Going beyond the average treatment effect: estimating the diverse impacts of a home visiting intervention on infant breastfeeding, hemoglobin, and dietary diversity
Loading...
Authors
Balakrishnan, Siva
Issue Date
2025
Type
Dissertation
Language
en_US
Keywords
Anaemia , Breastfeeding , China , Mediation , Moderation , Time series clustering
Alternative Title
Abstract
Introduction: Global goals in public health include improving maternal and infant health. Numerous interventions have been employed to achieve these goals, many of them involving educational home visits from community health workers (CHWs). The evaluations of these interventions predominantly estimate their average treatment effect, ignoring intervention mechanisms and potentially varying subgroup effects. We conducted a secondary analysis of the Healthy Future community-based home visiting intervention in rural China, which had positive effects on breastfeeding behaviors and maternal outcomes, such as mental health. We aimed to answer three additional questions: (1) Who benefited the most and least from the intervention? (2) How did the intervention improve breastfeeding behaviors? (3) What was the effect of the intervention on overall trends in breastfeeding? Paper 1 - Heterogeneous treatment effects: We used the sorted effects method to estimate individual program effects on exclusive breastfeeding (EBF), infant hemoglobin, and dietary diversity (DDS) using linear and logistic models for continuous and binary outcomes, respectively. Models included treatment, selected baseline characteristics, and first-degree interactions between treatment and baseline characteristics. Participants were clustered into top 20% and bottom 20% treatment effect groups and the characteristics of these two groups were compared. Mothers with high baseline caregiving knowledge, who gave birth vaginally, and had less social support saw the greatest improvements in EBF. Mothers who were pregnant at baseline, had less social support, and low knowledge saw the most benefits on hemoglobin. Mothers with infants born at baseline, more social support, and low knowledge had greatest improvements to DDS. Paper 2 - Mediation analysis: We decomposed the total effect of the program on a recommended breastfeeding practices composite index (mean 0, SD 1) into total direct effects (TDE) and pure indirect effects (PIE). The TDE is the average difference of counterfactual outcomes Y[a=0, ma=0] and Y[a=1, ma=0]). The PIE is the average difference of counterfactual outcomes Y[a=1, ma=0] and Y[a=1, ma=1]. Counterfactual outcomes were imputed from mediator and outcome models, and direct and indirect effects estimated, with bootstrapped 95% confidence intervals. Among all infants, breastfeeding knowledge (PIE = 0.012 SD; 95% CI 0.003, 0.022) and social support (PIE = 0.006 SD; 95% CI 0.001, 0.012) significantly mediated 16.4% and 5.3%, respectively, of the total effect of the program on the breastfeeding index. Paper 3 - Trends in breastfeeding: For each of their first six months postpartum, infants were labelled as EBF, mixed feeding (MF, feeding breast milk and other foods/liquids), or not breastfeeding (NBF). We used dynamic time warping to cluster feeding trends during the six-month period. We found seven distinct trends: always EBF (prevalence = 21.3%), always MF (33.2%), never breastfed (5.3%), EBF until the 5th month (15.1%), MF until the 5th month (13.6%), mostly EBF (6.9%), and NBF from the 3rd month (4.7%). Treatment was associated with improved odds of always EBF (ROR = 2.61, 95% CI 1.25, 5.42), MF until the 5th month (ROR = 2.52, 95% CI 1.18, 5.39), and NBF from the 3rd month (ROR = 2.82, 95% CI 1.16, 6.87) compared to being never breastfed. Mothers who always exclusively breastfed had higher education attainment and pre-intervention caregiving knowledge compared to mothers who did not. Conclusion: We identified several strengths and possible improvements to the Healthy Future program. The program successfully improved BF practices in part through modules correcting BF misconceptions and informing on BF timings. The program improved the odds of EBF in mothers with high baseline knowledge. In contrast, mothers with low baseline knowledge saw greater benefits from the program on infant hemoglobin and dietary diversity. CHWs potentially supported mothers lacking in social support, resulting in greater benefits to EBF and hemoglobin. The intervention could be improved by better targeting mothers with low baseline knowledge, possibly through modifying messaging for first-time mothers and collaborating with hospital staff during delivery. At the fifth month postpartum, the program could add messaging for mothers to avoid early supplementation. Finally, the program could add more thoughtful messaging to family members to strengthen maternal support and improve infant outcomes.