Submission ID 118524
| Issue/Objective | SEEDNet (Settlement-level Epidemiological Estimates Datasets for Network Analysis) is an open-source data library providing settlement-level estimates of human health and demographics. It is designed for network-science research, traditional epidemiological studies, and program planning, monitoring, and evaluation. SEEDNet offers small-area estimates that can be tailored for program implementation and evaluation by incorporating functional units into intervention designs. Additionally, it enables the measurement of spill-over effects often overlooked in conventional monitoring and evaluation frameworks. |
|---|---|
| Methodology/Approach | Our covariate-free method utilizes georeferenced national surveys to generate small-area estimates of health indicators through local inverse-distance weighted interpolation. This approach includes an algorithm for the comprehensive identification of population settlements of all sizes globally. The methodology ensures accurate and detailed health data at the settlement level, facilitating targeted interventions and assessments. |
| Results | SEEDNet was validated using three approaches: benchmarking against previously published small-area estimates, leave-one-out cross-validation using survey clusters, and out-of-sample predictions for settlements. (1) Benchmarking: The initial validation compared SEEDNet estimates of measles vaccination coverage to those obtained using a published Bayesian Geostatistical Model (BGM) and to regional estimates from DHS surveys. The SEEDNet estimates showed high agreement with both, with acceptable Root Mean Square Distance (RMSD) values. (2) Leave-One-Out Cross-Validation: This validation involved comparing SEEDNet estimates for each georeferenced DHS cluster to the LIDW-based estimates for the same clusters. The results indicated low Bias, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) for most indicators, demonstrating the accuracy of SEEDNet predictions. (3) Out-of-Sample Predictions: The validation included 10,917 settlements across ten countries. The predicted estimates were compared to direct estimates, showing high similarity and low RMSD and MAD values. Indicators measured with higher precision in the original surveys performed better. Overall, the SEEDNet method produced reliable small-area estimates of health indicators, with good performance across validation metrics. |
| Discussion/Conclusion | SEEDNet's settlement-level database provides critical insights into health disparities and intervention opportunities at a granular level. The implications for global health policy include enhanced precision in program planning and evaluation, and the potential for scaling up activities based on localized data. By integrating settlement-level data into global health strategies, SEEDNet supports more effective and equitable health interventions. |
| Presenters and affiliations | Jean-Luc Kortenaar Centre for Global Child Health, Hospital for Sick Children Diego Bassani Centre for Global Child Health, Hospital for Sick Children Amir Hossein Darooneh Centre for Global Child Health, Hospital for Sick Children |
