BENEFIT AND BURDEN OF MULTI-CANCER DETECTION - A BAYESIAN, SIMULATION-BASED ANALYSIS
Author(s)
Reinier Meester, MSc, PhD1, Niels Dunnewind, MSc2, Freek van Delft, MSc, PhD3, David K. Edwards, 5th, PhD1, Signe Fransen, MSc1, Erik Koffijberg, MSc, PhD3.
1Freenome Holdings, Inc, San Francisco, CA, USA, 2Independent consultant, Rotterdam, Netherlands, 3University of Twente, Enschede, Netherlands.
1Freenome Holdings, Inc, San Francisco, CA, USA, 2Independent consultant, Rotterdam, Netherlands, 3University of Twente, Enschede, Netherlands.
OBJECTIVES: Multiomic tests are an emerging option for early cancer detection with uncertain diagnostic performance and outcomes. Ongoing multi-cancer detection (MCD) trials use late-stage cancer reduction as the principal surrogate for mortality benefit. Models can be used to estimate long-term associated mortality benefits but should account for uncertainty in the natural history of disease and the diagnostic performance of tests.
METHODS: A multi-disease, discrete-event simulation model was developed using age-, sex- and site-specific incidence, stage and survival data from the U.S. Surveillance Epidemiology and End Results program (SEER; 2010-2015). Unobserved preclinical progression was informed by Bayesian calibration vs randomized clinical trials for screened cancers, and imputation for other cancers. Outcomes were cross-validated. We highlight an example analysis evaluating annual MCD in average-risk adults aged 40-80 y, with site- & stage-specific diagnostic performance based on the Circulating Cell-free Genome Atlas study. MCD follow-up was limited to 3 exams based on cancer signal origin. Outcomes include lifetime (late-stage) cancer cases, deaths, required tests, follow-up exams, and tests or exams per life-years gained (LYG) for 12 priority cancers (“number-needed-to-screen" or "-treat”), with 95% credible intervals reflecting uncertainty in key model inputs.
RESULTS: Without screening, for every 1000 adults, 336 (335-337) adults developed one of 12 selected cancers, 59 (58-60) were diagnosed in stage III, 84 (84-85) in stage IV, and 148 (147-148) died from the disease. MCD detected 133 (123-141) cancer cases, reduced stage III-IV diagnoses by 42% (38-47%), stage IV diagnoses by 58% (54-61%), deaths by 29% (26-32%) and yielded 780 (714-850) LYG. MCD had an estimated number-needed-to-screen of 43 (39-46) and number-needed-to-treat of 0.9 (0.6-1.4).
CONCLUSIONS: A novel multi-cancer model accounting for uncertainty in cancer development and test performance suggests that MCD may meaningfully improve long-term cancer outcomes. Future modeling studies should incorporate forthcoming clinical utility data and explore different possible applications.
METHODS: A multi-disease, discrete-event simulation model was developed using age-, sex- and site-specific incidence, stage and survival data from the U.S. Surveillance Epidemiology and End Results program (SEER; 2010-2015). Unobserved preclinical progression was informed by Bayesian calibration vs randomized clinical trials for screened cancers, and imputation for other cancers. Outcomes were cross-validated. We highlight an example analysis evaluating annual MCD in average-risk adults aged 40-80 y, with site- & stage-specific diagnostic performance based on the Circulating Cell-free Genome Atlas study. MCD follow-up was limited to 3 exams based on cancer signal origin. Outcomes include lifetime (late-stage) cancer cases, deaths, required tests, follow-up exams, and tests or exams per life-years gained (LYG) for 12 priority cancers (“number-needed-to-screen" or "-treat”), with 95% credible intervals reflecting uncertainty in key model inputs.
RESULTS: Without screening, for every 1000 adults, 336 (335-337) adults developed one of 12 selected cancers, 59 (58-60) were diagnosed in stage III, 84 (84-85) in stage IV, and 148 (147-148) died from the disease. MCD detected 133 (123-141) cancer cases, reduced stage III-IV diagnoses by 42% (38-47%), stage IV diagnoses by 58% (54-61%), deaths by 29% (26-32%) and yielded 780 (714-850) LYG. MCD had an estimated number-needed-to-screen of 43 (39-46) and number-needed-to-treat of 0.9 (0.6-1.4).
CONCLUSIONS: A novel multi-cancer model accounting for uncertainty in cancer development and test performance suggests that MCD may meaningfully improve long-term cancer outcomes. Future modeling studies should incorporate forthcoming clinical utility data and explore different possible applications.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR90
Topic
Clinical Outcomes, Medical Technologies, Methodological & Statistical Research
Disease
Oncology