Abstract
In sub-Saharan Africa, adolescent girls and young women aged 15 to 24 (AGYW) experience a high risk of early and unintended pregnancy. We assessed the impact of youth-friendly health services (YFHS) on pregnancy risk among AGYW who participated in the Girl Power study. In 2016, Girl Power randomly assigned 4 government-run health centers in Lilongwe, Malawi, to provide a standard (n = 1) or youth-friendly (n = 3) model of service delivery. At 6 and 12 months, study participants (n = 250 at each health center) self-reported their current pregnancy status and received a urine pregnancy test. Because of missing pregnancy test results, we used multiple imputation to correct for outcome misclassification in self-reported pregnancy status and applied the parametric g-formula on the corrected data to estimate the effect of YFHS on the 12-month risk of pregnancy. After correcting for outcome misclassification, the risk of pregnancy under the scenario where all health centers offered YFHS was 15.8% compared to 23.2% under the scenario where all health centers offered standard of care (risk difference: –7.3%; 95% CI, –15.5% to 0.8%). Access to a model of YFHS that integrates provider training with youth-friendly clinic modifications and community outreach activities may decrease risk of pregnancy among AGYW relative to standard of care.
Keywords: young people, health services, pregnancy, Sub-Saharan Africa
Introduction
Early and unintended pregnancies are common in sub-Saharan Africa (SSA), where the age-specific fertility rate among adolescents is more than twice the global average,1 and approximately 42% of all pregnancies are unplanned.2 Pregnancy complications are a leading cause of morbidity and mortality among adolescent girls and young women aged 15 to 24 (AGYW) in SSA.3 In 2019, 1.7 million disability-adjusted life years lost and 17% of all deaths in this population were attributed to maternal conditions.3 Early and unintended pregnancy can also compromise educational attainment, reduce future economic opportunities, and negatively impact the health and well-being of children.4
High unmet need for contraception and suboptimal contraceptive utilization contribute to the high risk of early and unintended pregnancy among AGYW in SSA.5 Structural features of the health system, including provider-imposed contraceptive eligibility restrictions,6 judgmental provider attitudes,7,8 inadequate privacy and confidentiality,8,9 long wait times,10,11 and inconvienent clinic hours,11,12 are ubiquitous across the region and contribute to low levels of contraceptive use in this population. Youth-friendly health services (YFHS) that address these barriers may reduce risk of pregnancy by increasing contraceptive use.
Facility-based models of YFHS that integrate provider training with youth-friendly clinic modifications and community sensitization are recommended by the World Health Organization (WHO) as a strategy to increase the use of sexual and reproductive health (SRH) services among young people.13 While such approaches have increased acceptability and uptake of contraceptives among AGYW across SSA, little is known about their impact on pregnancy.14,15 To our knowledge, only one study has explored how access to a model of YFHS that integrated provider training, clinic modifications, and community sensitization shaped pregnancy outcomes among AGYW in a SSA context.16 This study reported a reduction in adolescent childbearing following implementation of the National Adolescent Friendly Clinic Initiative (NAFCI) in South Africa but was unable to isolate intervention effects from other contemporaneous trends.
The Girl Power study offers a unique opportunity to evaluate the effect of a facility-based model of YFHS on the risk of pregnancy among AGYW in a more controlled environment. This study, which assigned health centers to provide SRH services using a standard or youth-friendly model of service delivery, reported considerably greater uptake of contraception among AGYW with access to YFHS compared to those without.17 Using information on self-reported pregnancy status and results from pregnancy tests, we assessed whether access to YFHS also decreased the 12-month risk of pregnancy relative to standard of care.
Methods
Study design
Girl Power—a multisite study implemented in Lilongwe, Malawi, and Western Cape, South Africa (2016-2017)—was designed to compare the uptake of SRH services among AGYW under four different models of service delivery.18 Given differences in intervention design and outcome ascertainment between countries, we restricted the present analysis to Malawi, where four comparable government-run health centers were randomized to offer standard of care (SOC; n = 1) or YFHS (n = 3).
In Girl Power–Malawi, the SOC health center (clinic 1) provided free HIV testing, free contraception, and syndromic management of sexually transmitted infections in a typical, adult-oriented clinical environment. Health centers offering YFHS (clinics 2-4) provided the same set of services as clinic 1 but used a youth-friendly model of service delivery. Services at clinics 2 to 4 were integrated, provided in youth-dedicated spaces, available in the afternoons and on select Saturdays, and delivered by health care providers who received training on medical and psychosocial support for young clients. Clinics 2 to 4 also employed peer navigators to help assess SRH needs and assist with clinical navigation. In addition, a sociobehavioral intervention consisting of monthly, facilitator-led sessions with and without a cash transfer was offered at clinics 3 and 4, respectively. Since clinics served catchment areas located ≥7 km apart, risk of between-clinic crossovers was minimal.
