Data analytics and artificial intelligence to predict length of stay, readmission and mortality after colorectal cancer surgery

Shamsul Masum*, Adrian Alan Hopgood, Samuel Stefan, Karen Flashman, Jim Khan

*Corresponding author for this work

Research output: Contribution to journalMeeting Abstractpeer-review


Background: Accurate prediction of length of stay (LoS), readmission, and mortality remains a key challenge after successful colorectal cancer surgery and has huge resource implications for clinical and management teams. This study has investigated data analytics and artificial intelligence (AI) in predicting all three outcomes after elective cancer surgery.

Methods: A prospectively maintained colorectal cancer database was used, covering 4336 patients who underwent colorectal cancer surgery between 2003 and 2019. The 33 patient parameters included demographics, peri- and postoperative outcomes, surgical approaches, complications and mortality. Ten data analytics and AI algorithms were compared. Extra tree regressor/classifier, LASSO algorithms, and correlation matrix with heat map were investigated to identify the most important parameters. Ten-fold cross-validation was used for validation.

Results: The significant predictors of LoS were age, ASA grade, operative time, presence or absence of a stoma, robotic or laparoscopic approach to surgery, complications, and operation type. The model with support vector regressor (SVR) algorithms predicted the LoS with an accuracy of 83% and root mean squared error (RMSE) of 12.52 days. The significant predictors of mortality were age, ASA grade, BMI, laparoscopic procedure, formation of a stoma, preop T Stage, preop M Stage, neoadjuvant chemotherapy, curative resection, and LoS. Bidirectional long short-term memory (Bi-LSTM) outperformed other algorithms. It predicted mortality with 80% accuracy (or 78% using only the parameters identified as important), 84% sensitivity and 75% specificity. The model predicted the 31 days mortality with 96% accuracy, 93% sensitivity and 100% specificity. The prediction of the 91 days mortality was possible with 94% accuracy, 91% sensitivity and 96% specificity.
A Bi-LSTM model predicted readmission with 87.5% accuracy, 84% sensitivity and 90%. specificity.

Conclusions: This study demonstrates how AI can help clinicians in the prediction of LoS, readmission, and mortality after colorectal cancer surgery. Accuracies of at least 80% were achieved.
Original languageEnglish
Article numberESSO20V-0163
Pages (from-to)e5
Number of pages1
JournalEuropean Journal of Surgical Oncology
Issue number2
Early online date29 Jan 2021
Publication statusPublished - 1 Feb 2021


Dive into the research topics of 'Data analytics and artificial intelligence to predict length of stay, readmission and mortality after colorectal cancer surgery'. Together they form a unique fingerprint.

Cite this