Dynamic weight loss trajectories and machine learning-based prediction of malnutrition in colorectal cancer patients after curative surgery
- 1NHC key Laboratory of Cancer Metabolism, Department of Radiology, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, China
- 2NHC key Laboratory of Cancer Metabolism, Department of Colorectal Surgery, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, China
Abstract
Objective: This study aimed to identify distinct postoperative weight loss trajectories and develop a machine learning-based model for predicting malnutrition risk at 6 months after curative resection for colorectal cancer (CRC).
Methods: This prospective observational study enrolled 205 CRC patients who underwent curative resection. Weight loss percentages were measured at four postoperative time points relative to preoperative baseline: T1 (postoperative day 7 or hospital discharge), T2 (1 month), T3 (3 months), and T4 (6 months). Latent class growth modeling (LCGM) was employed to identify distinct weight change trajectory patterns. Malnutrition was assessed using the Global Leadership Initiative on Malnutrition (GLIM) criteria at 6 months post-surgery. Core predictors for 6-month malnutrition were screened by taking the intersection of Lasso regression and the Boruta algorithm. Ten machine learning models were constructed and systematically compared, with the optimal model interpreted using SHapley Additive exPlanations (SHAP).
Results: LCGM identified three distinct weight change trajectories: persistent weight loss (28.3%), gradual recovery (38.5%), and rapid recovery (33.2%). Multinomial logistic regression (reference category: persistent weight loss group) showed that age > 65 years and ASA III–IV were associated with lower odds of gradual recovery, whereas colon cancer (right or left) was associated with higher odds. For rapid recovery, female, age > 65 years, ASA III–IV, and adjuvant chemotherapy were associated with lower odds, whereas colon cancer (right or left) was associated with higher odds. The overall incidence of GLIM-defined malnutrition at 6 months was 33.2%. Among the ten machine learning models, XGBoost achieved the best comprehensive performance, with an AUC of 0.836 (95% CI: 0.779–0.894). Calibration curves demonstrated good agreement between predicted and observed probabilities, and decision curve analysis confirmed net clinical benefit across a 9–91% threshold range. SHAP analysis revealed that lower preoperative BMI, receipt of adjuvant chemotherapy, older age, and rectal tumor location were associated with increased malnutrition risk.
Conclusion: This study revealed three distinct postoperative weight loss trajectories in CRC patients. The developed XGBoost model, incorporating four readily available clinical variables, demonstrated favorable predictive performance and clinical utility for early identification of patients at high risk of long-term malnutrition, pending further external validation.
Related
- Individual-Level Double Burden of Malnutrition among Ethiopian Women: Prevalence, Machine-Learning Prediction and Explainability Using the 2024/25 Ethiopian Demographic and Health Survey
- Machine Learning for Risk Prediction in Colorectal Surgery—Game Changer or Unnecessary? An Institutional Pilot Study
- Prediction of Parkinson disease progression from sparse longitudinal trajectories using multiclass likelihood contrast learning
- Single-Case Design for Network-based BCI Rehabilitation: Individual Learning Trajectories and Effects of Motor Imagery Timing in Stroke Patients
- Explainable Machine Learning for Prediction of Future High-Severity States Using Longitudinal APACHE-II Trajectories: Internal Validation and Cross-Cohort Portability Assessment.
BibTeX
@article{chen2026dynamic,
title = {{Dynamic weight loss trajectories and machine learning-based prediction of malnutrition in colorectal cancer patients after curative surgery}},
author = {Xingfa Chen and Jinliang Jian and Zhenmeng Lin and Mingming Xie and Yangming Li},
journal = {Frontiers in Nutrition},
year = {2026},
month = oct,
doi = {10.3389/fnut.2026.1982995},
url = {https://doi.org/10.3389/fnut.2026.1982995},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Dynamic weight loss trajectories and machine learning-based prediction of malnutrition in colorectal cancer patients after curative surgery
AU - Xingfa Chen
AU - Jinliang Jian
AU - Zhenmeng Lin
AU - Mingming Xie
AU - Yangming Li
PY - 2026
DA - 2026/10/21
JO - Frontiers in Nutrition
DO - 10.3389/fnut.2026.1982995
UR - https://doi.org/10.3389/fnut.2026.1982995
AB - Objective: This study aimed to identify distinct postoperative weight loss trajectories and develop a machine learning-based model for predicting malnutrition risk at 6 months after curative resection for colorectal cancer (CRC). Methods: This prospective observational study enrolled 205 CRC patients who underwent curative resection. Weight loss percentages were measured at four postoperative time points relative to preoperative baseline: T1 (postoperative day 7 or hospital discharge), T2 (1 month), T3 (3 months), and T4 (6 months). Latent class growth modeling (LCGM) was employed to identify distinct weight change trajectory patterns. Malnutrition was assessed using the Global Leadership Initiative on Malnutrition (GLIM) criteria at 6 months post-surgery. Core predictors for 6-month malnutrition were screened by taking the intersection of Lasso regression and the Boruta algorithm. Ten machine learning models were constructed and systematically compared, with the optimal model interpreted using SHapley Additive exPlanations (SHAP). Results: LCGM identified three distinct weight change trajectories: persistent weight loss (28.3%), gradual recovery (38.5%), and rapid recovery (33.2%). Multinomial logistic regression (reference category: persistent weight loss group) showed that age > 65 years and ASA III–IV were associated with lower odds of gradual recovery, whereas colon cancer (right or left) was associated with higher odds. For rapid recovery, female, age > 65 years, ASA III–IV, and adjuvant chemotherapy were associated with lower odds, whereas colon cancer (right or left) was associated with higher odds. The overall incidence of GLIM-defined malnutrition at 6 months was 33.2%. Among the ten machine learning models, XGBoost achieved the best comprehensive performance, with an AUC of 0.836 (95% CI: 0.779–0.894). Calibration curves demonstrated good agreement between predicted and observed probabilities, and decision curve analysis confirmed net clinical benefit across a 9–91% threshold range. SHAP analysis revealed that lower preoperative BMI, receipt of adjuvant chemotherapy, older age, and rectal tumor location were associated with increased malnutrition risk. Conclusion: This study revealed three distinct postoperative weight loss trajectories in CRC patients. The developed XGBoost model, incorporating four readily available clinical variables, demonstrated favorable predictive performance and clinical utility for early identification of patients at high risk of long-term malnutrition, pending further external validation.
SN - 2296-861X
ER -