Details on recruitment and study procedures have been published previously.18 Briefly, community leaders from each clinic’s catchment area were informed about Girl Power–Malawi before implementation. Each clinic enrolled 250 AGYW aged 15 to 24 years who were living in the clinic’s catchment area and willing to participate in a research study. At each clinic, participants were recruited using a combination of community outreach, participant referral, and self-referral. While current sexual activity was not an eligibility requirement, peer navigators conducting community outreach explained that the study would be most helpful for those in relationships, resulting in greater enrollment of AGYW who were sexually active. Participants were followed for 12 months with study visits at baseline, 6 months, and 12 months. Most study visits occurred in a private room at the clinic of enrollment; a small number of study visits at 12 months occurred in the community as part of 12-month tracing procedures. At each visit, participants received an interviewer-administered questionnaire and a small research incentive (~$2 USD). Participants were also asked to provide a urine sample for pregnancy testing at 6 and 12 months.
The Girl Power–Malawi study received approval from the University of North Carolina Institutional Review Board and the Malawi National Health Sciences Research Committee. All study participants were fully informed of the study procedures. Participants aged 18 to 24 provided informed written consent. Participants aged 15 to 17 provided informed written assent and permission by a parent, guardian, or legally authorized representative.
Analytic sample
For the present study, we restricted our analysis to the subset of participants who reported not being pregnant at the time of enrollment.
Measures
Our exposure of interest was access to YFHS, which was determined based on the clinic of enrollment. Participants who enrolled at the clinic offering SOC were classified as unexposed while participants enrolled at the three clinics offering YFHS were classified as exposed. The model of YFHS implemented in Girl Power–Malawi was designed to address barriers to care seeking among AGYW in Malawi.19 It was developed in accordance with the WHO guidance on the provision of health services to young people,20 and incorporated insights from Malawi’s Youth Friendly Health Services Training Manual,21 the Lilongwe District Adolescent Coordinator, and AGYW participating at each clinic’s youth club.
Our outcome of interest, pregnancy, was measured in two ways. At 6 and 12 months, participants reported their current pregnancy status on an interviewer-administered questionnaire. Participants who reported being pregnant at either study visit were classified as having an incident pregnancy; those who reported not being pregnant at both study visits were classified as not having an incident pregnancy. This self-reported measure of pregnancy was available for most participants and expected to be measured with error due to social desirability bias, fertility preferences, and challenges with identifying early pregnancy.
Our second measure of pregnancy was generated using results from urine pregnancy tests (UPTs) conducted at the 6- and 12-month study visits. We considered UPTs as the gold-standard test to confirm pregnancy status.22 Participants with a positive UPT at either study visit were classified as having an incident pregnancy, while participants who tested negative on a UPT at both study visits were classified as not having an incident pregnancy. UPTs were inconsistently administered at study visits due to site operational challenges, including inconsistent access to bathrooms and busy providers. Additionally, UPTs were not conducted if the 12-month study visit occurred in the community. As a result, our UPT-based measure of pregnancy was only available for 52% of participants.
Although Girl Power–Malawi randomly assigned health centers to different models of service delivery, a previous analysis reported imbalances in the distribution of baseline covariates between AGYW who enrolled at the SOC clinic and those enrolled at the YFHS clinics.17 We constructed a causal diagram to identify potential confounders of the relationship between access to YFHS and pregnancy. The minimally sufficient adjustment set for this analysis included baseline measures of age, socioeconomic status (constructed using principal components analysis on durable good ownership and presence of utilities in home),23 school enrollment, multiple partners in the past year, marital status, living children, history of using nonbarrier contraception, and perceived risk of pregnancy in the next year. Continuous covariates—age and socioeconomic status—were modeled using restricted cubic splines with knots at the 5th, 35th, 65th, and 95th percentiles.24 All other covariates were structured as dichotomous variables.
Analytic approach
Because our gold-standard measure of pregnancy was missing for nearly half of participants, we used a two-step approach to estimate the effect of YFHS on the 12-month risk of pregnancy. First, we used multiple imputation for measurement error (MIME) to correct for potential outcome misclassification in self-reported pregnancy.25,26 Following this correction, we applied the parametric g-formula on the corrected data to adjust for confounding and estimate effects of interest.27 We also applied the parametric g-formula directly on observed data to estimate the effect of YFHS on pregnancy using our self-reported measure of pregnancy and our UPT-based measure of pregnancy. These results were contrasted with results from our measurement error–corrected analysis to illustrate the impact of MIME on our inferences. The measurement error correction and parametric g-formula are described in detail below.
Measurement error correction: While most participants had our self-reported measure of pregnancy observed, only a subset of these participants also had our UPT-based measure of pregnancy observed. This formed a validation subgroup in which our possibly misclassified, self-reported measure of pregnancy (
) could be related to our UPT-based measure of pregnancy (
) and facilitated the use of multiple imputation to correct for outcome misclassification.25,26 To implement the MIME correction, we used a logistic regression model with Firth’s correction to estimate the probability of 
given 
and 
(our exposure: access to YFHS) in the validation subgroup. The model for the predictive values included an interaction term between 
and 
to flexibly model outcome misclassification between exposure groups and all variables from our minimally sufficient adjustment set described above (
; equation (1)).
 |
(1) |
We drew 40 sets of regression coefficients from the resulting posterior predictive distribution of parameters and used these values to impute the MIME-corrected outcome, 
, where 
indexes each of the 
imputations. For participants with our UPT-based measure of pregnancy observed, 
for all imputations. For participants with only our self-reported measure of pregnancy observed, 
was imputed based on a random draw from a Bernoulli distribution with probability, 
, generated using regression coefficients estimated in the validation subgroup from the kth draw (equation (2)).
