africa: An HPTN 068 analysis. AIDS, 33, 83-91. doi:10.1097/QAD.0000000000002037

32.

Stoner

M. C. D.

Pettifor

Edwards

J. K.

Aiello

A. E.

Halpern

C. T.

Julien

. . . Kahn

(2017). The effect of school attendance and school dropout on incident HIV and HSV-2 among young women in rural South Africa enrolled in HPTN 068. AIDS, 31, 2127-2134. doi:10.1097/QAD.0000000000001584

33.

Stroeken

Remes

De Koker

Michielsen

Van Vossole

Temmerman

(2012). HIV among out-of-school youth in Eastern and Southern Africa: A review. AIDS Care, 24, 186-194. doi:10.1080/09540121.2011.596519

34.

Timæus

I. M.

Moultrie

T. A.

(2013). Teenage childbearing and dropout from school in South Africa: Evidence from a household panel. Retrieved from https://iussp.org/sites/default/files/event_call_for_papers/Teenage%20Childbearing%20Busan.pdf

35.

Timæus

I. M.

Moultrie

T. A.

(2015). Teenage childbearing and educational attainment in South Africa. Studies in Family Planning, 46, 143-160. doi:10.1111/j.1728-4465.2015.00021.x

36.

Timæus

I. M.

Simelane

Letsoalo

(2012). Poverty, race, and children’s progress at school in South Africa. Journal of Development Studies, 49, 270-284. doi:10.1080/00220388.2012.693168

37.

Westreich

Cole

S. R.

Tien

P. C.

Chmiel

J. S.

Kingsley

Funk

M. J.

. . . Jacobson

L. P.

(2010). Time scale and adjusted survival curves for marginal structural Cox models. American Journal of Epidemiology, 171, 691-700. doi:10.1093/aje/kwp418

38.

Willian

(2013). A review of teenage pregnancy in South Africa: Experiences of schooling, and knowledge and access to sexual & reproductive health services. Retrieved from http://www.hst.org.za/publications/NonHST%20Publications/Teenage%20Pregnancy%20in%20South%20Africa%20Final%2010%20May%202013.pdf

39.

Xie

Liu

(2005). Adjusted Kaplan-Meier estimator and log-rank test with inverse probability of treatment weighting for survival data. Statistics in Medicine, 24, 3089-3110. doi:10.1002/sim.2174

40.

Zwang

Garenne

. (2008). Social context of premarital fertility in rural South Africa. African Journal of Reproductive Health, 12, 98-110. doi:10.2307/25470653

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.33 MB