Veilleux et al. ^{Reference Veilleux, Demchuk and Sutherland1} surveyed Canadian clinicians to investigate their practice patterns for spontaneous lobar intracerebral hemorrhage (typically nonsurgical) and whether the results of the recently published ENRICH trial on minimally invasive evacuation^{Reference Pradilla, Ratcliff and Hall2} (positive for surgery) would make them change their practice. One problem may be that clinicians were under-impressed by the use of mean utility-weighted modified Rankin scale (uw-mRS) scores and Bayesian credible intervals.

The primary outcome is the most important aspect of trial design, and the way it is analyzed should be as simple and comprehensible as possible. In stroke trials, the mRS is most frequently used, and clinicians are familiar with the meaning of the various grades. The mRS has often been analyzed as a dichotomous endpoint set at 0–2 versus 3–6 (patients are dependent or not), but recently, an “mRS shift analysis,” more sensitive for small differences between groups, is more difficult to understand.^{Reference Lees, Zivin, Ashwood, Davalos, Davis and Diener3,Reference Saver and Gornbein4} For ENRICH, the ordinal mRS score was transformed into the continuous uw-mRS, which permits the use of statistical tests that are more likely to be significant. But what does a “mean uw-mRS” mean? The transformation of an ordinal endpoint into a continuous variable is suspect; in the past, it has led to aberrations, such as trials of unbelievably small sizes.^{Reference Raymond, Darsaut, Eneling and Chagnon5}

In ENRICH, the mean uw-mRS at 180 days improved from 0.374 in the control (no surgery) group (somewhere between mRS 3 and 4) to 0.458 in the surgery group (also between mRS 3 and 4).

For alert readers of ENRICH, this should immediately raise the concern of statistical significance versus clinical significance. One could interpret the result such that it is not worth the time, effort and resources of surgery if the end result is still a dependent patient.

The conversion of a familiar and clinically meaningful endpoint into a new continuous variable (and the use of less-familiar Bayesian statistics) may explain why the Canadian survey responders were not impressed. Had the trialists chosen frequentist statistics and a traditional 0–2 dichotomized mRS, the result would have been more easily interpretable: Figure 2 in [Reference Pradilla, Ratcliff and Hall2] shows that independent patients were significantly more numerous in the surgical group (47/147 [32%; 95% CI: 25–40]) than in the medical group (26/139 [19%; 95% CI: 13–26]), with a P-value of 0.014 (Fisher’s exact test).

We endorse the authors’ suggestions for more clinical trials to be done in this field, using outcome measures that speak to clinician and patient interests.