Applications of Machine Learning Model for Prediction of Outcomes in Primary Pontine Hemorrhage.

BACKGROUND

Primary pontine hemorrhage (PPH) is a rare intracranial hemorrhage with a wide range of mortality rate. Predicting the prognosis of PPH is still challenging. Previous prognostic scoring tests have not been widely used due to limited external validation. This study applied machine learning (ML) algorithms to develop predictive models for mortality and prognosis of patients with PPH.

METHODS

Data of patients with PPH were retrospectively reviewed. Seven ML models were used to train and validate for predicting outcomes of PPH including 30-day mortality rate, 30-day, and 90-day functional outcomes. Accuracy, sensitivity, specificity, positive and negative predictive value, F1 score, Brier score, and area under the curve (AUC) of the receiver operating characteristic were calculated. The models with the highest AUC were then selected to evaluate the testing data.

RESULTS

One hundred and fourteen patients with PPH were included. Mean hematoma volume was 7 ml and most patients had hematoma in the central part of the pons. The 30-day mortality rate was 34.2% and favorable outcomes were observed in 71.1% and 70.2% during 30-day and 90-day follow-up. The ML model could predict 30-day mortality with an AUC of 0.97 using an artificial neural network. Regarding functional outcome, the gradient boosting machine could predict both 30-day and 90-day outcomes with an AUC of 0.94.

CONCLUSIONS

ML algorithms achieved a high performance and accuracy in predicting PPH outcomes. Despite the need for further validation, ML models are promising tools for clinical applications in the future.