Random effects adjustment in machine learning models for cardiac surgery risk prediction: a benchmarking study

ABSTRACT Objectives There is an ongoing debate over whether a procedural specific (e.g. Society of Thoracic Surgeons (STS)) or universal model (e.g. EuroSCORE II (ES II)) should be used for patient selection in cardiac surgery. Recently, we showed that ES II suffers from severe performance drift across several important metrics and that ML approaches such as Xgboost and Random Forest are substantially more resistant to dataset drift. With the growing interest in big data and its leverage through the use of ML approaches that are not limited by linear statistical assumptions, the number of clinical variables can theoretically increase exponentially. In addition, the variations and residual confounding that historically hindered the usefulness of cardiac risk stratification scores can potentially be taken into account. Here, we assess these possibilities on a large United Kingdom (UK) database. Methods A retrospective analysis of prospectively routinely gathered data on adult patients undergoing cardiac surgery in the UK between 2012-2019. We temporally split the data 70:30 into a training and validation subset. Two sets of seven ML mortality prediction models, with and without variable selection were assessed for consensus Clinical Effective Metric (CEM) overall performance and performance within each of CEM’s consistuent metrics. Confounding and potential causal relationships between covariates and outcomes were evaluated using bayesian network analysis. Results A total of 227,087 adults underwent cardiac surgery during the study period with a mortality rate of 2.76%. For non-variable selected (NVS) risk scores with 102 variables, Xgboost with adjustment for hospital variation was superior to the Xgboost without adjustment (p < 2e-16). Both NVS and the 18 variables selected (VS) Xgboost with adjustment for hospital variation risk scores were superior to the Xgboost (ES II 18 variables) model (p < 6.3e-15), with NVS Xgboost with adjustment for hospital variation having the best performance, followed by the VS Xgboost with adjustment for hospital variation (CEM Difference: 0.0150 and 0.0023, respectively). Conclusions We have identified an ML adjusted risk score comprising 102 variables that increases risk stratification performance on hold out dataset, removing the need to perform variable selection and reduction. This paves the way for further research that utilises this new set of variables with hospital-based adjustments for the safer selection of patients undergoing cardiac surgery.