 |
(2) |
Within each imputed data set, we used the parametric g-formula to adjust for confounding and estimate risk differences (RDs) and risk ratios (RRs) comparing the 12-month risk of pregnancy between participants with and without access to YFHS (equations (3) and (4)).
 |
(3) |
 |
(4) |
G-formula: The parametric g-formula is a powerful model-based form of standardization used to estimate population-level causal effects in the presence of confounding.27 This is achieved by modeling the outcome in the observed data and then using the relationships between the exposure, covariates, and outcome in the observed data to predict the probability of the outcome for each participant under all exposure scenarios. Averaging predicted probabilities according to exposure scenario produces standardized means that are weighted to reflect the distribution of covariates in the overall study population. This facilitates the estimation of marginal effects that are adjusted for confounding. Under assumptions of consistency, conditional exchangeability, positivity, and no model misspecification,28 the effects generated by the parametric g-formula can be interpreted as the population average causal effects.
To implement the parametric g-formula in our setting,27 we used a logistic regression model to predict the probability of pregnancy for each participant under a scenario where all clinics offered YFHS and a scenario where all clinics offered SOC. Our prediction model adjusted for the minimally sufficient adjustment set and interaction terms between access to YFHS and age, marital status, living children, and history of using nonbarrier methods of contraception. We averaged predicted probabilities according to exposure scenario and used these averages to estimate marginal RDs and RRs corrected for confounding. Confidence intervals (CIs) were constructed as the effect estimate ±1.96 times the standard deviation from 200 bootstrap resamples. For our MIME-corrected analysis, we incorporated multiple imputation into our bootstrap using the Boot-MI algorithm, with 40 imputed datasets per resample.29
To ascertain whether the effect of YFHS differed according to age, we averaged predicted probabilities according to exposure status and age group (AGYW aged 15-19 and AGYW aged 20-24) and compared stratum-specific effect estimates and CIs.
Sensitivity analysis
To isolate the effect of YFHS from potential effects of the sociobehavioral intervention and cash transfer, we restricted our analysis to only AGYW enrolled at clinics 1 and 2 and implemented the same analysis described above.
All statistical analyses were conducted with SAS version 9.4 software (SAS Institute).
Results
Of the 1000 AGYW who enrolled in Girl Power–Malawi, 962 reported not being pregnant at enrollment. Nineteen participants (2%) were excluded due to missing baseline covariate information, resulting in an analytic sample of 943. Nearly all participants had experienced sexual debut (>99%), and 75% reported sexual activity in the 30 days before enrollment. Participants who enrolled at YFHS clinics were older, more likely to be married, more likely to have a child, more likely to have used nonbarrier methods of contraception, and less likely to believe they would become pregnant in the next year than those who enrolled at the SOC clinic (Table 1).
Table 1.
Baseline characteristics of nonpregnant adolescent girls and young women enrolled in Girl Power–Malawi.
| YFHS (n = 707), No. (%) | SOC (n = 236), No. (%) | ||
|---|---|---|---|
| Age | 15-17 years | 154 (22) | 117 (50) |
| 18-20 years | 310 (44) | 67 (28) | |
| 21-24 years | 243 (34) | 52 (22) | |
| School enrollment | Completed 12 years of school or enrolled in school | 419 (59) | 167 (71) |
| Completed less than 12 years of school and not enrolled in school | 288 (41) | 69 (29) | |
| Socioeconomic status | Quartile 4 (highest SES) | 196 (28) | 40 (17) |
| Quartile 3 | 162 (23) | 80 (34) | |
| Quartile 2 | 175 (25) | 61 (26) | |
| Quartile 1 (lowest SES) | 172 (24) | 55 (23) | |
| Number of partners in the past 12 months | Zero partners | 46 (7) | 16 (7) |
| One partner | 530 (75) | 154 (65) | |
| More than 1 partner | 131 (19) | 66 (28) | |
| Sexually active in the past 30 days | Yes | 525 (74) | 174 (74) |
| History of risky alcohol consumption | Yes | 92 (13) | 34 (14) |
| Currently married | Yes | 156 (22) | 30 (13) |
| Ever pregnanta | Yes | 312 (44) | 70 (30) |
| Any living children | Yes | 279 (39) | 63 (27) |
| Perceived chance of pregnancy in next year | No chance | 583 (82) | 101 (43) |
| Some chance | 124 (18) | 135 (57) | |
| History of using nonbarrier methods of contraceptionb | Never used nonbarrier methods | 414 (59) | 170 (72) |
| Prior or current user of nonbarrier methods | 293 (41) | 66 (28) | |
| History of using condoms | Never used condoms | 129 (18) | 45 (19) |
| Prior or current user of condoms | 578 (82) | 191 (81) | |
Abbreviations: SES, socioeconomic status; SOC, standard of care; YFHS, youth-friendly health services.
a Missing for two participants.
b Includes oral contraceptive pills, injectables, implants, and intrauterine devices.
