Towards Early Warning Systems for Shigellosis in Africa: Multi-Horizon Forecasting of Incidence in Rural Western Kenya Using Statistical State-Space and Machine Learning Models
- 1Kenya Medical Research Institute
Abstract
IntroductionDespite the significant burden of Shigellosis, routine surveillance systems in resource-limited settings lack predictive capabilities to guide proactive public health responses. Consequently, we developed and evaluated the accuracy, uncertainty and computational cost of multi-horizon forecasting models for Shigella incidence among children <5 years in rural western Kenya using 16-year longitudinal incidence time-series.
MethodsWe estimated monthly Shigella incidence using data from three pediatric enteric studies (2008- 2024), supplemented with Kenya Health Information System during inter-study gaps. Incidence estimates incorporated sampling, healthcare-seeking, stool-testing, and pathogen-specific weights, with bootstrap-derived 95% confidence intervals. Predictors included Google Trends health-related indicators, climate, temporal trends, and seasonality. Forecasts at 1-, 3-, and 6-month horizons were evaluated using rolling-origin validation. Following horizon-specific predictor selection using Boruta, 14 state-space, machine-learning, and hybrid models were compared using point-forecast and probabilistic metrics, bootstrap confidence intervals, and Model Confidence Sets. Adaptive conformal prediction recalibrated prediction intervals, while Shapley Additive exPlanations assessed predictor contributions.
ResultsMean monthly Shigella incidence was 2.51 per 100 child-years (standard deviation=1.85). Shigella dynamics in Kenya can be forecast beyond seasonal persistence alone for up to six months reducing absolute errors by [~] 33%. However, optimal model architectures and predictor sets shift systematically across forecasting horizons. Temporal trend and recent incidence were the dominant predictors, while predictor importance shifted with forecast horizon toward syndromic, pharmaceutical, seasonal, and lagged environmental signals. Extreme Gradient Boosting performed best at 1 month (Root mean squared error [RMSE]=1.86), remained competitive at 3 months (RMSE=1.89), whereas exponential smoothing (ETS) performed best at 6 months (RMSE=2.05). Forecast accuracy declined with increasing horizon. ETS and TBATS provided the best-calibrated prediction intervals, whereas machine-learning models generally under-covered. Model Confidence Sets demonstrated horizon-dependent model distinguishability.
ConclusionHorizon-specific model selection balancing predictive accuracy, uncertainty calibration, and computational feasibility could strengthen operational early warning systems for shigellosis.
Author SummaryShigella infections remain an important cause of illness among young children in low-resource settings, yet public health systems often lack tools to anticipate changes in disease burden before they occur. Early warning systems that can reliably predict future Shigella incidence could help health authorities plan surveillance, allocate resources, and prepare for periods of increased transmission. We used 16 years of data from children <5 years in rural western Kenya to develop and compare different approaches for forecasting monthly Shigella incidence 1-, 3-, and 6-months ahead. We combined disease surveillance data with information on seasonal patterns, weather, and health-related internet searches. We found that Shigella incidence could be predicted beyond what would be expected from seasonal patterns alone, with useful forecasts possible up to six months ahead. However, the best forecasting approach depended on how far into the future predictions were required. Machine-learning methods provided the most accurate short-term predictions, while simpler statistical models performed better for longer-term forecasts and provided more reliable estimates of uncertainty. These findings highlight the importance of selecting forecasting methods according to the intended prediction horizon and operational requirements. Such approaches could strengthen early warning and preparedness for childhood Shigella infections in resource-limited settings.
Related
- Meteorological forecasting of respiratory diseases incidence with geospatial and machine learning methods
- Forecasting Ethiopia's Neonatal Mortality Rate Toward the Sustainable Development Goal Target 3.2 Using Statistical and Deep Learning Models: A Time-Series Forecasting Study.
