This research study, conducted at the Technical University of Munich and registered at ClinicalTrials.gov (NCT04092933), was designed to develop a machine-learning model that enables personalized perioperative risk prediction by using preoperative data. It was created in accordance with the TRIPOD statement for multivariable prediction models for individual diagnosis, approved by the Ethics Committee of the Medical Faculty of the Technical University of Munich, and conducted adhering to ethical guidelines, German legal regulations, and recommendations of the German Ethics Council. De-identified patient records were used in the research, given that the law on data protection was followed.
The study period was between June 2014 and March 2020 and the endpoints evaluated were in-hospital mortality. Data from different sources such as the hospital information system, laboratory information system and patient data management system were incorporated into the model. This included laboratory values, blood orders, surgical procedure (OPS) codes, medical history and patient medications. Additionally, a time window of two weeks before the respective surgery was used to find the laboratory value closest to the surgery date. Missing values were not imputed and a dichotomous feature of each variable included information about its availability.
The extreme gradient boosting (XGBoost) algorithm was used to develop the prediction models with hyperparameters tuned to optimize the area under the precision-recall curve (AUPRC). This resulted in a dataset of 47,205 used as a training cohort and 11,801 as a test cohort. Additionally, data collected between September 2019 and March 2020 was used to validate the model and the final model included 201 predictors, with an AUROC of 0.93 and an AUPRC of 0.76 for the validation set.
Individual risk profiles were provided for exemplary patients and the variables contributing to the prediction were also visualized in an importance plot. Additionally, partial dependence plots and waterfall plots showed the change in risk as well as the impact of individual variables on the overall risk prediction for each patient, respectively.
The Technical University of Munich’s main focus is faculty-led research, providing students the tools to relish their learning experiences. The Medical Faculty is just one of the many research departments within the University and many of the studies focus on developing new and more efficient medical solutions. Founded in 1868, the Medical Faculty is among the first few departments of the university and has made significant contributions to the research field. In particular, Professor Christian Trautwein, who has experience in the healthcare sector, is the lead researcher of this study. He is acclaimed for his work in data sciences, artificial intelligence and machine learning. His research is highly appreciated and his contributions to the field have been commendable.