Overall, 794 (84.2%) participants completed a 6-month visit and 818 (86.7%) participants completed a 12-month visit. Retention was similar between AGYW who enrolled at YFHS clinics and those enrolled at the SOC clinic at 6 months (84% vs 85%) and 12 months (88% vs 84%).
In total, 248 participants (26.2%) had pregnancy measured by self-report only, 55 (5.8%) by pregnancy test only, and 434 (46.0%) by both self-report and pregnancy testing. Of the 682 participants with our self-reported measure of pregnancy observed, 119 (17.4%) reported being currently pregnant at either the 6- or 12-month study visit. Participants who enrolled at a YFHS clinic were somewhat less likely to report a current pregnancy (85 of 513; 16.6%) than participants who enrolled at the SOC clinic (34 of 169; 20.1%). Participants at YFHS clinics were also somewhat less likely to test positive for pregnancy at the 6- or 12-month study visit (55 of 355; 15.5%) than their counterparts at the SOC clinic (23 of 134; 17.2%).
Among the 434 participants for whom both measures of pregnancy were available (i.e., the validation subgroup), the diagnostic accuracy of self-reported pregnancy was reasonably high and comparable between exposure groups (Table 2). Within this subgroup, the expected 12-month risk of pregnancy (self-reported) was similar under both models of service delivery (YFHS: 18.8% vs SOC: 19.4%; RD: –0.5%; 95% CI, –10.0% to 8.9%). Estimated risks were similar when pregnancy was measured using results from UPTs (YFHS: 17.2% vs SOC: 19.3%; RD: –2.1%; 95% CI, –11.4% to 7.1%). Notably, there were differences in the distribution of baseline covariates between members of the validation subgroup and those who only had our self-reported measure of pregnancy (Table S1).
Table 2.
Diagnostic accuracy of self-reported pregnancy.
| YFHS (n = 315) | SOC (n = 119) | |
|---|---|---|
| Did not report a current pregnancy | 259 | 98 |
| Did not test positive on a urine pregnancy test | 254 | 96 |
| Tested positive on a urine pregnancy test | 5 | 2 |
| Reported a current pregnancy | 56 | 21 |
| Did not test positive on a urine pregnancy test | 9 | 3 |
| Tested positive on a urine pregnancy test | 47 | 18 |
| Sensitivity | 0.90 (95% CI, 0.82-0.98) | 0.90 (95% CI, 0.77-1.00) |
| Specificity | 0.97 (95% CI, 0.94-0.99) | 0.97 (95% CI, 0.94-1.00) |
| Positive predictive value | 0.84 (95% CI, 0.74-0.94) | 0.86 (95% CI, 0.71-1.00) |
| Negative predictive value | (95% CI, 0.96-1.00) | 0.98 (95% CI, 0.95-1.00) |
Abbreviations: CI, confidence interval; SOC, standard of care; YFHS, youth-friendly health services.
In the full analytic sample, the expected 12-month risk of pregnancy (self-reported) under the scenario where all clinics offered YFHS was 17.7% (95% CI: 14.2% to 21.3%) compared to 21.6% (95% CI: 14.9% to 28.2%) under the scenario where all clinics offered SOC (RD: –3.8%; 95% CI, –11.8% to 4.2%). The effect of YFHS was strengthened when pregnancy was measured using results from UPTs (16.7% vs 21.9%; RD: –5.2%; 95% CI, –14.4% to 4.0%). After correcting for outcome misclassification, the expected 12-month risk of pregnancy under the scenario where all clinics offered YFHS was 15.8% (95% CI: 12.5% to 19.2%) compared to 23.2% (95% CI: 16.0% to 30.4%) under the scenario where all clinics offered SOC (RD: –7.3%; 95% CI, –15.5% to 0.8%). This represents a 30% decrease in the 12-month risk of pregnancy relative to the standard model of service delivery (RR: 0.7; 95% CI, 0.5 to 1.0; Table 3). Regardless of how pregnancy was measured, the effect of YFHS was smaller among adolescent girls aged 15 to 19 than among young women aged 20 to 24, although stratum-specific confidence intervals overlapped considerably (Figure 1).
Table 3.
Impact of the youth-friendly health service intervention on the 12-month risk of pregnancy among adolescent girls and young women enrolled in Girl Power–Malawi.
| Validation subgroup (n = 434) | Full study population (n = 943) | |||||||
|---|---|---|---|---|---|---|---|---|
| Risk (YFHS), % | Risk (SOC), % | Risk difference, % | Risk ratio | Risk (YFHS), % | Risk (SOC), % | Risk difference, % | Risk ratio | |
| Pregnancy (MIME-corrected) | . | . | . | . | 15.8 (12.5 to 19.2) | 23.2 (16.0 to 30.4) | –7.3 (–15.5 to 0.8) | 0.7 (0.5 to 1.0) |
| Pregnancy (self-report) | 18.8 (14.3 to 23.3) | 19.4 (11.7 to 27.0) | –0.5 (–10.0 to 8.9) | 1.0 (0.6 to 1.6) | 17.7 (14.2 to 21.3) | 21.6 (14.9 to 28.2) | –3.8 (–11.8 to 4.2) | 0.8 (0.6 to 1.2) |
| Pregnancy (UPT) | 17.2 (13.0 to 21.4) | 19.3 (11.4 to 27.3) | –2.1 (–11.4 to 7.1) | 0.9 (0.5 to 1.5) | 16.7 (12.9 to 20.4) | 21.9 (13.9 to 29.8) | –5.2 (–14.4 to 4.0) | 0.8 (0.5 to 1.2) |
Abbreviations: MIME, multiple imputation for measurement error; SOC, standard of care; UPT, urine pregnancy test; YFHS, youth-friendly health services.
