OBJECTIVES: Population-level cancer registries report diagnosed incident cases. However, the estimate of underlying prevalence rate of cancer is important to assess the observed and potential impact of diagnostic or screening technologies for early tumor detection. We developed a method to estimate the hidden prevalence of cancer using data-driven modeling.

METHODS: We use a backward induction approach to estimate the hidden prevalence rate of cancer. The National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) program provides incidence rates by cancer type and stage, which were convoluted with background mortality rates in a backwards fashion. Convolution periods were defined by dwell times, which were synthesized in an ensemble approach from published literature and empirical estimates. We demonstrate our method on pancreatic and colorectal cancers, for which we calculate rates and relative proportions of incident and hidden prevalent cases.

RESULTS: The SEER observed incidence rates for stage I to III are 4, 11, and 4 per 100,000 persons in the age 50+ US population for pancreatic cancer and 34, 33, and 32 per 100,000 for colorectal cancer. In comparison, the proposed method estimates hidden prevalence (including observed incidence) rates for the same stages of 48, 44, and 32 per 100,000 for pancreatic cancer and 137, 102, and 66 per 100,000 for colorectal cancer, respectively. This suggests that for each identified stage I pancreatic cancer there are 11 additional cases that exist undetected; for each identified stage I colorectal cancer, there are 3 additional undetected cases. The stagewise proportions of disease that remain undetected are 92%, 76%, and 88% for stages I to III for pancreatic cancer and 75%, 67%, and 51% for colorectal cancer, respectively.

CONCLUSIONS: Our approach reveals a large undiagnosed burden of cancer and consequent early detection opportunity, which is important when assessing the economic and population-level impact of screening and diagnostic technologies.