Forecasting cancer incidence and prevalence using age–period–cohort and survivorship models: a practical, flexible, and interpretable framework

Age-period-cohort (APC) model outputs have been used extensively to produce forecasts of cancer incidence, identify emerging public health concerns, and quantify the impact of potential interventions. However, these models have not been extended to forecast cancer <i>prevalence</i>-the number of cancer survivors per capita. Recent advancements in cancer screening and therapeutics have substantially improved survival for many malignancies, leading to an increased need to gauge the future health resource needs of cancer survivors. Concurrent shifts in cancer incidence trends require new methods to identify the separate and joint impacts of incidence and survival changes. In this paper, we formalize methods for forecasting incidence and introduce novel forecasting methods for prevalence that are highly flexible and interpretable. Our approach has three steps. First, we model cancer incidence trends by age, period, and birth cohort using the New APC Model. Second, we model all-cause mortality by age at diagnosis and year of diagnosis using flexible regression splines. Third, we estimate cancer prevalence as the convolution of cancer incidence and all-cause mortality, accounting for the need for backward projection of incidence to estimate prevalence during early periods. We illustrate our methods using data on invasive female breast cancer, stratified by estrogen receptor status, based on 1992-2019 SEER data. Our analysis illustrates how to calculate the relative impact of period vs. cohort effects on future incidence trends, the contributions of incidence trends and survival trends on future prevalence trends, and total case count estimation.