Abstract
Postoperative complications are still hard to predict despite the efforts towards the creation of clinical risk scores. The published scores contribute for the creation of specialized tools, but with limited predictive performance and reusability for implementation in the oncological context. This work aims to predict postoperative complications risk for cancer patients, offering two major contributions. First, to develop and evaluate a machine learning-based risk score, specific for the Portuguese population using a retrospective cohort of 847 cancer patients undergoing surgery between 2016 and 2018, for 4 outcomes of interest: (1) existence of postoperative complications, (2) severity level of complications, (3) number of days in the Intermediate Care Unit (ICU), and (4) postoperative mortality within 1 year. An additional cohort of 137 cancer patients from the same center was used for validation. Second, to improve the interpretability of the predictive models. In order to achieve these objectives, we propose an approach for the learning of risk predictors, offering new perspectives and insights into the clinical decision process. For postoperative complications the Receiver Operating Characteristic Curve (AUC) was 0.69, for complications’ severity AUC was 0.65, for the days in the ICU the mean absolute error was 1.07 days, and for 1-year postoperative mortality the AUC was 0.74, calculated on the development cohort. In this study, predictive models which could help to guide physicians at organizational and clinical decision making were developed. Additionally, a web-based decision support tool is further provided to this end.
Highlights
Cancer is a major health problem worldwide and it is among the leading death causes of the 2 1st century
In this study, we tested the predictive performance of machine learning (ML) models for four main postoperative outcomes derived from our cancer patient population, in order to to facilitate prehabilitation strategies and manage hospital resources more efficiently
We investigated the use of machine learning techniques in the surgical risk prediction of cancer patients
Summary
Cancer is a major health problem worldwide and it is among the leading death causes of the 2 1st century. The traditional risk scores (e.g., P-POSSUM [6], ARISCAT [7] and ACS score [8]) are important in choosing the course of actions, such as prehabilitation or supportive measures, to be taken during the preoperative, intraoperative and postoperative periods [5]. Their limited predictive performance is clear, in the geriatric population. Most of these risk scores were constructed based on simple linear models with inherent limitations for high-dimensional and multi-variate data
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
