Cancer Mortality Projections: A Comparative Analysis of Statistical Models for Evaluating Uncertainty and Improved Mortality Forecasting in Korea

Hoejun Kwon, National Cancer Center Korea
Hyunsoon Cho, National Cancer Center Korea

Cancer is a leading cause of death worldwide, and because mortality patterns vary by cancer type and are strongly influenced by age structure, mortality prediction must accurately reflect this complexity. This study aims to identify by comparing various statistical models and identify the model that shows the most consistent cancer mortality projections. Age- and sex-specific cancer mortality data from the Korea Central Cancer Registry and Statistics Korea were obtained. Compared statistical models which based on age-period-cohort factors, Joinpoint models, and time series models. The analysis utilized cancer mortality data from 1983 to 2002 to project future trends up to the year 2022. To evaluate the prediction accuracy of each model, we measured the absolute difference between the predicted value and the actual observations. This study evaluated the variability in cancer mortality predictions across multiple models. While the NordPred model demonstrated the most consistent results, it was often not considered the best fit in certain cases. Other models exhibited greater deviations and inconsistencies across various groups. Relying on a single model can result in more deviations, in subgroup-specific outcomes, complicating disease burden predictions and healthcare planning. A multi-model approach is necessary to ensure accurate and reliable for informed public health decision-making.

Keywords: Mortality and Longevity, Census data, Population projections, forecasts, and estimations, Comparative methods

See extended abstract.