Paper co-authored with Arshad Aminu Yakasai, Steve Kretschmer, Allie Gugliotti, and Trisha Wood Santos. Presented at ICFP 2025 in Bogota, Colombia
Background
Social vulnerabilities significantly influence contraceptive uptake, method switching, and discontinuation. Material poverty, misinformation, provider bias, low agency, and gendered power dynamics often hinder access and informed choice. These challenges are further shaped by psycho-behavioral elements such as norms, stigma, relationship quality, and coping styles, which can subtly or overtly affect contraceptive behaviors. While vulnerability is often treated as a static construct, it is dynamic and context-specific - its influence varying by age, geography, method preference, and cultural setting.
This study leverages a multi-country, large-scale dataset to explore how both socio-economic (e.g., education, occupation, decision-making power) and psycho-behavioral (e.g., trust in partner, contraceptive secrecy, satisfaction with current method) vulnerabilities interact with women’s contraceptive start/stop decisions. By applying machine learning approaches, we aim to surface nuanced patterns and high-impact leverage points for reproductive health programs. Our analysis offers cross-country comparisons to identify both universally relevant and contextually specific vulnerability dynamics, with implications for targeting method choice interventions more effectively.
Main Question
What is the relative impact of socio-economic and psycho-behavioral vulnerability factors on women’s contraceptive initiation and discontinuation? This study investigates how these attributes interact across geographies, life stages, and relationship contexts to either facilitate or hinder contraceptive uptake. We hypothesize that age-specific dynamics, interpersonal influences, and socio-cultural beliefs intersect in distinct ways to shape decisions—especially among adolescents and young women—necessitating tailored programmatic responses.
Methodology
Data was collected in 2022-2023 from 8 Countries across 3 Continents, yielding a total sample size of n~24,000 women of reproductive age. In the 7 LMICs - namely, Mexico, India (Bihar state), Burkina Faso, Ethiopia, Kenya, Nigeria, Zambia - CAPI survey interviews were conducted with multi-stage random cluster samples drawn with probability proportional to size (PPS), stratified by urban-rural strata. In contrast, the USA sample was a calibrated blend of probability-based and non-probability web panels. Random forest ML models with gradient boosting, hyperparameter tuning, and multi-fold cross validation were performed to identify vulnerability dynamics in uptake and discontinuation of contraceptives.
Results
Complex interaction effects emerged between age, agency, partner dynamics, and perceptions of contraceptive methods. Among women under 20, relationship quality strongly influenced contraceptive decisions: emotional support from a partner increased daily pill use, while criticism or emotional abuse raised discontinuation risk. For women over 20, relationship dynamics had less predictive power, suggesting greater individual autonomy in contraceptive behaviors.
Psycho-behavioral attributes significantly shaped method uptake and continuation. Embarrassment around seeking contraception predicted pill discontinuation among youth, while the belief that regular menstruation is essential for health led to injection discontinuation. Trust in a partner amplified the positive association between high sexual frequency and implant use, whereas distrust neutralized this relationship.
Satisfaction with current methods consistently emerged as a barrier to switching—especially among urban users of traditional or on-demand methods—suggesting contentment with familiar options often outweighs perceived benefits of new products. Peer and maternal influence reduced the likelihood of rejection, highlighting the importance of interpersonal influencers in contraceptive adoption.
Education showed non-linear effects, whereby both lowest and highest levels (of women and their partners) were linked to lower initiation and discontinuation of methods, compared to couples with average median years of schooling. In some contexts, highly educated women aged 20–24 were less likely to adopt implants, contrasting with trends in other age groups.
Across countries, rural women were more likely to experience vulnerabilities tied to secrecy, gender norms, and method misconceptions, while urban women’s choices were more shaped by relational dynamics and satisfaction with their current method.
Contribution
This study demonstrates that social vulnerability is a multi-dimensional, context-specific construct with dynamic effects on contraceptive behavior. By applying machine learning across large, harmonized datasets from eight countries, we reveal complex, often non-linear interactions between socio-economic and psycho-behavioral factors that shape contraceptive initiation and discontinuation.
Our analysis underscores that psycho-behavioral attributes—such as trust in partners, emotional support, secrecy, and perceived method risks—play as critical a role as structural barriers like poverty or education. The influence of these factors varies significantly by age: adolescent and young adult women are more affected by partner dynamics and social norms, while older women’s choices reflect higher autonomy and method satisfaction.
The non-linear effects of education challenge conventional assumptions about its protective influence, showing that both the least and most educated women may avoid or discontinue certain methods. Additionally, satisfaction with current methods—especially among urban traditional method users—emerges as a consistent deterrent to adopting new options, even when alternatives are available.
By surfacing these cross-cutting and context-specific patterns, our findings offer actionable insights for reproductive health programs aiming to improve method uptake and continuation. Tailored strategies that address relational and behavioral dimensions—particularly among youth—are critical to closing gaps in contraceptive access and autonomy. This research contributes to a growing evidence base advocating for nuanced, data-driven approaches to reproductive health that account for the interplay of personal, relational, and structural vulnerabilities.