Thrombolysis in acute ischaemic stroke: Author response to IST-3 investigators
Subgroup analyses can be informative but not definitive. We should neither consider them the last word nor ignore them. In the original IST-3 trial report, subgroup results by time window warrant greater attention because (1) the trial entry criteria excluding patients with a “clear indication” for alteplase (at a time when administration within 3 hours was a major factor in indications) results in potentially different populations and (2) the results reported for the primary analysis are substantially different in these subgroups.
The IST-3 trial reported a primary outcome of being alive and independent (Oxford Handicap Score 0-2) at 6 months, equivalent to modified Rankin Scale score of 0-2. Among 849 patients randomized within 3 hours of stroke onset, 30.6% those treated with alteplase and 22.7% controls were alive and independent, a 7.9% absolute increase (NNTB 13). Among 1,177 patients randomized 3-4.5 hours after stroke onset, 31.5% those treated with alteplase and 37.7% controls were alive and independent, a 6.2% absolute decrease (NNTH 16).
A test of heterogeneity of treatment effect across the three subgroups by time from randomization (0-3 hours, 3-4.5 hours, 4.5-6 hours) was not statistically significant using a multivariate risk-based analysis for this prespecified primary subgroup analysis.[1,2,3] Logistic regression has been considered unreliable for detecting differences in causal effects between subgroups if the control event rates are not rare and are different between subgroups, and the control event rates in the key subgroups of interest here were 22.7% and 37.7%. An unadjusted analysis finds a highly statistically significant subgroup interaction, but would be considered a post hoc analysis. An interaction by time window is neither reliably established nor excluded, just uncertain with a failure to demonstrate with statistical significance in the prespecified statistical approach. The best test of validity of subgroup-treatment effect interactions is not statistical significance but reproducibility in other trials.
In reporting the absolute results for the 3-4.5 hour subgroup (NNTH 16) the finding is statistically significant if using a 95% confidence interval and not statistically significant if using a 99% confidence interval. The selection of the statistical threshold to report for this analysis is irrelevant for objective understanding of the results and misleading if used as the key conclusion in either direction. Interpreting statistical significance as a definitive sign of harm from a single outcome (even the primary outcome) from a single trial would be overinterpretation. Interpreting absence of statistical significance for this analysis (or for an analysis of interaction by subgroup) as negating this result would be underinterpretation.
The findings from IST-3 for the primary outcome provide a suggestion that alteplase could be harmful and an even more substantial suggestion of absence of benefit in the 3-4.5 hour time window. This provides uncertainty and concern that use of alteplase 3-4.5 hours after stroke could be harmful without benefit. This concern is not alleviated by reports of secondary outcomes, overall 0-6 hour analyses, or prespecified analyses for other findings.
For the best interests for patient care we should consider these findings a significant call for caution, determine the most valid way to analyze available data for the specific question of safety and efficacy of alteplase 3-4.5 hours after symptom onset (rather than rely on previously reported analyses not specific for this focus), and conduct that analysis for a thorough view with reporting of many outcomes (mRS 0-1, mRS 0-2, ordinal shift analysis, symptomatic intracranial hemorrhage, mortality). The result of such an analysis is likely to change guidance and target additional research. In the interim clinicians should consider this uncertainty while trying to make optimal clinical care decisions for individual patients.
 Sandercock P, Wardlaw JM, Lindley RI, Dennis M, Cohen G, Murray G, Innes K, Venables G, Czlonkowska A, Kobayashi A, Ricci S, Murray V, Berge E, Slot KB, Hankey GJ, Correia M, Peeters A, Matz K, Lyrer P, Gubitz G, Phillips SJ, Arauz A. The benefits and harms of intravenous thrombolysis with recombinant tissue plasminogen activator within 6 h of acute ischaemic stroke (the third international stroke trial [IST-3]): a randomised controlled trial. Lancet 2012 June 23;379(9834):2352-63.
 Sandercock P, Lindley R, Wardlaw J, Whiteley W, Murray G, IST-3 Collaborative Group. Statistical analysis plan IST-3. IST-3 SAP v5.3.2. 2012. Int J Stroke 2012;7:186-7.
 Kent DM, Rothwell PM, Ioannidis JPA, Altman DG, Hayward RA. Assessing and reporting heterogeneity in treatment effects in clinical trials: a proposal. Trials 2010;11:85.
 Shrier I, Pang M. Confounding, effect modification, and the odds ratio: common misinterpretations. J Clin Epidemiol. 2015;68:470-4.
 Sullivan KM, Dean A, Soe MM. OpenEpi: A web-based epidemiologic and statistical calculator for public health. Public Health Rep. 2009;124:471-4.
 Rothwell PM. Treating Individuals 2. Subgroup analysis in randomised controlled trials: importance, indications, and interpretation. Lancet 2005;365:176-86.
Competing interests: Authors of the original BMJ article; BSA, MMM, and JSM work for EBSCO Health which publishes DynaMed; no financial competing interests with any stroke-related products