Scholarlink

Introduction <p>Suboptimal clinical response (SCR) after Metabolic Bariatric Surgery (MBS) remains a significant clinical challenge, yet reliable preoperative prediction tools are lacking. Artificial intelligence (AI) methods may improve early identification of high-risk patients and support personalized surgical planning.</p> Methods <p>Using data from the Tehran Obesity Treatment Study (TOTS) a large prospective registry, we developed and evaluated multiple machine learning (ML) and deep learning (DL) models to predict SCR one year after MBS. Preoperative demographic, anthropometric, and clinical variables were used as predictors. Models included logistic regression, AdaBoost, multilayer perceptron, Support Vector Machine, Support Vector Machine with class weighting, k-Nearest Neighbors and Naïve Bayes. Performance was assessed with recall, F1-score, and area under the receiver operating characteristic curve (AUC). Model interpretability was examined using SHapley Additive exPlanations (SHAP).</p> Results <p>Logistic regression achieved a recall of 0.70, F1-score of 0.24, and an AUC of 0.7747, providing the best overall balance for identifying SCR. AdaBoost followed with recall of 0.64, F1-score of 0.24, and the highest AUC of 0.7765. The MLP achieved a recall of 0.70, F1-score of 0.23, and an AUC of 0.7534, showing comparable performance to logistic regression. SHAP analysis revealed that baseline BMI, surgery type, and age were the most influential predictors.</p> Conclusion <p>This study demonstrates the feasibility of AI-based machine learning and deep learning models for preoperative prediction of suboptimal clinical response after MBS. While traditional approaches such as logistic regression showed comparable performance, the overall results indicate that these models can meaningfully support clinicians in patient selection, risk stratification, and informed preoperative decision-making.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine and Deep Learning Models for Preoperative Prediction of Suboptimal Clinical Response One Year after Metabolic Bariatric Surgery: Findings from the Tehran Obesity Treatment Study

  • 期刊论文

  • 发表日期：

    2026年3月28日

查看全文
  • Maryam Barzin,
  • Arya Derakhshesh,
  • Omid Kohandel Gargari,
  • Maryam Mahdavi,
  • Sobhan Kazemi,
  • Majid Valizadeh,
  • Azra Ramezankhani

摘要

Introduction

Suboptimal clinical response (SCR) after Metabolic Bariatric Surgery (MBS) remains a significant clinical challenge, yet reliable preoperative prediction tools are lacking. Artificial intelligence (AI) methods may improve early identification of high-risk patients and support personalized surgical planning.

Methods

Using data from the Tehran Obesity Treatment Study (TOTS) a large prospective registry, we developed and evaluated multiple machine learning (ML) and deep learning (DL) models to predict SCR one year after MBS. Preoperative demographic, anthropometric, and clinical variables were used as predictors. Models included logistic regression, AdaBoost, multilayer perceptron, Support Vector Machine, Support Vector Machine with class weighting, k-Nearest Neighbors and Naïve Bayes. Performance was assessed with recall, F1-score, and area under the receiver operating characteristic curve (AUC). Model interpretability was examined using SHapley Additive exPlanations (SHAP).

Results

Logistic regression achieved a recall of 0.70, F1-score of 0.24, and an AUC of 0.7747, providing the best overall balance for identifying SCR. AdaBoost followed with recall of 0.64, F1-score of 0.24, and the highest AUC of 0.7765. The MLP achieved a recall of 0.70, F1-score of 0.23, and an AUC of 0.7534, showing comparable performance to logistic regression. SHAP analysis revealed that baseline BMI, surgery type, and age were the most influential predictors.

Conclusion

This study demonstrates the feasibility of AI-based machine learning and deep learning models for preoperative prediction of suboptimal clinical response after MBS. While traditional approaches such as logistic regression showed comparable performance, the overall results indicate that these models can meaningfully support clinicians in patient selection, risk stratification, and informed preoperative decision-making.

Graphical Abstract