Health system resilience: quantifying the dynamic impact of environmental shocks on health service utilization using an interrupted time series and time-series forecasting approach in Western Province, Zambia

Background: In low- and lower-middle-income nations, advancements toward Universal Health Coverage (UHC) are progressively jeopardized by the combined effects of acute shocks, such as floods and pandemics, alongside chronic health system stressors. Although the disruptive effects of individual conditions are well understood, a significant gap exists in the empirical quantification of their synergistic interactions over time. This study employs an innovative multi-model time-series methodology to dynamically forecast health service utilization and assess the impacts of major shocks in Western Province (WP), Zambia.Methods: We conducted a longitudinal ecological study utilizing 73 months of routine health data collected from 62 healthcare facilities, spanning October 2017 to September 2023. Initially, a comparative analysis of four time-series models—ARIMA, SARIMA, ETS, and Prophet—was performed to determine the most accurate forecasting approach. Subsequently, a Hierarchical Time Series (HTS) model was employed to generate reconciled forecasts. Finally, an Interrupted Time Series (ITS) analysis was implemented to quantify the impacts of three distinct shocks: a drought in 2019, the COVID-19 pandemic in 2021, and a complex double-peak flood event in 2023.Findings: The Prophet model was identified as the most accurate forecasting tool, achieving a mean absolute percentage error (MAPE) of 7·22%. The ITS analysis demonstrated that each shock distinctly affected health service utilization. The drought was associated with a significant immediate decline in utilization (level change=-0·105, p