Machine Learning for Prediction of Mortality and Unfavorable Outcome in Traumatic Brain Injury: A CENTER-TBI Study

Abstract

Background: Prediction of outcome in traumatic brain injury is important for risk-stratification of patients. Most prediction models were thus far based on traditional statistical techniques. We aimed to explore the added value of common machine learning algorithms (ML) for prediction of outcome for moderate and severe traumatic brain injury. Methods: We performed logistic regression (LR), lasso regression, and ridge regression with key baseline predictors in the IMPACT database (15 studies, n=11,022). ML algorithms included support vector machines, random forests, gradient boosting machines, and artificial neural networks, and were trained using the same predictors. Regression and ML algorithms were all trained to optimize the log-likelihood. We performed cross-validation by leaving out each study once. The algorithms were further validated externally on the recent CENTER-TBI study (patients with GCS<13, n = 1,554, enrolled between 2014 and 2018). Calibration was assessed graphically and summarized with estimates of calibration-in-the-large and calibration slope. Discrimination was quantified using the c-statistic. Findings: In the IMPACT database, 3,332/11,022 (30%) died and 5,233 (48%) had unfavourable outcome. Outcomes were similar in the CENTER-TBI study, where 348/1,554 (29%) died and 651 (54%) had unfavourable outcome. Discrimination and calibration varied widely between the studies, and ML algorithms did not perform better than regression modeling techniques in most cohorts. The mean c-statistic over all algorithms and validation sets was 0·80 for mortality and 0·79 for unfavourable outcome (0·82 and 0·77 in CENTER-TBI respectively). Interpretation: Flexible machine learning algorithms may not outperform traditional regression approaches in a low-dimensional setting for outcome prediction after moderate or severe TBI. Rather than relying on new algorithms, prediction research in TBI should focus on validation of existing models and their extension with new predictors (biomarkers, imaging, genomics) with strong incremental value. Clinical Trial Registration: ClinicalTrials.gov Identifier: NCT02210221. Funding Statement: Data used in preparation of this manuscript were obtained in the context of CENTER-TBI, a large collaborative project with the support of the European Union 7th Framework program (EC grant 602150). Declaration of Interests: The authors declare to have no competing interests. Ethical Approval Statement: The authors declare that all participants signed informed consent to be included in the study. Ethical approval was obtained for each recruiting sites.