Prediction models adjusted for access of YFHS (dichotomous; clinics 2-4 vs clinic 1), age (continuous; years), socioeconomic status (continuous; score based on durable goods ownership and presence of utilities in home), school enrollment (dichotomous; 12+ years of education completed or < 12 years of education completed and enrolled in school vs < 12 years of education completed and not enrolled in school), multiple partnerships (dichotomous; 2+ partners in past 12 months vs < 2 partners in past 12 months), marital status (dichotomous; married vs not married), number of living children (dichotomous; at least 1 living child vs no living children), history of using nonbarrier methods contraception (dichotomous; prior or current use of oral contraceptive pills, injectables, implants, or intrauterine devices vs never used oral contraceptive pills, injectables, implants, or intrauterine devices), perceived chance of pregnancy (dichotomous; some chance of pregnancy vs no chance of pregnancy), and interaction terms between access to YFHS (our exposure) and age, marital status, history of using nonbarrier methods of contraception, and number of living children.
Figure 1.
Effect of the youth-friendly health service intervention on the 12-month risk of pregnancy among adolescent girls and young women enrolled in Girl Power–Malawi, according to age at enrollment. SOC, standard of care; UPT, urine pregnancy tests; YFHS, youth-friendly health services.
Restricting our analytic sample to only AGYW enrolled at clinics 1 and 2 returned a smaller effect, although estimates were considerably less precise due to the smaller sample size. After correcting for outcome misclassification, the 12-month risk of pregnancy under the scenario where both clinics provided YFHS was 18.4% (95% CI: 11.7% to 25.0%) compared to 24.2% (95% CI: 16.4% to 32.0%) under the scenario where both clinics offered SOC (RD: –5.8%; 95% CI, –16.7% to 5.0%; RR: 0.8; 95% CI, 0.5 to 1.3).
Discussion
In this secondary analysis, we evaluated the effect of facility-based YFHS on the risk of pregnancy among AGYW in Lilongwe, Malawi. The 12-month risk of pregnancy among AGYW enrolled in Girl Power–Malawi was high. Our findings were generally consistent with a decrease in the 12-month risk of pregnancy when AGYW were offered a model of YFHS that included provider training and clinic modifications relative to SOC. Our results also demonstrated that self-reported pregnancy is prone to measurement error and may introduce bias into analyses.
Overall, we observed trends toward decreased pregnancy risk among AGYW in SSA when clinics offered SRH services using a youth-friendly compared to a standard model of service delivery. Prior evidence regarding the effect of YFHS in SSA has largely focused on the uptake of SRH services and acceptability and uptake of contraception among AGYW.14,15 One observational study from South Africa did find that living within one kilometer of an NAFCI-accredited clinic during adolescence resulted in an 8-percentage point reduction in the risk of childbirth by age 18.16 However, since the NAFCI program was implemented during a period of substantial investment in the health care system,30 observed declines may be attributed to more than just the presence of NAFCI-accredited clinics.31‑33 By capitalizing on a controlled environment with a well-defined intervention that enhanced care-seeking experiences relative to standard of care,34 our findings provide more direct evidence that expanding access to YFHS likely reduces the risk of pregnancy among AGYW in SSA.
Such evidence is timely in light of DREAMS (Determined, Resilient, Empowered, AIDS-free, Mentored, and Safe), a multibillion-dollar initiative designed to reduce the risk of HIV, pregnancy, and violence among AGYW in 14 SSA countries.35 To increase uptake of SRH services, DREAMS supports the establishment and expansion of YFHS within the existing health care infrastructure.35 Our results suggest this component of DREAMS may decrease pregnancy risk, although the impact may differ according to the model of YFHS implemented and the availability of contraception. Siloed funding mechanisms have contributed to frequent stockouts of contraception within DREAMS,36 which we expect would limit overall impact. Additionally, the impact of YFHS may differ according to the populations targeted by the intervention. Our results reflect the effect of YFHS on pregnancy risk in Girl Power–Malawi, a study that enrolled AGYW residing in urban and periurban settings in central Malawi. The transportability of these results depends on the extent to which participants in Girl Power–Malawi represent populations targeted with YFHS.37
Under the current SOC, we estimate that nearly one in four AGYW would have become pregnant over follow-up. This is likely an underestimate of incidence owing to pregnancy losses between study visits,38 which we expect would be similar between exposure groups. While our findings are imprecise, they suggest that expanding access to the Girl Power–Malawi model of YFHS could decrease pregnancy risk by approximately 30% among sexually active AGYW residing in urban and periurban settings in Malawi. A decrease in pregnancy risk of this magnitude is meaningful. However, access to YFHS did not eliminate adolescent pregnancy. Moreover, prior work in Malawi suggests that > 50% of pregnancies are unintended,38,39 which exceeds the estimated interventional effect. As only 54% of Girl Power–Malawi participants with access to YFHS received hormonal methods of contraception during follow-up,17 future research should consider layering interventions that generate demand for—and agency to use—contraception, alongside YFHS.40 Vocational and life-skills training, socioeconomic asset building, educational subsidies, and community mobilization have increased contraceptive use among AGYW in some SSA contexts.41,42 Aligning combination strategies with local contextual factors may have an even greater influence on early and unintended pregnancy.
