早期检测的潜在生存获益：多癌种模拟模型的开发与结果

The rapid development of blood-based or liquid-biopsy cancer screening tests may change the landscape of how cancer screening is performed. Statistical models that study optimal screening strategies are usually of a single cancer type. We present the development of a multi-cancer model and results associated with early detection of the top five adult cancers by annual mortality in White and Black racial groups: lung, breast, prostate, colorectal and pancreas. We developed a cross-cohort discrete event simulation model and calibrated it to SEER age-adjusted cancer incidence for individuals aged 18-80 from birth cohorts 1939-2001 for each cancer type, race and sex. The natural history comprised of the following trajectories: from the healthy state to cancer onset or all-cause mortality; from the cancer state to cancer death or non-cancer death, including the possibility of long-time survivors who return to the healthy state. Our model is optimized to learn a cancer, sex, and race-specific time-to-cancer distribution through a simulated annealing approach. To capture simultaneous risk of multiple cancers in a population, we modeled cancers as competing risks and internally validated model behavior against individually calibrated site-specific cancers. Early detection benefit was estimated by stage-shifting all cancer patients to SEER localized stage and advancing detection times using literature-derived mean sojourn time estimates from primary screening studies. We confirmed the validity of our competing cancer risk model by comparing cancer incidence and mortality to individually calibrated cancers. Aggregate incidence and mortality in the multi-cancer model closely approximated the sum of individually calibrated site-specific cancers, reflecting expected competition for first diagnosis. Early detection for the top five most common cancers was associated with an approximately threefold reduction in overall cancer mortality compared to no screening, with similar results seen across males and females. White individuals showed an approximately 3.7-fold reduction in cancer death rates (males: 4.77% vs 1.29%; females: 4.87% vs 1.30%), which was substantially greater than the approximately 1.8-fold reduction seen for black individuals (males: 6.45% vs 3.58%; females: 6.22% vs 3.59%). Overall, our modeling results find projected benefit using our framework for assessing population-level mortality reduction from early multi-cancer detection, with differential effects observed across racial groups. We present the results of our multi-cancer model developed and calibrated to SEER data. Our model, under ideal screening conditions, projects substantial cancer mortality benefits which vary by cancer type and correlate with estimated mean sojourn time, a proxy for tumor aggressiveness. Prospective clinical studies that incorporate performance characteristics of real screening tests and clinical data are necessary to confirm clinical benefits. Future work should incorporate the potential harms of false positive tests and resource utilization associated with population level screening.