Purpose Major trials assessing the clinical efficacy of mechanical thrombectomy show some unexpected variability in statistically significant variables.^1-4 This abstract will assess the underlying variability of National Institute of Health stroke scale (NIHSS) subscores, and provide proof of concept for its effect on major clinical trials.

Methods 300 consecutive patients evaluated with CT perfusion for stroke were retrospectively analyzed. Patients were included if thrombectomy was attempted and excluded if the pre-intervention dataset was incompletely documented, including incomplete documentation of NIHSS. 53 patients were included in the study. Principal component analysis was performed on the NIHSS subscores for feature reduction with retention of 95% of variance. Three models were made, using composite score (CS), individual subscores (IS), and principal components (PC). Using data available prior to the decision to intervene, whole data logistic regression models were produced to analyze effect of NIHSS on discharge outcome as assessed by modified rankin scale. Discharge location was also assessed. Backward stepwise elimination was used to reduce included variables, with a p-value threshold of 0.1 used for inclusion. Furthermore sampling with repletion was used to divide the data into training and testing subsets, with a 0.8 sampling ratio. Logistic regression models were built on the training data using composite score, individual subscores, and principal components. T-testing with Bonferroni multiple testing correction compared models for accuracy and receiver operating characteristic curves area under the curve (AUC) based on testing data.

Results Predictive model analysis of mRS, PC based models (accuracy 61.79+/-9.90 AUC.704+/-0.111) outperformed IS (accuracy 60.88+/-10.42 AUC 0.605+/-0.130) and composite score (accuracy 59.23+/-10.49, AUC 0.632+/-0.132) with a p value<0.001 for AUC. For accuracy PC also showed a statistically significant improvement in accuracy over CS with p<0.02. Similar results were seen for discharge location, with statistically significant improvement of PC models over CS models in AUC and accuracy. For both discharge location and mRS, analytic handling of NIHSS subscores produces variation in what factors are statistically significant, altering included variables and their p values.

Conclusions NIHSS composite score is a suboptimal predictor for assessed patient outcomes. Principal component analysis improves predictive value of assessment with a statistically significant effect for pre-intervention patient outcomes analysis. Statistically significant variables are heavily dependent of NIHSS subscores, and variation in the distribution of these subscore may represent a significant source of bias.

References

1. . OA Berkhemer, et al. MR CLEAN, a multicenter randomized clinical trial of endovascular treatment for acute ischemic stroke in the Netherlands: new england journal of med 372;1 nejm.org january 1, 2015.

2. . Campbell BCV, et al. Endovascular therapy for ischemic stroke with perfusion-imaging selection. N Engl J Med. 2015;372:1009–1018.

3. . Ciccone A, et al. Endovascular treatment for acute ischemic stroke. N Engl J Med2013;368:904–913

4. . Linfante I, et al. Predictors of poor outcome despite recanalization: a multiple regression analysis of the NASA registry. J Neurointerv Surg. 2016;8:224–229.

Disclosures S. Arndt: None. A. Albar: None. G. Bennett: None. J. Lavie: None. P. Gulotta: None. J. Milburn: None.