Estimation of the performance of a risk prediction model for gastric cancer occurrence in Japan: Evidence from a small external population

Introduction We recently developed a risk prediction model for gastric cancer which showed good performance in terms of discrimination. However, lack of external validation hampers the generalizability of our results. Methods The study population consisted of 1292 individuals from JPHC cohort I (Omonogawa town, Akita prefecture). The previously developed model was used to predict survival for each individual at 10 years of follow-up. Results Thirty-three cases of gastric cancer occurred during 17,246 person-years of follow-up (27 cases occurred during the first 10 years). The c-index was estimated at 0.798 at 10 years of follow-up. In terms of calibration, the Nam-d’Agostino test was non significant (p-value = 0.23). Discussion Our previously developed risk prediction model for gastric cancer showed good performance on an external population. This suggests it might be used for risk discrimination in the general Japanese population.