Whereas existing evidence on interventions to prevent pregnancy among AGYW in SSA has relied on self-reported measures of pregnancy,41‑43 Girl Power–Malawi collected self-reported and UPT-based measures of pregnancy. These measures had complementary strengths and weaknesses. Our self-reported measure was available for more participants, but our UPT-based measure was a more valid measure of current pregnancy status. To maximize precision and minimize bias, we capitalized on the presence of an internal validation subgroup in which our self-reported measure of pregnancy could be related to our UPT-based measure and used multiple imputation to account for outcome misclassification in self-reported pregnancy.25,26
The MIME correction assumes that the gold-standard measure of the outcome (i.e., our UPT-based measure of pregnancy) is missing at random conditional on observed covariates. We observed important differences in the distributions of baseline covariates between AGYW included in the validation subgroup and those with only a self-reported measure of pregnancy recorded. Specifically, those in the validation subgroup were younger, less likely to report any living children, and more likely to report never having used nonbarrier methods of contraception than those with only a self-reported measure of pregnancy recorded. We therefore included all measured predictors of inclusion in the validation subgroup in our imputation model, and results were robust to the addition of other measured sociodemographic and behavioral characteristics. Our results illustrate that errors in self-reported pregnancy can introduce bias into analyses, a finding with important implications for future research.
This secondary analysis used data from a study designed to compare uptake of SRH services under different models of service delivery. The small number of clinics in the parent study limited our ability to account for cluster-level effects, and the relatively small number of participants enrolled at each clinic contributed to imbalances in baseline covariates between exposure groups. While we included all measured confounders in our analysis, unmeasured and residual confounding remain possible. Additionally, as the parent study was not powered to definitively detect differences in the risk of pregnancy between exposure groups, our estimates lacked precision; nevertheless, results demonstrate clear trends toward a decrease in pregnancy risk. Finally, because pregnancy tests were not administered at enrollment, prevalent pregnancies may have been misclassified as incident pregnancies. Given differences in the distribution of age and marital status at baseline, we expect this risk is somewhat higher among AGYW enrolled at clinics offering YFHS than those enrolled at the SOC clinic, biasing effect estimates toward the null.
Preventing early and unintended pregnancy is an urgent public health priority in SSA that is expected to bring a triple dividend of benefit: for the immediate health of AGYW today, for the future health of the adults they will become, and for the health of the next generation. Facility-based models of YFHS that integrate provider training and youth-friendly clinic modifications are effective for delivering SRH services to young people.14 In this evaluation, we show that access to such models of YFHS may also decrease the risk of pregnancy relative to standard models of service delivery in SSA. Adopting, expanding, and sustaining access to high-quality, facility-based models of YFHS is an important step toward protecting—and promoting—health and well-being within this large and growing population in the region.
Supplementary material
Supplementary material is available at the American Journal of Epidemiology online.
Supplementary Material
Contributor Information
Lauren A Graybill, Institute for Global Health and Infectious Diseases, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, United States.
Daniel Westreich, Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, United States.
Bertha Maseko, Clinical Research Programme, Malawi-Liverpool-Wellcome Trust, Blantyre 312225, Malawi.
Twambilile Phanga, UNC Project-Malawi, University of North Carolina at Chapel Hill, Lilongwe 207233, Malawi.
Tiyamike Nthani, UNC Project-Malawi, University of North Carolina at Chapel Hill, Lilongwe 207233, Malawi.
Dhrutika Vansia, UNC Project-Malawi, University of North Carolina at Chapel Hill, Lilongwe 207233, Malawi.
Benjamin H Chi, Department of Obstetrics and Gynecology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, United States.
Julie L Daniels, Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, United States.
Jennifer H Tang, Department of Obstetrics and Gynecology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, United States.
Linda-Gail Bekker, Desmond Tutu HIV Centre, University of Cape Town, Cape Town 7925, South Africa.
Audrey E Pettifor, Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, United States.
Nora E Rosenberg, Department of Health Behavior, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, United States.
Funding
This work was supported by Evidence for HIV Prevention in Southern Africa (EHPSA), a Department for International Development program managed by Mott MacDonald. Additional investigator support was provided by the National Institute of Allergy and Infectious Diseases at the National Institutes of Health (R01AI131060, K24AI120796, and T32AI007001) and the National Institute of Mental Health at the National Institutes of Health (R21MH125705).
Conflict of interest
L.A.G. and B.H.C. received consultancy fees from UNICEF. All other authors reported no conflicts of interest.
Disclaimer
The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.
Data availability
Deidentified data underlying this article will be shared on reasonable request to Dr Nora E. Rosenberg. Approval of requests for deidentified data will require a data use agreement.
