Using Causal Inference to Investigate Contraceptive Discontinuation in Sub-Saharan Africa

Victor Akinwande, Megan MacGregor, Celia Cintas, Ehud Karavani, Dennis Wei, Kush R. Varshney, Pablo A. Nepomnaschy

Abstract

Discontinuation rates vary by family planning method and across socio-economic contexts. Understanding these variations and their causes is paramount for developing and implementing policies aimed at curbing discontinuation rates. Randomized controlled trials (RCTs) are ideal for obtaining this information, but this design can be extremely expensive and logistically complex. The ongoing collection of comprehensive data sets, such as Demographic and Health Surveys (DHS data), when combined with machine learning methods, present an alternative and relatively cost-effective means of evidence gathering for policy development. Here, we use causal inference to estimate the effect of injectable contraceptive use on discontinuation over the 12-month period that follows its adoption. To that aim, we use retrospective observational data from seven sub-Saharan African countries captured by the DHS’ Contraceptive Calendar. We use machine learning methods to characterize data regions that share common covariate support. We find that the use of injectables increased the risk of discontinuation in four of the seven countries analyzed. Consistent with existing literature, we find that concerns with the side-effects of injectables appear to be the most frequent reason for discontinuation. However, these risks decreased after adjusting for socio-economic factors. As risk estimates may not apply uniformly within populations, we characterized the sub-populations for robust estimations by their geographical region, level of unmet needs, marital status, level of education, and age of first sex.

Introduction

Family planning is a cornerstone of global health, yet discontinuation rates for modern contraceptives remain high in developing regions. Understanding why women stop using a method is critical for policy, but the gold standard for evidence—Randomized Controlled Trials (RCTs)—is often prohibitively expensive and logistically difficult to implement at scale. Most existing research relies on associative studies, which can be misleading due to confounding factors like wealth or education. This study bridges the gap by using causal inference to estimate the true effect of injectable contraceptives on discontinuation over a 12-month period.

Methodology

The study utilized retrospective observational data from the Demographic and Health Surveys (DHS) Contraceptive Calendar across seven Sub-Saharan African countries: Burundi, Ethiopia, Liberia, Nigeria, Sierra Leone, Uganda, and Zambia. The sample included 20,703 individuals.

To move beyond simple correlation, we employed a potential outcomes framework:

  • Causal Weighting: Inverse Probability Weighting (IPW) and Overlap Weighting (OW) were used to adjust for 23 variables, ensuring the treatment (injectables) and control (other modern methods) groups were statistically comparable.
  • OverRule Algorithm: This machine learning method identified the "overlap region"—the subset of the population where women had a realistic probability of choosing either method. OverRule generates interpretable Boolean rules to define these sub-populations.
  • Propensity Modeling: Calibrated Logistic Regression (CLR) with ridge regularization estimated the probability of method choice.

Results

The analysis revealed that the "naive" marginal risk of stopping injectables was consistently higher than the adjusted causal risk. This suggests a "socio-economic buffer," where factors like education and information access significantly influence discontinuation rates previously attributed solely to the method itself.

Risk ratio estimates for injectable discontinuation across seven countries, comparing full and overlap samples using IPW and OW models.

The use of injectables increased the risk of discontinuation due to health concerns in four of the seven countries. Average Treatment Effect (ATE) risk ratios ranged from 1.08 to 1.69. A ratio of 1.69 (found in Nigeria) implies that users of injectables are 69% more likely to stop due to health concerns than users of other modern methods.

Detailed risk ratios and 95% confidence intervals for full and overlap cohorts across different weighting models.

The OverRule algorithm confirmed high coverage, with overlap cohorts representing 91% to 95.7% of the total survey respondents. Placebo tests validated the model, yielding risk ratios close to 1.0 when treatment was replaced with random noise.

Discussion and Conclusion

While injectables are a popular choice for re-initiating family planning, they also exhibit high discontinuation rates due to side effects. However, our findings highlight that socio-economic factors play a substantial role in these outcomes.

Limitations include potential recall bias in retrospective DHS data and the assumption that socio-economic factors remained stable over the five-year survey period. Future research should leverage more frequent datasets like PMA2020 and expand to other reproductive life stages. By providing interpretable demographic profiles, this approach offers policymakers actionable insights to target interventions for specific sub-populations at risk of discontinuation.

Summary updated

Abstract

Discontinuation rates vary by family planning method and across socio-economic contexts. Understanding these variations and their causes is paramount for developing and implementing policies aimed at curbing discontinuation rates. Randomized controlled trials (RCTs) are ideal for obtaining this information, but this design can be extremely expensive and logistically complex. The ongoing collection of comprehensive data sets, such as Demographic and Health Surveys (DHS data), when combined with machine learning methods, present an alternative and relatively cost-effective means of evidence gathering for policy development. Here, we use causal inference to estimate the effect of injectable contraceptive use on discontinuation over the 12-month period that follows its adoption. To that aim, we use retrospective observational data from seven sub-Saharan African countries captured by the DHS’ Contraceptive Calendar. We use machine learning methods to characterize data regions that share common covariate support. We find that the use of injectables increased the risk of discontinuation in four of the seven countries analyzed. Consistent with existing literature, we find that concerns with the side-effects of injectables appear to be the most frequent reason for discontinuation. However, these risks decreased after adjusting for socio-economic factors. As risk estimates may not apply uniformly within populations, we characterized the sub-populations for robust estimations by their geographical region, level of unmet needs, marital status, level of education, and age of first sex.
BibTeX
@inproceedings{akinwande2024using,
  author = {Victor Akinwande and Megan MacGregor and Celia Cintas and Ehud Karavani and Dennis Wei and Kush R. Varshney and Pablo A. Nepomnaschy},
  title = {{Using Causal Inference to Investigate Contraceptive Discontinuation in Sub-Saharan Africa}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2024},
  url = {https://openreview.net/forum?id=Whd4u9BVvL},
}
Figure