The following is a summary of “Incidence of asymptomatic catheter-related thrombosis in intensive care unit patients: a prospective cohort study,” published in the November 2023 issue of Critical Care by Fan et al.
Prognosis is essential for spinal cord injury (SCI) patients, especially those who need critical care.
Researchers performed a retrospective study to develop and validate a machine-learning (ML) model to predict the discharge destination of spinal cord injury patients in the ICU.
Patient data from an ICU database for those diagnosed with SCI was extracted. The initial patient records were used to create 98 ML classifiers to predict the discharge destination (e.g., death, further medical care, home). The main evaluation metric was the micro-average area under the curve (AUC) for discrimination. The best classifier for average AUC and the best for death sensitivity were combined into an ensemble classifier. The ensemble classifier’s discrimination performance was compared to top death-sensitivity and average-AUC classifiers. Prediction consistency and clinical utility were also evaluated.
The study included 1,485 SCI patients. The ensemble classifier achieved a micro-average AUC of 0.851, slightly lower than the best average AUC classifier (P=0.10). The ensemble classifier had better death sensitivity (0.452) than the best average-AUC classifier. The ensemble classifier’s death sensitivity was lower than the top 8 death-sensitivity classifiers (P<0.05), even though the classifier had a higher micro-average AUC. In decision curve analysis, the ensemble classifier showed a similar Brier score and better Net benefit compared to the original classifiers’ performance.
The study found that the Ensemble classifier predicts discharge destinations better than other methods and could help manage critical SCI patients early on.
Source: journals.lww.com/spinejournal/abstract/9900/early_prognostication_of_critical_patients_with.498.aspx