References
- 1. United Nations, Department of Economic and Social Affairs, Population Division . World Population Prospects (2022); 2022: Data Sources. (UN DESA/POP/2022). [Google Scholar]
- 2. Bearak J, Popinchalk A, Ganatra B, et al. Unintended pregnancy and abortion by income, region, and the legal status of abortion: estimates from a comprehensive model for 1990-2019. Lancet Glob Health. 2020;8(9):e1152–e1161. 10.1016/S2214-109X(20)30315-6 [DOI] [PubMed] [Google Scholar]
- 3. Global Burden of Disease Collaborative Network . Global Burden of Disease Study 2019 Results. Institute for Health Metrics and Evaluation; 2020. [Google Scholar]
- 4. Patton GC, Sawyer SM, Santelli JS, et al. Our future: a Lancet commission on adolescent health and wellbeing. Lancet. 2016;387(10036):2423–2478. 10.1016/S0140-6736(16)00579-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Sully E, Biddlecom A, Darroch J, et al. Adding It Up: Investing in Sexual and Reproductive Health 2019. Guttmacher Institute; 2020. [Google Scholar]
- 6. Solo J, Festin M. Provider bias in family planning services: a review of its meaning and manifestations. Glob Health Sci Pract. 2019;7(3):371–385. 10.9745/GHSP-D-19-00130 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Wood K, Jewkes R. Blood blockages and scolding nurses: barriers to adolescent contraceptive use in South Africa. Reprod Health Matters. 2006;14(27):109–118. 10.1016/S0968-8080(06)27231-8 [DOI] [PubMed] [Google Scholar]
- 8. Chandra-Mouli V, Lenz C, Adebayo E, et al. A systematic review of the use of adolescent mystery clients in assessing the adolescent friendliness of health services in high, middle, and low-income countries. Glob Health Action. 2018;11(1):1536412. 10.1080/16549716.2018.1536412 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Self A, Chipokosa S, Misomali A, et al. Youth accessing reproductive health services in Malawi: drivers, barriers, and suggestions from the perspectives of youth and parents. Reprod Health. 2018;15(1):108. 10.1186/s12978-018-0549-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Edwards PV, Roberts ST, Chelwa N, et al. Perspectives of adolescent girls and young women on optimizing youth-friendly HIV and sexual and reproductive health care in Zambia. Front Glob Womens Health. 2021;2:723620. 10.3389/fgwh.2021.723620 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Abuosi AA, Anaba EA. Barriers on access to and use of adolescent health services in Ghana. J Health Res. 2019;33(3):197–207. 10.1108/JHR-10-2018-0119 [DOI] [Google Scholar]
- 12. Nmadu AG, Mohammed S, Usman NO. Barriers to adolescents' access and utilisation of reproductive health services in a community in north-western Nigeria: a qualitative exploratory study in primary care. Afr J Prim Health Care Fam Med. 2020;12(1):e1–e5. 10.4102/phcfm.v12i1.2307 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Dick B, Ferguson BJ, Chandra-Mouli V, et al. A review of the evidence for interventions to increase young people’s use of health services in developing countries. In: Ross DA, Dick B, Ferguson BJ, eds. Preventing HIV/AIDS in Young People. World Health Organization; 2006. [PubMed] [Google Scholar]
- 14. Denno DM, Hoopes AJ, Chandra-Mouli V. Effective strategies to provide adolescent sexual and reproductive health services and to increase demand and community support. J Adolesc Health. 2015;56(suppl 1):S22–S41. 10.1016/j.jadohealth.2014.09.012 [DOI] [PubMed] [Google Scholar]
- 15. Mazur A, Brindis CD, Decker MJ. Assessing youth-friendly sexual and reproductive health services: a systematic review. BMC Health Serv Res. 2018;18(1):216. 10.1186/s12913-018-2982-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Branson N, Byker T. Causes and consequences of teen childbearing: evidence from a reproductive health intervention in South Africa. J Health Econ. 2018;57:221–235. 10.1016/j.jhealeco.2017.11.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Rosenberg NE, Bhushan NL, Vansia D, et al. Comparing youth-friendly health services to the standard of care through "Girl Power-Malawi": a quasi-experimental cohort study. J Acquir Immune Defic Syndr. 2018;79(4):458–466. 10.1097/QAI.0000000000001830 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Rosenberg NE, Pettifor AE, Myers L, et al. Comparing four service delivery models for adolescent girls and young women through the 'Girl Power' study: protocol for a multisite quasi-experimental cohort study. BMJ Open. 2017;7(12):e018480. 10.1136/bmjopen-2017-018480 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Evidence to Action Project. Evaluation of Youth-Friendly Health Services in Malawi. United States Agency for International Development; 2014. [Google Scholar]
- 20. World Health Organization . Making Health Services Adolescent Friendly. World Health Organization; 2012. [Google Scholar]
- 21. Ministry of Health, Republic of Malawi . Youth Friendly Health Services Training Manual. Ministry of Health; 2009. [Google Scholar]
- 22. Ministry of Health, Republic of Malawi . Malawi National Reproductive Health Service Delivery Guidelines: 2014-2019. Retrieved from: https://scorecard.prb.org/wp-content/uploads/2022/03/Malawi-National-Reproductive-Health-Service-Delivery-Guidelines-2014-2019.pdf
