Scholarlink

Purpose <p>Postoperative length of stay (LOS) after colorectal cancer (CRC) surgery is an important indicator of recovery, discharge readiness, patient experience, and healthcare utilization. Although LOS is influenced by multiple clinical and organizational factors, accurate prediction may support perioperative planning and resource allocation. This systematic review aimed to identify and critically appraise artificial intelligence (AI)- and machine learning (ML)-based prediction models for postoperative LOS after CRC surgery, with emphasis on outcome definitions, predictor timing, model performance, validation, and clinical applicability.</p> Methods <p>A&#xa0;systematic search of PubMed/Medline, Embase, and Scopus was conducted from database inception to 17&#xa0;April 2026, with supplementary Google Scholar and reference screening. Peer-reviewed studies reporting AI/ML-based models developed and/or validated to predict postoperative LOS as a&#xa0;continuous outcome, or prolonged/delayed LOS as a&#xa0;categorical outcome, in adults undergoing CRC surgery were included. Risk of bias and applicability were assessed using PROBAST. Owing to substantial heterogeneity in study design, outcome definitions, predictor timing, and reported performance measures, findings were synthesized narratively.</p> Results <p>Six studies met the eligibility criteria. Included studies were heterogeneous in design, setting, cohort size, LOS definitions, predictor timing, modeling methods, and validation strategies. Outcomes were modeled variably as continuous LOS, prolonged LOS, or delayed discharge, with nonstandardized thresholds across studies. Predictors commonly included age, comorbidity burden, American Society of Anesthesiologists (ASA) score, operative approach, stoma creation, operative duration, blood loss, enhanced recovery pathway variables, and postoperative complications or early recovery measures. Internal validation frequently showed moderate-to-high predictive performance, particularly in models incorporating intraoperative or early postoperative information; however, calibration reporting, external validation, and prospective impact evaluation were uncommon. Overall, methodological concerns included inconsistent reporting of missing-data handling, small sample size in some cohorts, and limited evidence of transportability across settings.</p> Conclusion <p>Current evidence suggests that AI/ML models may help estimate postoperative LOS after CRC surgery, particularly for perioperative or early postoperative risk stratification. However, the evidence base remains limited and methodologically heterogeneous, and the scarcity of external validation and incomplete reporting reduce confidence in routine clinical implementation. Future studies should use standardized outcomes, align predictor timing with intended clinical use, and prioritize transparent reporting, robust validation, and prospective evaluation.</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

AI prediction of length of stay after colorectal cancer surgery: a systematic review

  • 期刊论文

  • 发表日期：

    2026年5月27日

查看全文
  • Arshia Bozorgnia,
  • Alireza Ghattan,
  • Mohammad Ali Arj,
  • Abbas Arj

摘要

Purpose

Postoperative length of stay (LOS) after colorectal cancer (CRC) surgery is an important indicator of recovery, discharge readiness, patient experience, and healthcare utilization. Although LOS is influenced by multiple clinical and organizational factors, accurate prediction may support perioperative planning and resource allocation. This systematic review aimed to identify and critically appraise artificial intelligence (AI)- and machine learning (ML)-based prediction models for postoperative LOS after CRC surgery, with emphasis on outcome definitions, predictor timing, model performance, validation, and clinical applicability.

Methods

A systematic search of PubMed/Medline, Embase, and Scopus was conducted from database inception to 17 April 2026, with supplementary Google Scholar and reference screening. Peer-reviewed studies reporting AI/ML-based models developed and/or validated to predict postoperative LOS as a continuous outcome, or prolonged/delayed LOS as a categorical outcome, in adults undergoing CRC surgery were included. Risk of bias and applicability were assessed using PROBAST. Owing to substantial heterogeneity in study design, outcome definitions, predictor timing, and reported performance measures, findings were synthesized narratively.

Results

Six studies met the eligibility criteria. Included studies were heterogeneous in design, setting, cohort size, LOS definitions, predictor timing, modeling methods, and validation strategies. Outcomes were modeled variably as continuous LOS, prolonged LOS, or delayed discharge, with nonstandardized thresholds across studies. Predictors commonly included age, comorbidity burden, American Society of Anesthesiologists (ASA) score, operative approach, stoma creation, operative duration, blood loss, enhanced recovery pathway variables, and postoperative complications or early recovery measures. Internal validation frequently showed moderate-to-high predictive performance, particularly in models incorporating intraoperative or early postoperative information; however, calibration reporting, external validation, and prospective impact evaluation were uncommon. Overall, methodological concerns included inconsistent reporting of missing-data handling, small sample size in some cohorts, and limited evidence of transportability across settings.

Conclusion

Current evidence suggests that AI/ML models may help estimate postoperative LOS after CRC surgery, particularly for perioperative or early postoperative risk stratification. However, the evidence base remains limited and methodologically heterogeneous, and the scarcity of external validation and incomplete reporting reduce confidence in routine clinical implementation. Future studies should use standardized outcomes, align predictor timing with intended clinical use, and prioritize transparent reporting, robust validation, and prospective evaluation.