Abstract W P267: Different Strokes: Causality and Outcomes in the NINDS-tPA Trials

Abstract

Background: Predictive models of outcome after acute ischemic stroke utilize regression methods that discern only associations between variables. Causality, or the nature of relationships between myriad variables collected in clinical trials, has not been considered. We investigated whether NIHSS strata of stroke severity disclose distinct interactions amongst clinical variables, imaging, treatment and outcomes in the NINDS tPA Trials. Methods: The public dataset of the NINDS tPA Trials was analyzed based on NIHSS strata of <10, 10-14, 15-19 and ≥20. For each cohort, causal links were established for relationships between age, sex, co-morbidities, imaging features, timing features, thrombolytic treatment and subsequent 90-day mRS. Each variable was normalized to the interval [0, 1] and missing input samples replaced using regression predictions. The PC causal discovery algorithm used conditional independence tests between variables, incorporating temporal details. Statistical dependencies were established at p-value of 0.1 on the Fisher's z test. Results: 624 subjects (mean age 66.9 years, 58% female) were included. Overall, significant causal links were identified between clinical variables, CT findings and clinical outcomes after thrombolysis. Subgroup analyses on NIHSS strata, however, revealed distinct causal links at each strata. For NIHSS≥20, age and 7-day imaging findings were directly linked with 90-day mRS, followed by treatment and clinical or imaging findings. At NIHSS 15-19, the strength of treatment interaction was not as prominent. At NIHSS 10-14, treatment and 7-day imaging were closest linked with outcome, over other variables. At NIHSS<10, 7-day imaging and clinical variables other than age were closest linked with outcome. Initial imaging variables and even time from onset to treatment exhibited different relationships with outcomes when causality was considered. Conclusions: NIHSS strata of stroke severity in the NINDS-tPA Trials demonstrate different relationships between clinical, imaging and treatment variables with outcomes. Causal relationships, not just associations, provide novel insight on stroke outcomes to inform future clinical trial design and decision-making.