Forecasting and Public Health Implications of Cervical Cancer Mortality in Canada: A Time-Series and Machine Learning Study

Cervical cancer, a highly preventable disease primarily caused by HPV, remains a critical global health issue. In Canada, its incidence rate grew fastest among all cancers from 2015 to 2023, highlighting a pressing public health challenge. To inform future planning, this study forecasts national cervical cancer mortality by comparing traditional and machine learning time-series models. The author analyzed annual age-standardized mortality data (1950-2022) from the Public Health Agency of Canada, evaluating the Exponential Smoothing (ETS), ARIMA, Neural Network Autoregression (NNAR), and eXtreme Gradient Boosting (XGBoost) models using rolling-window cross-validation. Results showed that traditional models outperformed their machine learning counterparts in predictive accuracy. The ETS and ARIMA models achieved lower Root Mean Square Errors (RMSEs of 0.45 and 0.44, respectively) compared to NNAR (0.65) and XGBoost (1.39), indicating that simpler methods better captured the linear historical trend. Despite its lower overall accuracy, XGBoost provided a crucial epidemiological insight: it identified a strong 16-18 year lag in mortality rates, which accounted for over 83% of the model's feature importance, suggesting deep-seated cohort effects or the long-term impact of prevention programs. These forecasts offer actionable data for Canadian health authorities to optimize screening and vaccination strategies. Moreover, the demonstrated lag between intervention and outcome provides a vital evidence-based framework for other countries, supporting the global goal of reducing cervical cancer mortality through timely, data-driven policy.