Multivariate Models using quantitative CT assessment for prediction of outcome after Subarachnoid Hemorrhage

ObjectiveTo diagnose if a patient has subarachnoid hemorrhage (SAH) and to predict if he/she will develop angiographic vasospasm/delayed cerebral ischemia (DCI) or will die using only admission cranial computed tomography (CT) and World Federation of Neurological Surgeons (WFNS) or Hunt-Hess clinical grades. MethodsWe developed a semiautomatic method of quantification of SAH volume, surface, sphericity, fractal dimension, and other parameters by retrospectively analyzing admission CT scans of 127 SAH patients. A multivariable model to predict the development of vasospasm/DCI and death was fitted by combining CT and admission neurological grade with derived data. It was also developed and then validated with bootstrapping and with a separate group of patients. ResultsThe diagnosis of SAH can be given by surface to volume ratio of the 60-80 HU volume of interest for blood AUC 0.98 (95% CI 0.95-1), hemorrhage volume 0.92 (95% CI 0.85-0.99), sphericity 0.87 (95% CI 0.73-1) and surface area 0.75 (95% CI 0.56-0.94). For vasospasm/DCI prediction AUC of biomarker panel was 0.731 (95%CI 0.626-0.836) which was significantly different from random (p<0.001) while Hunt-Hess gave 0.653 (95%CI 0.535-0.772). We established one individual image derived predictor of death by the surface to volume ratio at the level of AUC (Area under the curve) of 0.777 (95% CI 0.667 - 0.886), while a biomarker panel gave the value of AUC of 0.857 (95%CI 0.769-0.944). These two biomarkers allow the correct prediction of the death of nine of ten patients with 36% and 59% probability, respectively. ConclusionsThe model allows rapid, objective, and quantitative assessment of the risk of vasospasm/DCI and death in SAH patients. Both assessments, being fully automatic, can be easily introduced to CT scanner reconstruction software. Key messageO_LIWhat is already known on this topic - Subarachnoid hemorrhage (SAH) has an incidence of 7.9 per 100 000 inhabitants per year and morbidity of almost 90%. It is typically diagnosed with a cranial CT-scan by a radiologist. The amount of blood on the CT scan is assessed qualitatively using the Fisher score and clinical evaluation through the WFNS and Hunt-Hess scales. The objective was to establish some unitary predictor for diagnosis and biomarker model predictors of death and vasospasm after distinguishing from control cases. C_LIO_LIWhat this study adds - Contrary to common sense knowledge the most significant factor in increasing of intra-hospital death probability is not the hemorrhage volume, but its surface-to-volume ratio derived from the image volume-of-interest delineated at 60-80 HU. C_LIO_LIHow this study might affect research, practice or policy - This model could help physicians to take decisions on redirecting the patient to more monitored hospital areas, anticipate possibilities of complications and apply endovascular treatment with calcium antagonists or balloon angioplasty. C_LI