- 23. Vyas S, Kumaranayake L. Constructing socio-economic status indices: how to use principal components analysis. Health Policy Plan. 2006;21(6):459–468. 10.1093/heapol/czl029 [DOI] [PubMed] [Google Scholar]
- 24. Howe CJ, Cole SR, Westreich DJ, et al. Splines for trend analysis and continuous confounder control. Epidemiology. 2011;22(6):874–875. 10.1097/EDE.0b013e31823029dd [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Edwards JK, Cole SR, Troester MA, et al. Accounting for misclassified outcomes in binary regression models using multiple imputation with internal validation data. Am J Epidemiol. 2013;177(9):904–912. 10.1093/aje/kws340 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Cole SR, Chu H, Greenland S. Multiple-imputation for measurement-error correction. Int J Epidemiol. 2006;35(4):1074–1081. 10.1093/ije/dyl097 [DOI] [PubMed] [Google Scholar]
- 27. Snowden JM, Rose S, Mortimer KM. Implementation of G-computation on a simulated data set: demonstration of a causal inference technique. Am J Epidemiol. 2011;173(7):731–738. 10.1093/aje/kwq472 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Hernán M, Robins J. Causal Inference: What If. Chapman & Hall/CRC; 2020. [Google Scholar]
- 29. Schomaker M, Heumann C. Bootstrap inference when using multiple imputation. Stat Med. 2018;37(14):2252–2266. 10.1002/sim.7654 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. World Bank . World Development Indicators: External Health Expenditure per Capita and Domestic General Government Health Expenditure, PPP (Current International $) Between 2000 and 2010, South Africa. 2021. World Bank.
- 31. Kahn-Lang A, Lang K. The promise and pitfalls of differences-in-differences: reflections on 16 and pregnant and other applications. J Bus Econ Stat. 2020;38(3):613–620. 10.1080/07350015.2018.1546591 [DOI] [Google Scholar]
- 32. Brugha R, Simbaya J, Walsh A, et al. How HIV/AIDS scale-up has impacted on non-HIV priority services in Zambia. BMC Public Health. 2010;10(1):540. 10.1186/1471-2458-10-540 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Larson E, O'Bra H, Brown JW, et al. Equitable distribution of PEPFAR-supported HIV/AIDS Services in South Africa. Am J Public Health. 2011;101(8):1349–1351. 10.2105/AJPH.2011.300242 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Phanga T, Bhushan NL, Maseko B, et al. Perception of the Girl Power Clinic Versus Other Settings in Malawi: A Qualitative Study. AIDSImpact; 2017. [Google Scholar]
- 35. Saul J, Bachman G, Allen S, et al. The DREAMS core package of interventions: a comprehensive approach to preventing HIV among adolescent girls and young women. PLoS One. 2018;13(12):e0208167. 10.1371/journal.pone.0208167 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Fleischman J. Five Years of DREAMS and What Lies Ahead: How to Address the Intersecting Crises of HIV, Gender Inequality, and Health Security. Center for Strategic and International Studies; 2021. [Google Scholar]
- 37. Westreich D, Edwards JK, Lesko CR, et al. Transportability of trial results using inverse odds of sampling weights. Am J Epidemiol. 2017;186(8):1010–1014. 10.1093/aje/kwx164 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Polis CB, Mhango C, Philbin J, et al. Incidence of induced abortion in Malawi, 2015. PLoS One. 2017;12(4):e0173639. 10.1371/journal.pone.0173639 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Hall JA, Barrett G, Phiri T, et al. Prevalence and determinants of unintended pregnancy in Mchinji District, Malawi; using a conceptual hierarchy to inform analysis. PLoS One. 2016;11(10):e0165621. 10.1371/journal.pone.0165621 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Chandra-Mouli V, Parameshwar PS, Parry M, et al. A never-before opportunity to strengthen investment and action on adolescent contraception, and what we must do to make full use of it. Reprod Health. 2017;14(1):85. 10.1186/s12978-017-0347-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Hindin MJ, Kalamar AM, Thompson T-A, et al. Interventions to prevent unintended and repeat pregnancy among young people in low- and middle-income countries: a systematic review of the published and Gray literature. J Adolesc Health. 2016;59(3):S8–S15. 10.1016/j.jadohealth.2016.04.021 [DOI] [PubMed] [Google Scholar]
- 42. Burchett HED, Kneale D, Griffin S, et al. Which structural interventions for adolescent contraceptive use have been evaluated in low- and middle-income countries? Int J Environ Res Public Health. 2022;19(18):11715. 10.3390/ijerph191811715 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Phillips SJ, Mbizvo MT. Empowering adolescent girls in Sub-Saharan Africa to prevent unintended pregnancy and HIV: a critical research gap. Int J Gynecol Obstet. 2016;132(1):1–3. 10.1016/j.ijgo.2015.10.005 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Deidentified data underlying this article will be shared on reasonable request to Dr Nora E. Rosenberg. Approval of requests for deidentified data will require a data use agreement.