Frailty prediction in patients with chronic digestive system diseases: based on multi-task learning model

Area of Science:

• Artificial Intelligence in Medicine
• Gerontology
• Gastroenterology

Background:

• Chronic digestive system diseases (CDSD) are a significant global health concern, increasing morbidity and mortality.
• Assessing patient prognosis through the frailty index is critical for effective healthcare management.
• Proactive healthcare strategies are needed to manage CDSD and associated frailty.

Purpose of the Study:

• To develop and evaluate a multi-timepoint frailty prediction model for patients with CDSD.
• To compare the performance of various machine learning models, including the Multi-Gate Mixture-of-Experts (MMoE) framework, for frailty prediction.
• To provide insights into frailty progression in CDSD patients for improved clinical decision-making.

Main Methods:

• Data from 565 CDSD patients, including frailty assessments at 3 and 6 years, were analyzed.
• Five models were developed and evaluated: Tab Transformer, Convolutional Neural Network (CNN), Deep Neural Network (DNN), Extreme Gradient Boosting (XGBoost), and Random Forest (RF).
• The Multi-Gate Mixture-of-Experts (MMoE) framework was utilized to enhance predictive capabilities.

Main Results:

• The MMoE framework consistently outperformed single models in predicting 3-year and 6-year frailty indices.
• Random Forest (RF) achieved perfect performance (Micro-AUC 1.000) in both training and test sets for both prediction intervals.
• Tab Transformer also demonstrated high predictive accuracy, with Micro-AUC values up to 0.999 on the test set.

Conclusions:

• The MMoE-based approach effectively predicts frailty at key time points, offering valuable insights into disease progression.
• Integrating this AI model into CDSD management can facilitate early interventions and personalized treatment plans.
• This predictive model supports proactive healthcare and improved patient outcomes in CDSD management.