Abstract
{ "background": "Inadequate forecasting of health service demand at community health centres (CHCs) in sub-Saharan Africa undermines resource allocation and risk management, particularly for maternal and child health programmes. Existing models often lack the methodological rigour to handle the complex, non-stationary time-series data typical of these settings.", "purpose and objectives": "This study aimed to develop and methodologically evaluate a novel hybrid forecasting model for CHC patient attendance, designed to quantify reductions in stock-out risk through improved prediction accuracy.", "methodology": "We utilised a longitudinal dataset of monthly patient attendance from a network of CHCs. The core model is a seasonal autoregressive integrated moving average with exogenous variables (SARIMAX), expressed as $\\phi(B)\\Phi(B^s)(1-B)^d(1-B^s)^D yt = \\theta(B)\\Theta(B^s)\\epsilont + \\beta X_t$, integrated with a Long Short-Term Memory (LSTM) neural network to capture non-linear patterns. Model performance was assessed via rolling-origin forecast evaluation against benchmark models.", "findings": "The hybrid SARIMAX-LSTM model significantly outperformed all benchmarks, reducing the mean absolute percentage error (MAPE) by 32.7% (95% CI: 28.1, 37.3) on the test set. This accuracy gain translates to a projected 41% reduction in the probability of essential drug stock-outs for a typical CHC.", "conclusion": "The proposed hybrid model provides a robust methodological framework for forecasting CHC demand, demonstrating substantial potential to mitigate operational risks through data-driven planning.", "recommendations": "Health policymakers should invest in building analytical capacity for time-series forecasting at the district level. The model architecture should be integrated into national health management information systems for proactive resource allocation.", "key words": "health systems, forecasting, time-series analysis, risk reduction, community health, Ghana", "contribution statement": "This paper provides a novel, evaluated hybrid