Emergency surgery accounted for 80.8 per cent (6862 of 8492 patients) and benign disease for 57.3 per cent (4868 of 8492) ( Table S1).Ĭohort study patient inclusion and model derivation and validation flow
The most common surgical procedures were abdominal (40.6 per cent, 3446 of 8492 patients), orthopaedic (33.8 per cent, 2867 of 8492) and head and neck surgery (9.8 per cent, 835 of 8492). Results Patientsįrom 10 029 patients in the cohort study, data from 8492 patients entered the machine learning processes ( Fig.
STEFAN KAISER ESSENTIA FULL
A full description of methods is available in Appendix S1. Finally, to decide which model to select, performance was evaluated through the mean area under the receiver operating characteristic curve (AUROC) value. In order to ascertain the model’s stability, this training and testing split was randomly repeated 100 times (bootstraps). These models were tuned with a 10-fold cross-validation, fitted in the 75 per cent split of the derivation set and assessed in the remaining 25 per cent.
These features were then combined into 26 different predictor sets, which were fitted through three different algorithms (logistic regression, decision trees and random forest), generating a total of 78 different models. Five features were selected following variable importance measurements from both linear and non-linear modelling. Sixteen patient and operative variables were entered into the analysis and, based on time, data were split into a derivation set, where all analysis and modelling were performed, and a validation set, where the final model was assessed. Machine learning methods were used to analyse the COVIDSurg Cohort Study dataset. Full study methodology has been published previously 2. Patients undergoing any type of surgery were included from 1 February 2020 to 31 July 2020.
Methods Cohort study designĪn international, multicentre, prospective, cohort study (COVIDSurg Cohort Study) included consecutive patients who were diagnosed with SARS-CoV-2 in the 7 days before or the 30 days after surgery. The authors aimed to develop and validate a machine learning-based risk score to predict postoperative mortality risk in patients with perioperative SARS-CoV-2 infection. To inform consent and shared decision-making, a robust, globally applicable score is needed to predict individualized mortality risk for patients with perioperative SARS-CoV-2 infection. There is an urgent need to restart surgery safely in order to minimize the impact of untreated non-communicable disease.Īs rates of SARS-CoV-2 infection in elective surgery patients range from 1–9 per cent 3–8, vaccination is expected to take years to implement globally 9 and preoperative screening is likely to lead to increasing numbers of SARS-CoV-2-positive patients, perioperative SARS-CoV-2 infection will remain a challenge for the foreseeable future. Since the beginning of the COVID-19 pandemic tens of millions of operations have been cancelled 1 as a result of excessive postoperative pulmonary complications (51.2 per cent) and mortality rates (23.8 per cent) in patients with perioperative SARS-CoV-2 infection 2.
0 kommentar(er)
