Determining the LVO Denominator: Estimating the Proportion of Large Vessel Ischemic Strokes When CTA is Unavailable
Ananya Iyyangar1, Kimberly Geronimo2, Deepa Dongarwar1, Anjan Ballekere1, Ngoc Mai Le1, Hussain Azeem1, Rania Abdelkhaleq1, Amanda Jagolino-Cole1, Jo Ann Soliven2, Sunil Sheth1, Sushanth R Aroor1
1Neurology, UT Health Houston, 2The Medical City, Pasig, Phillipines
Objective:
To derive, test and externally validate LVO incidence measures using widely available data points, clinical findings, and imaging features on non-contrast head CT.
Background:
Precise determination of the region-specific prevalence of large vessel occlusion (LVO) acute ischemic stroke (AIS) is essential for system-wide planning and identifying access gaps for thrombectomy. Given the variability in LVO reporting (ranging from 13% to 52% of AIS) and absence of CT angiogram availability in most international regions, estimating true prevalence remains challenging.
Design/Methods:
A retrospective chart review of all AIS patients from March to December 2021 at UT Health Houston was performed and data on demographics, initial CT ASPECTS, presence of cortical infarct on CT or MRI done at least 6 hours of presentation, NIHSS, and VAN score were collected. Gold standard was CTA-based definition of LVO, and the model was built using multivariable logistic regression and ROC analysis. Subsequently, the model was externally validated in a cohort of patients with AIS who received vascular imaging at the Medical City, Pasig, Philippines from January to October 2022.
Results:
Among 336 patients that met inclusion criteria at UTHealth, median age was 65, 48.9% were female and median NIHSS was 9. 54.8% patients had LVO confirmed on CTA. A multivariable regression using dichotomized NIHSS (>10), VAN (positive or negative) and presence of cortical changes on CT showed high classification performance (AUC 0.85) with sensitivity of 84% and specificity of 72%. Among the 200 patients that met criteria in the Philippines, a multivariable regression using the dichotomized NIHSS (>10), VAN (positive or negative) and presence of cortical changes on CT also showed high classification performance (AUC 0.79) with sensitivity of 64% and specificity of 86%.
Conclusions:
Using widely available clinical data and non-contrast head CT features, LVO incidence can be accurately identified and may help under resourced settings identify accessibility gaps.