Hierarchical Statistical Models to Represent and Visualize Survey Evidence for Program Evaluation: iCCM in Malawi

Area of Science:

• Public Health
• Health Policy and Management
• Epidemiology

Background:

• Maternal, newborn, and child health program evaluation is complex.
• Large household surveys offer data but present analytical challenges.
• Evaluating interventions like integrated community case management (iCCM) requires robust methods.

Purpose of the Study:

• To propose and apply a hierarchical modeling approach for cross-sectional survey data in program evaluation.
• To assess the impact of scaling up iCCM in Malawi on careseeking behaviors.
• To provide empirical Bayes estimates for district-level careseeking at two time points.

Main Methods:

• Utilized hierarchical models for cross-sectional survey data.
• Applied the approach to data from Malawi's iCCM scale-up.
• Analyzed careseeking for diarrhea, pneumonia, and malaria from any source and specifically from health surveillance assistants (HSAs).

Main Results:

• No evidence that increased trained HSAs correlated with higher overall careseeking for pneumonia, diarrhea, or malaria.
• A positive association was found between trained HSAs and careseeking specifically from HSAs.
• An increase of 100 trained HSAs per district corresponded to a 2% average increase in careseeking from HSAs.

Conclusions:

• Hierarchical models offer a flexible framework for evaluating iCCM programs using national survey data.
• The findings suggest targeted strategies may be needed to boost overall careseeking beyond HSA utilization.
• The methodology can be extended for causal analyses and evaluations of similar health programs.