Optimizing risk cutoffs in joint endoscopic screening for upper gastrointestinal cancers: a data-driven approach from...

Although risk prediction models for esophageal cancer (EC) and gastric cancer (GC) exist, determining appropriate cutoffs under predetermined endoscopy capacity remains challenging. We aimed to develop a strategy for selecting optimal high-risk cutoffs in model-based joint screening for EC and GC.

We built and validated the strategy based on four population-based screening cohorts from different regions of China and two diagnostic models for predicting prevalent EC and GC. The proposed strategy consisted of two steps: determining the relative high-risk coverage depending on the incidence rate ratio (IRR) of EC to GC and identifying the optimal cutoffs through an exhaustive search approach.

Totally, 43,156 individuals who had undergone endoscopic examination were included in the analysis. Taking two opportunistic screening cohorts as a benchmark dataset, we built the strategy for selecting optimal cutoffs for EC and GC at an expected total high-risk coverage of 50%. The results showed that under different IRR scenarios (range: 0.2-5.0), the optimal cutoffs could be identified by this strategy to achieve a total high-risk coverage of 50% and the detection rates of both EC and GC were improved (detection rate ratios range: 1.72-5.77). External validation results indicated that the high-risk cutoffs determined by this strategy are applicable to other populations. Finally, we developed a user-friendly online tool to aid cutoff selection ( https://pkucr.shinyapps.io/cutoff_selection/ ).

This study provided a strategy for the selection of high-risk cutoffs in EC and GC joint screening, which may hold promise for advancements in model-based joint screening.