New perspectives in the prediction of postoperative complications for high-risk ulcerative colitis patients: machine learning preliminary approach.

OBJECTIVE: Patients with acute severe and medical refractory ulcerative colitis have a high risk of postoperative complications after total abdominal colectomy (TAC). The objective of this retrospective study is to use machine learning to analyze and predict short-term outcomes. PATIENTS AND METHODS: 32 patients with ulcerative colitis were treated with total abdominal colectomy between 2011 and 2017. Biographical data, preoperative therapy, blood chemistry, nutritional status, surgical technique, blood transfusion and preoperative length of stay were the features selected for the statistical analyses and were used as input for the machine learning algorithms to predict the rate of complications.
RESULTS: Traditional statistical analysis showed an overall postoperative morbidity rate of 34% and a mortality rate of 3%. Preoperative low serum albumin levels (<2.5 g/dL) were related to a higher risk of minor infectious complications with statistical significance (p<0.05). Preoperative length of stay (>4 days), blood transfusions (≥1 unit) and body temperature (≥37.5°C) demonstrated a major impact on infectious morbidity with statistical significance (p<0.05). Patients treated with steroids and rescue therapy presented a higher risk of minor infectious complications (p<0.05). Evaluating only preoperative features, machine learning algorithms were able to predict minor postoperative complications with a high strike rate (84.3%), high sensitivity (87.5%) and high specificity (83.3%) during the testing phase.
CONCLUSIONS: Machine learning is demonstrated to be useful in predicting the rate of minor postoperative complications in high-risk ulcerative colitis patients, despite the small sample size. It represents a major step forward in data analysis by implementing a retrospective study from a prospective point of view.