Mark N. Rubin, MD
In acute stroke therapy studies, we are used to seeing “onset to treatment” (OTT) as an important variable. OTT gives the reader a sense of “The Machine:” how quickly can someone activate emergency care personnel, get transported to the appropriate hospital and have guideline-based treatment administered? When we scrutinize OTT between trials, or at our own stroke center meetings after practice improvement initiatives, we frequently compare median OTT. This is mostly probably because…it’s easy to understand, and perhaps we have a sense that – given the incredible heterogeneity from one case to the next from stroke syndrome to treatment approach – the median “corrects” these disparities.
Some colleagues in the UK suggest otherwise. They point out that descriptive statistics such as the median fail to clearly demonstrate the relative distribution a particular intervention may have had within a cohort. As an example, if median OTT goes down between two chronologically-consecutive cohorts, is it because everyone received earlier treatment? Did some get ultra-early treatment and the rest of the cohort stay relatively the same? These investigators suggest a statistical method novel to stroke research, relative distributions, may be more informative than descriptive statistics.
They applied the principles of relative distributions to the SITS-UK stroke registry, selecting a reference cohort from 1/2003-1/2007 with a median OTT of 160min and a comparator cohort from 10/2009-9/2010 with a median OTT of 145min. The natural hypothesis from comparison of the medians is that doctors and systems became more experienced and/or recognition among the community has increased over those time periods. Applying the relative distribution techniques demonstrated that only some of the latter cohort actually received faster treatment and that a fair amount of the cohort actually received later treatment as compared to the reference cohort. They acknowledge this is multifactorial and reasons are speculative, but the point is we can be more granular with our scrutiny of the system with an assessment based on relative distributions rather than medians, because it’s just not that simple!