RESUMEN
Background/Aims@#We aimed to develop a deep learning model for the prediction of the risk of advanced colorectal neoplasia (ACRN) in asymptomatic adults, based on which colorectal cancer screening could be customized. @*Methods@#We collected data on 26 clinical and laboratory parameters, including age, sex, smoking status, body mass index, complete blood count, blood chemistry, and tumor marker, from 70,336 first-time colonoscopy screening recipients. For reference, we used a logistic regression (LR) model with nine variables manually selected from the 26 variables. A deep neural network (DNN) model was developed using all 26 variables. The area under the receiver operating characteristic curve (AUC), sensitivity, and specificity of the models were compared in a randomly split validation group. @*Results@#In comparison with the LR model (AUC, 0.724; 95% confidence interval [CI], 0.684 to 0.765), the DNN model (AUC, 0.760; 95% CI, 0.724 to 0.795) demonstrated significantly improved performance with respect to the prediction of ACRN (p < 0.001). At a sensitivity of 90%, the specificity significantly increased with the application of the DNN model (41.0%) in comparison with the LR model (26.5%) (p < 0.001), indicating that the colonoscopy workload required to detect the same number of ACRNs could be reduced by 20%. @*Conclusions@#The application of DNN to big clinical data could significantly improve the prediction of ACRNs in comparison with the LR model, potentially realizing further customization by utilizing large quantities and various types of biomedical information.
RESUMEN
Background/Aims@#We aimed to develop a deep learning model for the prediction of the risk of advanced colorectal neoplasia (ACRN) in asymptomatic adults, based on which colorectal cancer screening could be customized. @*Methods@#We collected data on 26 clinical and laboratory parameters, including age, sex, smoking status, body mass index, complete blood count, blood chemistry, and tumor marker, from 70,336 first-time colonoscopy screening recipients. For reference, we used a logistic regression (LR) model with nine variables manually selected from the 26 variables. A deep neural network (DNN) model was developed using all 26 variables. The area under the receiver operating characteristic curve (AUC), sensitivity, and specificity of the models were compared in a randomly split validation group. @*Results@#In comparison with the LR model (AUC, 0.724; 95% confidence interval [CI], 0.684 to 0.765), the DNN model (AUC, 0.760; 95% CI, 0.724 to 0.795) demonstrated significantly improved performance with respect to the prediction of ACRN (p < 0.001). At a sensitivity of 90%, the specificity significantly increased with the application of the DNN model (41.0%) in comparison with the LR model (26.5%) (p < 0.001), indicating that the colonoscopy workload required to detect the same number of ACRNs could be reduced by 20%. @*Conclusions@#The application of DNN to big clinical data could significantly improve the prediction of ACRNs in comparison with the LR model, potentially realizing further customization by utilizing large quantities and various types of biomedical information.