Objective:

To develop and externally validate a liquid neural network (LNN)-based model for predicting peri-procedural hypoglycemia in hospitalized patients undergoing colonoscopy, and to develop a cross-platform web application integrating real-time SHAP-based interpretability analysis.

Approach:

• Model Development: Internal validation was performed using stratified five-fold cross-validation combined with out-of-fold (OOF) prediction, and LASSO feature selection and SMOTE class balancing were carried out within the training folds. Models constructed included logistic regression, decision tree, random forest, extreme gradient boosting, and LNN, with performance evaluated in terms of discrimination, calibration, and clinical utility.

Key Findings:

• LNN model achieved the highest internal-validation AUC of 0.851 (95% CI: 0.804–0.893) and best sensitivity-specificity balance.
• External validation of the web application yielded an AUC of 0.848 (95% CI: 0.758–0.921) with a sensitivity of 74.07% and specificity of 86.52%.

Interpretation:

Limitations:

• Study population limited to hospitalized patients undergoing colonoscopy.
• Retrospective design may introduce bias.
• External validation sample size was relatively small.

Conclusion:

The LNN-based model and web application provide interpretable, individualized risk assessment for peri-procedural hypoglycemia.

Sources:

• Web Application