基于机器学习构建结直肠癌术后第3个月营养不良风险的预测模型

Development and validation of a risk prediction model for malnutrition three months after colorectal cancer surgery

摘要

目的：基于患者主观整体营养评估（PG-SGA）诊断结直肠癌（CRC）患者的营养状况，构建快速且易于应用的术后第3个月营养不良预测模型。方法：纳入2024年5—10月山东大学齐鲁医院于门诊复查的303例CRC患者。收集患者的人体测量学指标、实验室检查结果及人体成分分析等相关数据。基于LASSO回归和单因素Logistic回归筛选变量，构建术后第3个月的多种机器学习模型。模型性能通过ROC曲线（AUC）及混淆矩阵相关指标，并结合Brier评分进行综合评估。结果：根据PG-SGA评分，术后第3个月营养不良发生率为44.88%；经LASSO回归及单因素Logistic回归分析，筛选出包括糖尿病、TNM分期、前白蛋白、左小腿围度及全身相位角等13个与营养不良显著相关的特征变量（P<0.05）。在7种机器学习算法中，随机森林（RF）算法表现最佳，AUC值达到0.912，特异度高达0.914；基于SHAP分析框架将模型可视化，全身相位角、左小腿围度、厌食评分及前白蛋白是术后第3个月营养不良风险的核心预测因子（均P<0.001），其中全身相位角的预测能力最强。结论：基于RF算法构建的营养不良风险预测模型，对结直肠癌患者术后第3个月营养不良的发生风险具有良好的预测能力，可为临床CRC患者的快速营养干预提供科学指导。

Abstract

Objective: To assess the nutritional status of colorectal cancer (CRC) patients using the Patient-Generated Subjective Global Assessment (PG-SGA) and to develop a rapid and easy-to-apply prediction model for malnutrition at the third postoperative month. Methods: This study enrolled 303 CRC patients during their outpatient follow-up at the third postoperative month at Qilu Hospital of Shandong University from May 2024 to October 2024. Relevant data, including anthropometric measurements, laboratory findings, and body composition analysis, were collected. Variables were selected using LASSO and univariate logistic regression to construct multiple machine learning models. Model performance was evaluated using the ROC curve (AUC), confusion-matrix-derived metrics, and the Brier score. Results: The prevalence of malnutrition at the third postoperative month, as diagnosed by PG-SGA, was 44.88%. LASSO regression combined with univariate logistic regression identified 13 feature variables significantly associated with malnutrition, including diabetes, TNM stage, prealbumin, left calf circumference, and whole-body phase angle（P<0.05）. Among the seven machine learning algorithms evaluated, the Random Forest (RF) model exhibited the best performance, achieving an Area Under the Curve (AUC) of 0.912 and a specificity of 91.4%. The model was visualized using the SHAP (Shapley Additive Explanations) analysis framework. Whole-body phase angle, left calf circumference, anorexia score, and prealbumin were identified as core predictors for malnutrition risk at the third postoperative month(P<0.001), with whole-body phase angle demonstrating the strongest predictive power. Conclusion: The prediction model based on the RF algorithm demonstrates excellent performance in predicting the risk of malnutrition in CRC patients three months postoperatively. It can provide scientific guidance for prompt and targeted nutritional interventions for these patients in a clinical setting.

关键词

结直肠癌 / 营养不良 / 预测模型 / 机器学习

Key words

Colorectal cancer / Malnutrition / Predictive model / Machine learning

参考文献

[1] Gupta A , Gupta E , Hilsden R , et al. Preoperative malnutrition in patients with colorectal cancer[J]. Can J Surg, 2021, 64（6）: E621-E629. DOI: 10.1503/cjs.016820.

[2] 王罡艳， 罗会华 . 营养不良相关因素及代谢相关通路对结直肠癌患者预后的影响[J]. 中国现代普通外科进展， 2019， 22（6）： 496-498. DOI: 10.3969/j.issn.1009—9905.2019.06.024.

[3] Lohsiriwat V . The influence of preoperative nutritional status on the outcomes of an enhanced recovery after surgery （ERAS） programme for colorectal cancer surgery[J]. Tech Coloproctol, 2014, 18（11）: 1075-1080. DOI: 10.1007/s10151—014—1198—2.

[4] Dai T , Wu D , Tang J , et al. Construction and validation of a predictive model for the risk of three—month—postoperative malnutrition in patients with gastric cancer: a retrospective case—control study[J]. J Gastrointest Oncol, 2023, 14（1）: 128-145. DOI: 10.21037/jgo—22—1307.

[5] Balderas—Peña LM , González—Barba F , Martínez—Herrera BE , et al. Body composition and biochemical parameters of nutritional status: correlation with health—related quality of life in patients with colorectal cancer[J]. Nutrients, 2020, 12（7）: 2110. DOI: 10.3390/nu12072110.

[6] Gallois C , Artru P , Lièvre A , et al. Evaluation of two nutritional scores' association with systemic treatment toxicity and survival in metastatic colorectal cancer: an AGEO prospective multicentre study[J]. Eur J Cancer, 2019, 119: 35-43. DOI: 10.1016/j.ejca.2019.07.011.

