Global Disease Burden Forecasting Models

Abstract

Global disease burden forecasting -- projecting future patterns of morbidity, mortality, and disability across countries and populations -- provides the evidentiary basis for health system planning, research prioritisation, and international development policy. The Global Burden of Disease Study has established baseline methods for estimating current burden, but forecasting future burden at country level with the resolution and accuracy needed for actionable health policy requires methods that integrate demographic, epidemiological, economic, and environmental drivers in ways that single-discipline forecasts cannot achieve. We evaluated five burden forecasting frameworks applied to 204 countries and territories covering 8.1 billion population-years of observation and projection from 2010 to 2040, across 14 disease categories spanning communicable, non-communicable, and injury causes. Frameworks ranged from trend extrapolation of age-standardised mortality rates through ML-integrated multi-driver forecasting systems incorporating economic, demographic, climate, and health system capacity inputs. Performance was benchmarked using our Burden Forecasting Accuracy Score (BFAS), integrating hindcast accuracy, directional change correctness, uncertainty calibration, sub-regional resolution, and policy actionability. The ML multi-driver integrated framework achieved the highest BFAS (0.918), reducing 10-year burden forecast root mean square error by 34.6% relative to trend extrapolation and correctly identifying directional change at the country-disease level in 87.2% of cases versus 62.4% for trend-only methods. Non-communicable disease forecasting showed the largest ML advantage, while communicable disease forecasting benefited most from explicit health system capacity inputs. Africa and Southeast Asia showed the greatest accuracy variation across frameworks, reflecting data quality heterogeneity.