Surgical recovery for gastric cancer patients often presents significant clinical challenges. Recent research highlights that integrating resting energy expenditure (REE) into postoperative complication prediction models can enhance accuracy. This study analyzed REE dynamics in 193 patients using machine learning algorithms. Specifically, researchers measured metabolic indices before surgery and on postoperative day 1. Consequently, they identified specific ratios, such as preH-B% and D1H-B%, as critical predictors for complications like Clavien-Dindo grade II or higher.

Furthermore, the study prioritized predictors using advanced techniques like random forest and support vector machines. Traditional models often overlook these metabolic fluctuations. However, the integrated model showed a marked improvement in discrimination. Specifically, the area under the curve (AUC) rose from 0.758 to 0.812 when adding REE metrics. This improvement suggests that metabolic monitoring provides a deeper layer of physiological insight than standard clinicopathological factors alone.

Implementing Postoperative Complication Prediction Models in Practice

Clinicians can now leverage metabolic data for better risk stratification. Notably, the parsimonious model developed in this study allows for easy clinical visualization. Additionally, internal validation through bootstrapping confirmed the model's robustness and calibration. Therefore, integrating indirect calorimetry into routine perioperative care might soon become a standard for identifying high-risk patients. This approach ensures that surgeons can implement early interventions and personalized recovery protocols effectively.

Frequently Asked Questions

How does REE improve the accuracy of surgical outcome predictions?

REE reflects a patient's metabolic response to surgical stress. By capturing these dynamic changes, postoperative complication prediction models can identify physiological deviations that traditional metrics like BMI might miss.

Which machine learning techniques were most effective in this study?

The researchers utilized random forest, support vector machines, and LASSO regression. These tools helped prioritize metabolic indices, which ultimately enhanced the model's predictive discrimination and reclassification capabilities.

Disclaimer: This content is for informational and educational purposes only. It is not intended to be a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of your physician or other qualified health provider with any questions you may have regarding a medical condition. Refer to the latest local and national guidelines for clinical practice.

References

Yang D et al. Dynamic machine learning model integrating resting energy expenditure for predicting postoperative complications after gastrectomy for gastric cancer. J Cancer Res Clin Oncol. 2026 May 10. doi: 10.1007/s00432-026-06492-y. PMID: 42107025.

Siriwardhana et al. The role of resting energy expenditure in cancer-related malnutrition. Clinical Nutrition. 2024.

Soguero-Ruiz et al. Machine learning for surgical site infection prediction. Journal of Biomedical Informatics. 2023.