Nomogram for predicting an individual prospective hemorrhage risk in untreated brainstem cavernous malformations

Abstract

OBJECTIVE In this study, the authors aimed to create a nomogram for precisely predicting the 5-year prospective hemorrhage risk in brainstem cavernous malformations (BSCMs). METHODS Patients with confirmed BSCMs in a single-center prospective observational series from January 2012 to December 2016 were included in the present study for nomogram building and validation. The concordance index (C-index), calibration curves, and decision curve analysis were used to evaluate the predictive accuracy, discriminative ability, and clinical usefulness of the nomogram. Then, a nomogram-based risk stratification model for untreated BSCMs was developed. RESULTS In total, 600 patients were included in the study; 417 patients who had been enrolled before July 2015 were divided into the training and validation cohorts, and 183 subsequently enrolled patients were used as the external validation cohort. By applying a backward stepwise procedure in the multivariable Cox model, variables, including prior hemorrhage (HR 1.69), hemorrhage on admission (HR 3.33), lesion size > 1.5 cm (HR 1.84), lesion depth (HR 2.35), crossing the axial midpoint (HR 1.94), and developmental venous anomaly (HR 2.62), were incorporated to develop a nomogram. The Harrell C-index values for a 5-year prospective hemorrhage were 0.752 (95% CI 0.687–0.816), 0.801 (95% CI 0.665–0.936), and 0.758 (95% CI 0.674–0.842) in the training, internal validation, and external validation cohorts, respectively. The nomogram performed well in terms of consistency between prediction and actual observation according to the calibration curve. The patients could be classified into three distinct (low, medium, and high) risk groups using the final score of this nomogram. CONCLUSIONS Independent predictors of the 5-year hemorrhage risk in untreated BSCMs were selected to create the first nomogram for predicting individual prospective hemorrhage. The nomogram was able to stratify patients into different risk groups and assist in clinical decision-making.