AI-based CRA-age risk-stratified screening reduced the screened population by 82.1% and improved positive predictive value from 7.6% to 15.9% compared to age-based screening.
Observational (n=7,672)
Yes
Does an AI-based risk stratification model integrating routine laboratory data with age improve the yield and cost-effectiveness of multi-cancer early detection screening compared to age-based screening?
An AI-based risk stratification model using routine laboratory data and age significantly improves the efficiency and cost-effectiveness of multi-cancer early detection screening by enriching the tested population.
Absolute Event Rate: 15.9% vs 7.6%
10538 Background: Early cancer detection improves survival, yet population-wide screening remains costly and logistically challenging. Traditional cancer risk models, such as lifestyle-based assessments and polygenic risk scores, offer limited predictive performance. Routine laboratory tests (complete blood count, urinalysis, and biochemical panels) are performed in primary care and may harbor latent cancer signals. We developed an AI-based cancer risk assessment (CRA) model using routine laboratory data to enrich cancer cases and improve the cost-effectiveness of downstream multi-cancer early detection (MCED) testing. Methods: Routine laboratory data from 7, 672 individuals (1, 245 cancer; 6, 427 non-cancer) across two hospitals were retrospectively collected and used to develop a random forest based CRA model incorporating 56 selected routine laboratory features. Among them, 5, 392 (70. 3%) also underwent OncoSeek test, a validated MCED assay based on seven protein tumor markers (e. g. , AFP, CEA). OncoSeek screening performance was estimated under two enrichment strategies: (1) age-based enrichment (≥50 years), and (2) risk-stratified enrichment using a joint CRA–age model. A two-dimensional CRA–age risk landscape was used to estimate cancer incidence across strata. Strata with a modeled cancer incidence ≥0. 57%, corresponding to the overall cancer incidence among Chinese adults (GLOBOCAN 2017), were classified as high risk. Performance metrics from both strategies were applied to simulate OncoSeek screening outcomes in a hypothetical cohort of 1, 000, 000 adults aged ≥20 years. Results: In the simulated cohort with an overall cancer incidence of 0. 57%, direct OncoSeek screening detected 2, 450 cancers with a PPV of 5. 6%. Age-based screening (≥50 years) required screening 536, 206 individuals (−46. 4%), increased cancer incidence to 0. 92%, and detected 2, 327 cancers. In contrast, CRA–age risk-stratified screening reduced the screened population to 178, 660 individuals (−82. 1%), enriched cancer incidence to 2. 69%, and detected 2, 402 cancers. Compared with age-based screening, CRA–age stratification improved PPV from 7. 6% to 15. 9%, reduced false positives by 55. 2% and lowered the cost per cancer detected from 18, 436 to 5, 950 (−67. 7%), while preserving overall cancer detection. Conclusions: AI-based CRA–age risk stratification enables efficient population-level cancer screening by restricting downstream MCED testing to a high-risk subset comprising approximately one-fifth of the population, while preserving overall cancer detection. This strategy substantially improves PPV, reduces false positives, and lowers the cost per cancer detected. These findings support a risk-stratified MCED screening paradigm as a scalable and economically sustainable alternative to population-wide or age-based population screening.
Mao et al. (Wed,) conducted a observational in Cancer (n=7,672). AI-based cancer risk assessment (CRA) model integrating routine laboratory data with age vs. Age-based screening (≥50 years) was evaluated on Positive predictive value (PPV) of cancer detection. AI-based CRA-age risk-stratified screening reduced the screened population by 82.1% and improved positive predictive value from 7.6% to 15.9% compared to age-based screening.