Background and purpose: Baseline stroke severity predicts outcomes among thrombolysed patients. The baseline National Institutes of Health Stroke Scale (NIHSS) thresholds are sometimes used to select patients for thrombolysis, clinical trial enrollment, or both. Using data lodged with Virtual International Stroke Trials Archive, we compared adjusted outcomes between thrombolysed and nonthrombolysed patients enrolled in neuroprotection trials (1998-2007) to assess the influence of various levels of baseline NIHSS. Method-We assessed the association of treatment with outcome, measured across the modified Rankin scale score distribution, in patients categorized by baseline NIHSS in increments of 4. We used an age and baseline NIHSS adjusted Cochran-Mantel-Haenszel test followed by proportional odds logistic regression analysis. We report the Cochran-Mantel-Haenszel P values and estimated odds ratios (OR) for improved modified Rankin scale score distribution with treatment for patients within each baseline NIHSS category.
Results: Data were available for 5817 patients (1585 thrombolysed and 4232 nonthrombolysed). Baseline severity was greater among thrombolysed than nonthrombolysed (median baseline NIHSS, 14 vs 13; P < 0.05). An association of treatment with outcome was seen independently and was of similar magnitude within each of the baseline NIHSS categories 5 to 8 (P=0.04; OR, 1.25; 95% confidence interval [CI], 1.0-1.6; N = 278/934 thrombolysed/nonthrombolysed), 9 to 12 (P = 0.01; OR, 1.3; 95% CI, 1.1-1.6; N = 404/942), 13 to 16 (P < 0.05; OR, 1.6; 95% CI, 1.3-2.1; N = 342/814), 17 to 20 (P < 0.05; OR, 1.7; 95% CI, 1.3-2.1; N = 311/736), and 21 to 24 (P < 0.05; OR, 1.6; 95% CI, 1.1-2.1; N = 178/466). No association was observed within baseline NIHSS categories 1 to 4 (P = 0.8; OR, 1.1; 95% CI, 0.3-4.4; N = 8/161) or ≥ 25 (P = 0.08; OR, 1.1; 95% CI, 0.7-1.9; N = 64/179).
Conclusions: In this nonrandomized comparison, outcomes after thrombolysis were significantly better than in untreated comparators across baseline NIHSS 5 to 24. The significant association was lost only at extremes of baseline NIHSS when sample sizes were small and confidence limits were wide.