Modeling the Predictive Performance of the ICH Score: Stepwise Selection and Logistic Modeling of Mortality, Functional Outcomes, Placement Destination, and Length of Stay

Authors

• Hussain Alkhars, Department of Neurological Surgery, The George Washington University Hospital, Washington, District of Columbia, USA.
• Sean M. Lee, Office of Clinical Statisticians, The George Washington University, Washington, District of Columbia, USA.
• Brij Kathuria, Department of Neurological Surgery, The George Washington University Hospital, Washington, District of Columbia, USA.
• Khashayar Mozaffari, Department of Neurological Surgery, UT Southwestern Medical Center, Dallas, Texas, USA.
• Peter Harris, Department of Neurological Surgery, The George Washington University Hospital, Washington, District of Columbia, USA.
• Max Fleisher, Department of Neurological Surgery, The George Washington University Hospital, Washington, District of Columbia, USA.
• Michael Rosner, Department of Neurological Surgery, The George Washington University Hospital, Washington, District of Columbia, USA.

Document Type

Journal Article

Publication Date

3-1-2026

Journal

Neurosurgery practice

Volume

7

Issue

1

DOI

10.1227/neuprac.0000000000000210

Keywords

Discharge placement; ICH score; Intracerebral hemorrhage; Length of stay; Modified Rankin scale; Mortality; Prognostication

Abstract

BACKGROUND AND OBJECTIVES: The intracerebral hemorrhage (ICH) score is widely used to predict mortality in ICH, but its ability to predict functional outcomes, length of stay (LOS), and discharge placement remains unclear. This study evaluates its predictive performance across these clinical outcomes. METHODS: This retrospective, single-center study included 273 patients admitted with ICH between 2018 and 2023. Logistic regression models assessed the ICH score's predictive performance for in-hospital mortality, modified Rankin Scale (mRS) at discharge, intensive care unit and hospital LOS, and discharge placement. Performance metrics, including accuracy, sensitivity, and specificity, were assessed using leave-one-out cross-validation. Stepwise selection was used to identify the predictive value for the individual ICH score components. RESULTS: The ICH score accurately predicted in-hospital mortality (accuracy: 0.89, sensitivity: 0.80, specificity: 0.91). It showed moderate performance for mRS at discharge (accuracy: 0.67, Kappa: 0.49), particularly for mortality (mRS 6, sensitivity: 0.84, specificity: 0.91), but was less reliable for predicting functional independence (mRS 0-3, sensitivity: 0.59, specificity: 0.85) and severe disability (mRS 4-5, sensitivity: 0.66, specificity: 0.70). The score poorly predicted intensive care unit LOS (accuracy: 0.70, Kappa: 0.39) and hospital LOS (accuracy: 0.53, Kappa: 0.04). While it identified favorable (home/acute rehab) and poor (hospice/death) discharge outcomes, it failed to predict intermediate placement (subacute rehab/long-term care). Stepwise selection consistently excluded infratentorial location, suggesting limited predictive value. CONCLUSION: The ICH score reliably predicts in-hospital mortality but has limited accuracy for functional outcomes, LOS, and discharge placement. While useful for risk stratification, its role in individualized prognostication is limited. Future studies incorporating additional clinical and socioeconomic factors may improve predictive accuracy.

APA Citation

Alkhars, Hussain; Lee, Sean M.; Kathuria, Brij; Mozaffari, Khashayar; Harris, Peter; Fleisher, Max; and Rosner, Michael, "Modeling the Predictive Performance of the ICH Score: Stepwise Selection and Logistic Modeling of Mortality, Functional Outcomes, Placement Destination, and Length of Stay" (2026). GW Authored Works. Paper 8658.
https://hsrc.himmelfarb.gwu.edu/gwhpubs/8658

Department

Neurological Surgery