A Systematic Review of Cancer Burden Forecasting Models: Evaluating Efficacy for Long-Term Predictions Using Annual Data

Abstract

This paper presents a comprehensive systematic review of forecasting models applied to cancer burden prediction, focusing on their efficacy for long-term predictions using annual data. Cancer represents a significant challenge to global healthcare systems, necessitating accurate forecasting models for effective planning and resource allocation. We evaluated various methodologies, including JoinPoint Regression, Age-Period-Cohort models, time series analysis, exponential smoothing, machine learning, and more, highlighting their strengths and weaknesses in forecasting cancer incidence, mortality, and Disability-Adjusted Life Years. Our literature search strategy involved a systematic search across major scientific databases, yielding a final selection of 10 studies for in-depth analysis. These studies employed diverse forecasting models, which were critically assessed for their predictive accuracy, handling of annual data limitations, and applicability to cancer epidemiology. Our findings indicate that no single model universally excels in all aspects of cancer burden forecasting. However, ARIMA models and their variants consistently demonstrated strong predictive performance across different cancers, countries, and projection periods. The evaluation also underscores the challenges posed by limited long-term data and the potential for complex models to overfit in sparse data scenarios. Importantly, the review suggests a need for further research into developing models capable of accurate longer-term forecasts, which could significantly enhance healthcare planning and intervention strategies. In conclusion, while ARIMA and its derivatives currently lead in performance, there is a pressing need for innovative models that extend predictive capabilities over longer horizons, improving the global healthcare sector's response to the cancer burden.