- Predicting hepatitis B incidence trends in China using Baidu Search Index and time series models: a comparative study of rolling forecasting and direct forecasting strategies
- An explainable machine-learning framework for seasonal early warning of West Nile virus using predictive climate data
- Epidemiological Characteristics and Incidence Trends of Hemorrhagic Fever With Renal Syndrome in Shandong Province, China: Retrospective Observational Study With Forecasting
BibTeX
@misc{ogwel2026towards,
title = {{Towards Early Warning Systems for Shigellosis in Africa: Multi-Horizon Forecasting of Incidence in Rural Western Kenya Using Statistical State-Space and Machine Learning Models}},
author = {B. Ogwel and B. O. Onyando and B. O. Nyawanda and A. O. Awuor and J. Ogutu and J. B. Ochieng and M. Garcia Quesada and P. B. Pavlinac and M. Lau and R. Omore and E. T. Rogawski McQuade},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.19.26365895},
url = {https://www.medrxiv.org/content/10.64898/2026.10.19.26365895v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Towards Early Warning Systems for Shigellosis in Africa: Multi-Horizon Forecasting of Incidence in Rural Western Kenya Using Statistical State-Space and Machine Learning Models
AU - B. Ogwel
AU - B. O. Onyando
AU - B. O. Nyawanda
AU - A. O. Awuor
AU - J. Ogutu
AU - J. B. Ochieng
AU - M. Garcia Quesada
AU - P. B. Pavlinac
AU - M. Lau
AU - R. Omore
AU - E. T. Rogawski McQuade
PY - 2026
DA - 2026/10/19
PB - medRxiv
DO - 10.64898/2026.10.19.26365895
UR - https://www.medrxiv.org/content/10.64898/2026.10.19.26365895v1
AB - IntroductionDespite the significant burden of Shigellosis, routine surveillance systems in resource-limited settings lack predictive capabilities to guide proactive public health responses. Consequently, we developed and evaluated the accuracy, uncertainty and computational cost of multi-horizon forecasting models for Shigella incidence among children <5 years in rural western Kenya using 16-year longitudinal incidence time-series. MethodsWe estimated monthly Shigella incidence using data from three pediatric enteric studies (2008- 2024), supplemented with Kenya Health Information System during inter-study gaps. Incidence estimates incorporated sampling, healthcare-seeking, stool-testing, and pathogen-specific weights, with bootstrap-derived 95% confidence intervals. Predictors included Google Trends health-related indicators, climate, temporal trends, and seasonality. Forecasts at 1-, 3-, and 6-month horizons were evaluated using rolling-origin validation. Following horizon-specific predictor selection using Boruta, 14 state-space, machine-learning, and hybrid models were compared using point-forecast and probabilistic metrics, bootstrap confidence intervals, and Model Confidence Sets. Adaptive conformal prediction recalibrated prediction intervals, while Shapley Additive exPlanations assessed predictor contributions. ResultsMean monthly Shigella incidence was 2.51 per 100 child-years (standard deviation=1.85). Shigella dynamics in Kenya can be forecast beyond seasonal persistence alone for up to six months reducing absolute errors by [~] 33%. However, optimal model architectures and predictor sets shift systematically across forecasting horizons. Temporal trend and recent incidence were the dominant predictors, while predictor importance shifted with forecast horizon toward syndromic, pharmaceutical, seasonal, and lagged environmental signals. Extreme Gradient Boosting performed best at 1 month (Root mean squared error [RMSE]=1.86), remained competitive at 3 months (RMSE=1.89), whereas exponential smoothing (ETS) performed best at 6 months (RMSE=2.05). Forecast accuracy declined with increasing horizon. ETS and TBATS provided the best-calibrated prediction intervals, whereas machine-learning models generally under-covered. Model Confidence Sets demonstrated horizon-dependent model distinguishability. ConclusionHorizon-specific model selection balancing predictive accuracy, uncertainty calibration, and computational feasibility could strengthen operational early warning systems for shigellosis. Author SummaryShigella infections remain an important cause of illness among young children in low-resource settings, yet public health systems often lack tools to anticipate changes in disease burden before they occur. Early warning systems that can reliably predict future Shigella incidence could help health authorities plan surveillance, allocate resources, and prepare for periods of increased transmission. We used 16 years of data from children <5 years in rural western Kenya to develop and compare different approaches for forecasting monthly Shigella incidence 1-, 3-, and 6-months ahead. We combined disease surveillance data with information on seasonal patterns, weather, and health-related internet searches. We found that Shigella incidence could be predicted beyond what would be expected from seasonal patterns alone, with useful forecasts possible up to six months ahead. However, the best forecasting approach depended on how far into the future predictions were required. Machine-learning methods provided the most accurate short-term predictions, while simpler statistical models performed better for longer-term forecasts and provided more reliable estimates of uncertainty. These findings highlight the importance of selecting forecasting methods according to the intended prediction horizon and operational requirements. Such approaches could strengthen early warning and preparedness for childhood Shigella infections in resource-limited settings.
ER -