MedWire News: Shift analysis is the best test for detecting relatively uniform benefits across a wide range of stroke severities, whereas dichotomized analysis performs better for detecting a predictable health transition, research suggests.
Shift analysis is also best used to detect unexpected treatment benefits, indicate the findings published in the journal Neurology. The power of dichotomized analysis to detect treatment benefits is partly dependent on the selected point of dichotomy.
Shift analysis assesses how the distribution of scores across an outcomes scale, such as the modified Rankin Scale (mRS), changes in patients treated with a study intervention versus placebo. Dichotomizing a scale such as the mRS focuses exclusively on transitions across the dichotomy point.
Jeffrey Saver and Jeffrey Gornbein from the University of California at Los Angeles, USA, constructed four randomized stroke trial models employing the mRS.
In their model of a neuroprotection trial, expected to confer mild benefits to patients regardless of stroke severity, using shift analysis would require less than half the number of patients needed to detect a treatment benefit if a dichotomized analysis was used.
For an intervention promoting early recanalization, the smallest sample size was possible when the mRS was dichotomized at 0–1 (excellent outcomes) versus 2–6. Shift analysis and the mRS dichotomized at 0–2 (favorable outcomes) versus 3–6, required slightly larger patient cohorts, although shift analysis was best overall if strokes were generally mild.
The benefits of late recanalization were best detected by the mRS dichotomized at favorable outcomes. Shift analysis required more than double the sample size and dichotomizing the mRS at excellent outcomes failed to detect a benefit.
Shift analysis was by far the best option for detecting unexpected benefits, where changes in outcome clustered at an unpredicted point on the mRS. Dichotomizing the mRS at excellent outcomes necessitated a 10-fold increase in sample size, and dichotomizing it at favorable outcomes could not detect the benefit.
The findings from these models held true when applied to six previously published stroke trials, note the researchers.
They conclude: “These insights can guide clinical trialists in selecting the prespecified primary mode of statistical analysis for acute stroke clinical trials.”
Free abstract
