Predicting hematoma expansion using machine learning: An exploratory analysis of the ATACH 2 trial

Arooshi Kumar; Jens Witsch; Jennifer Frontera; Adnan I Qureshi; Eric Oermann; Shadi Yaghi; Kara R Melmed

doi:10.1016/j.jns.2024.123048

Predicting hematoma expansion using machine learning: An exploratory analysis of the ATACH 2 trial

J Neurol Sci. 2024 May 12:461:123048. doi: 10.1016/j.jns.2024.123048. Online ahead of print.

Authors

Arooshi Kumar¹, Jens Witsch², Jennifer Frontera³, Adnan I Qureshi⁴, Eric Oermann³, Shadi Yaghi⁵, Kara R Melmed⁶

Affiliations

¹ Rush University Medical Center, Department of Neurology, Chicago, IL 60612, United States of America. Electronic address: arooshi.kumar@gmail.com.
² Hospital of the University of Pennsylvania, Department of Neurology, Philadelphia, PA 19104, United States of America.
³ NYU Langone Medical Center, Department of Neurology, New York, NY 10016, United States of America.
⁴ Zeenat Qureshi Stroke Institutes and Department of Neurology, University of Missouri, Columbia, MO 65201, United States of America.
⁵ Warren Alpert Medical School of Brown University, Department of Neurology, Providence, RI 02903, United States of America.
⁶ NYU Langone Medical Center, Department of Neurology, New York, NY 10016, United States of America; NYU Langone Medical Center, Department of Neurosurgery, New York, NY 10016, United States of America.

PMID: 38749281
DOI: 10.1016/j.jns.2024.123048

Abstract

Introduction: Hematoma expansion (HE) in patients with intracerebral hemorrhage (ICH) is a key predictor of poor prognosis and potentially amenable to treatment. This study aimed to build a classification model to predict HE in patients with ICH using deep learning algorithms without using advanced radiological features.

Methods: Data from the ATACH-2 trial (Antihypertensive Treatment of Acute Cerebral Hemorrhage) was utilized. Variables included in the models were chosen as per literature consensus on salient variables associated with HE. HE was defined as increase in either >33% or 6 mL in hematoma volume in the first 24 h. Multiple machine learning algorithms were employed using iterative feature selection and outcome balancing methods. 70% of patients were used for training and 30% for internal validation. We compared the ML models to a logistic regression model and calculated AUC, accuracy, sensitivity and specificity for the internal validation models respective models.

Results: Among 1000 patients included in the ATACH-2 trial, 924 had the complete parameters which were included in the analytical cohort. The median [interquartile range (IQR)] initial hematoma volume was 9.93.mm³ [5.03-18.17] and 25.2% had HE. The best performing model across all feature selection groups and sampling cohorts was using an artificial neural network (ANN) for HE in the testing cohort with AUC 0.702 [95% CI, 0.631-0.774] with 8 hidden layer nodes The traditional logistic regression yielded AUC 0.658 [95% CI, 0.641-0.675]. All other models performed with less accuracy and lower AUC. Initial hematoma volume, time to initial CT head, and initial SBP emerged as most relevant variables across all best performing models.

Conclusion: We developed multiple ML algorithms to predict HE with the ANN classifying the best without advanced radiographic features, although the AUC was only modestly better than other models. A larger, more heterogenous dataset is needed to further build and better generalize the models.

Keywords: Hematoma expansion; Intracerebral hemorrhage; Machine learning.