[7] 于鹏， 地力木热提·艾买提， 哈力木拉提·吾布力卡斯木， 等 . 左半结直肠癌术后吻合口瘘的LASSO回归分析[J]. 中国现代普通外科进展， 2024， 27（1）： 61-65. DOI: 10.3969/j.issn.1009—9905.2024.01.015.

[8] 薛亦诚， 刘超， 杨贵淞， 等 . 基于Logistic回归和支持向量机的早发性结直肠癌风险预测模型[J]. 中国现代普通外科进展， 2024， 27（3）： 195-198. DOI: 10.3969/j.issn.1009—9905.2024.03.007.

[9] Wimmer E , Glaus A . Early identification of cancer—related malnutrition in patients with colorectal cancer before and after surgery: a literature review[J]. Support Care Cancer, 2022, 30（11）: 8775-8783. DOI: 10.1007/s00520—022—07230—z.

[10] Álvaro Sanz E , Garrido Siles M , Rey Fernández L , et al. Nutritional risk and malnutrition rates at diagnosis of cancer in patients treated in outpatient settings: Early intervention protocol[J]. Nutrition, 2019, 57: 148-153. DOI: 10.1016/j.nut.2018.05.021.

[11] Hirahara N , Tajima Y , Fujii Y , et al. Preoperative prognostic nutritional index predicts long—term outcome in gastric cancer: a propensity score—matched analysis[J]. Anticancer Res, 2018, 38（8）: 4735-4746. DOI: 10.21873/anticanres.12781.

[12] 倪婷， 高红玉， 赵丹， 等 . 营养干预对结直肠癌术后辅助化疗患者营养状况及生活质量的影响[J]. 肿瘤防治研究， 2022， 49（12）： 1286-1290. DOI: 10.3971/j.issn.1000—8578.2022.22.0417.

[13] Lee DU , Fan GH , Hastie DJ , et al. The clinical impact of malnutrition on the postoperative outcomes of patients undergoing gastrectomy for gastric cancer: Propensity score matched analysis of 2011—2017 hospital database[J]. Clin Nutr ESPEN, 2021, 46: 484-490. DOI: 10.1016/j.clnesp.2021.09.005.

[14] Yan X , Zhu J , Wang J , et al. Development and validation of a novel prognostic prediction system based on GLIM—defined malnutrition for colorectal cancer patients post—radical surgery[J]. Front Nutr, 2024, 11: 1425317. DOI: 10.3389/fnut.2024.1425317.

[15] Cong M , Song C , Xu H , et al. The patient—generated subjective global assessment is a promising screening tool for cancer cachexia[J]. BMJ Support Palliat Care, 2022, 12（e1）: e39-e46. DOI: 10.1136/bmjspcare—2020—002296.

[16] Hãkonsen SJ , Pedersen PU , Bath—Hextall F , et al. Diagnostic test accuracy of nutritional tools used to identify undernutrition in patients with colorectal cancer: a systematic review[J]. JBI Database System Rev Implement Rep, 2015, 13（4）: 141-187. DOI: 10.11124/jbisrir—2015—1673.

[17] De Almeida C , Penna PM , Pereira SS , et al. Relationship between phase angle and objective and subjective indicators of nutritional status in cancer patients: a systematic review[J]. Nutr Cancer, 2021, 73（11—12）: 2201-2210. DOI: 10.1080/01635581.2020.1850815.

[18] Gulin J , Ipavic E , Mastnak DM , et al. Phase angle as a prognostic indicator of surgical outcomes in patients with gastrointestinal cancer[J]. Radiol Oncol, 2023, 57（4）: 524-529. DOI: 10.2478/raon—2023—0060.

[19] Wei J , Jiao J , Chen C L , et al. The association between low calf circumference and mortality: a systematic review and meta—analysis[J]. Eur Geriatr Med, 2022, 13（3）: 597-609. DOI: 10.1007/s41999—021—00603—3.

[20] Dellière S , Cynober L . Is transthyretin a good marker of nutritional status[J]. Clin Nutr, 2017, 36（2）: 364-370. DOI: 10.1016/j.clnu.2016.06.004.

[21] Aoyama T , Nakazono M , Segami K , et al. Clinical significance of the prealbumin level in gastric cancer patients who receive curative treatment[J]. J Gastrointest Cancer, 2023, 54（1）: 27-34. DOI: 10.1007/s12029—021—00777—w.

[22] Hu Q , Zhu KF , Sun J . Preoperative c—reactive protein to prealbumin ratio is independently associated with prognosis in patients with resectable colorectal cancer[J]. J Surg Oncol, 2022, 126（3）: 622-630. DOI: 10.1002/jso.26823.

[23] Fielding RA , Landi F , Smoyer KE , et al. Association of anorexia/appetite loss with malnutrition and mortality in older populations: a systematic literature review[J]. J Cachexia Sarcopenia Muscle, 2023, 14（2）: 706-729. DOI: 10.1002/jcsm.13186.