Graphic reanalysis of the two NINDS-tPA trials confirms substantial treatment benefit

Abstract

Background and purpose: Multiple statistical analyses of the 2 NINDS-tPA trials have confirmed study findings of benefit of fibrinolytic therapy. A recent graphic analysis departed from best practices in the visual display of quantitative information by failing to take into account the skewed functional importance of NIH Stroke Scale raw scores and by scaling change axes at up to 20 times the range achievable by individual patients.

Methods: Using the publicly available datasets of the 2 NINDS-tPA trials, we generated a variety of figures appropriate to the characteristics of acute stroke trial data.

Results: A diverse array of figures all visually delineated substantial benefits of fibrinolytic therapy, including: bar charts of normalized gain and loss; stacked bar, bar, and matrix plots of clinically relevant ordinal ranks; a time series stacked line plot of continuous scale disability weights; and line plot, bubble chart, and person icon array graphs of joint outcome table analysis. The achievable change figure showed substantially greater improvement among tPA than placebo patients, median 66.7% (interquartile range, 0 to 92.0) versus 50.0% (interquartile range, -7.1 to 80.0), P=0.003.

Conclusions: On average, under 3 hour patients treated with tPA recovered two-thirds while placebo patients improved only half of the way toward fully normal. Graphical analyses of the 2 NINDS-tPA trials, when performed according to best practices, is a useful means of conveying details about patient response to therapy not fully delineated by summary statistics, and confirms a valuable treatment benefit of under 3 hour fibrinolytic therapy in acute stroke.

Figure 1

High school test scores illustrating importance of incorporating variations of functional significance across scale ranges into visual data displays. A) Bar chart showing the nonlinear functional meaningfulness of high school test scores. All important letter grade transitions in subject mastery are clustered to the far right of the raw score distribution. B) Pie chart showing the effects on of a new teaching method on the grades of 66 students. Overall the intervention has substantial benefits. Compared with standard teaching approaches, the intervention yields higher grades for 22 (33%) students, including dramatically higher grades for 9 (14%), lower grades for2 (3%) students, and the same grades for 42 (64%) students. C) Graph of the effects of the teaching intervention using the display techniques of Hoffman and Schriger. Because of the extreme nonlinearity of raw test score functional importance, the intrinsic characteristics of cumulative distribution displays, and the axis scaling at a range up to 20 times the change score achievable by individual students, the educational intervention appears to have little effect. D) Stacked bar chart of final letter grade outcomes under standard and novel teaching strategies demonstrates the substantial benefit of the novel intervention obscured in the cumulative distribution plot.

Figure 2

Data from the two NINDS-TPA trials similarly demonstrates importance of incorporating variations of functional significance across scale ranges into visual data displays. A) Bar chart showing the nonlinear functional meaningfulness of NIHSS scores. Important transitions in global disability (reflected in modified Rankin Scale levels 0–6) are clustered to the far right or far left of the raw score distribution. R6 = dead, R5 = bedridden, needing constant care, R4 = dependent and unable to walk, R3 = dependent but able to walk, R2 = independent in daily living but vocationally disabled, R1 = neurologic symptoms without disability, R0 = symptom-free. B) Pie chart showing the findings of the NINDS TPA trials for the effect of TPA on final outcome compared with supportive care, using joint outcome table analysis. Overall the intervention has substantial benefits. For every 100 patients treated, TPA improves final outcomes in 32, including complete or near complete recovery in 8%, worsens final outcome in 3%, and does not alter final outcome in 65%. C) Cumulative distribution graph shown by Hoffman and Schriger of the delta NIHSS results in the two NINDS TPA trials. Green lines are TPA groups and blue lines placebo groups. Because of the extreme nonlinearity of raw NIHSS score functional importance, the intrinsic characteristics of cumulative distribution displays, and the axis scaling at a range up to 20 times the change score achievable by individual patients, the lytic intervention appears to have little effect. D) Stacked bar chart of final global disability outcomes in the two NINDS-TPA trials the benefit of lytic treatment obscured in the cumulative distribution plot.

Figure 3

Cumulative distribution curve comparing supportive care outcomes in the NINDS TPA Trials with Miraculous Plasminogen Activator, an impossibly effective treatment that cures every patient. The area between the curves, 17% of the area between the axes, is the highly compressed region in which therapeutic effects can be displayed (gray color). The regions outside the curves are dead space (red color), uninformative regarding treatment effects, compacting the informative zone.

Figure 4

Normalized gain and loss bar chart showing degree of improvement and worsening after treatment among all 624 patients enrolled in the two NINDS-TPA trials. TPA patients on average recover 2/3 of the way toward normal vs half-way toward normal for placebo patients, with a particular increase in the rate of complete recovery with TPA.

Figure 5

Heptile matrix bar charts showing baseline and day 90 Rankin levels (mapped from the NIHSS) for all 624 patients in the two NINDS TPA trials. The TPA matrix shows greater rightward shift toward normal and near normal outcomes across all pretreatment levels of stroke deficit severity. Cutpoints for mapping of NIHSS scores to Rankin levels were: Rankin 0 - NIHSS 0; Rankin 1 – NIHSS 1–2; Rankin 2 – NIHSS 3–4; Rankin 3 – NIIHSS 5–8; Rankin 4 – NIHSS 9–17; Rankin 5 – NIHSS 18–41; Rankin 6 – NIHSS 42. B = baseline. Rankin* = Rankin mapped from the NIHSS.

Figure 6

A) Graph of the prespecified primary hypothesis of NINDS Trial 2. The trial tested whether TPA would increase the proportion of patients achieving normal or near normal outcomes on four outcome scales. NIHSS – National Institute of Health Stroke Scale; BI – Barthel Index; mRS – modified Rankin Scale; GOS – Glasgow Outcome Scale. Consistent evidence of benefit was noted across all four endpoints. B) Graph depicting the absolute reduction in events by major acute myocardial infarction and cerebral infarction treatments. The absolute benefits of TPA observed in the NINDS-TPA trials are substantially greater than with cardiac interventions, explaining the lower sample size needed to demonstrate convincing treatment effects. Data from references 3, 26, 37, and 38., , , Favorable recovery = Rankin Scale 0–1. Reduced disability = lower score on the Rankin Scale by 1 or more levels. PTA – percutaneous transluminal angioplasty; MI – myocardial infarction.

Figure 7

Stacked area plot of disability burden throughout the first year after stroke among all patients enrolled in the two NINDS-TPA trials. Color bar shows the hue intensities assigned to the poles and midpoint of the WHO disability weight, and displayed colors reflect the exact value of the DW along this continuous spectrum. Patients in both treatment groups begin with severe disability burdens (red and orange predominate at 0 timepoint). Among those who survive their stroke (numerically more with TPA than placebo throughout the 1 year period), substantially less disability is experienced by TPA patients throughout the first poststroke year.

Figure 8

Joint outcome figures based on data from the two NINDS-TPA Trials. A) Line plot showing final global disability outcomes of 100 patients if they receive placebo or TPA treatment. Green indicates better outcome with TPA, red worse. B) Bubble plot of same data. C) Person icon array, decision-aid figure showing among 100 treated patients the change in outcomes that will occur with TPA rather than placebo. (Figure 8C reproduced with permission of UCLA Stroke Center).