Kevin S. Attenhofer, MD
As stroke neurologists, we are all intimately familiar with the modified Rankin Scale (mRS) as a measure of degree of disability. It is a common outcome measure in stroke research and can be statistically analyzed as a simple dichotomization or ordinal shift (among other options). The dichotomized outcome takes varied and complex neurological outcomes and simplifies them down to nominal variables of “good” or “bad.” This is statistically more straightforward, but does result in some outcome information being discarded. The ordinal shift retains more of this information, but typically requires larger sample size to maintain adequate power. Even when well powered, however, the mRS still has a disproportionate focus on motor function when compared to other neurological domains, such as cognition or patient metrics such as quality of life.
Detractors of the mRS note that it fails as an assessment of patient perception of quality of life (QoL) despite treatment plans becoming increasingly patient-centered. The mRS simply does not take into account a patient or family’s perceived QoL, partially because of the subjective nature of QoL. In recent years, researchers have sought to improve outcome measures by including patient perception of QoL. The most widely accepted patient-centered outcome measure is the utility approach to health-related quality of life. In the utility-weighted mRS (UW-mRS), utility weighting converts the spacing between ranks on the mRS from arbitrarily fixed intervals to distances that directly reflect patient and societal valuation of outcome and disability states (with values ranging from 0 (death) to 1 (perfect health)).
In this paper, Dijkland et al. used individual patient data from MR CLEAN (Multicenter Randomized Clinical Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands) to assess the statistical power of the UW-mRS in a simulation study. Utility values were determined using the EuroQol Group 5-Dimension Self-Report Questionnaire (EQ-5D-3L) responses of patient, proxy, or healthcare provider at 90 days after inclusion, simultaneously with the mRS. Utility weights for each mRS category were then determined by averaging the derived utilities of all patients within each mRS health state. In this study, the mean utility values for mRS categories 0 to 6 were: 0.95, 0.93, 0.83, 0.62, 0.42, 0.11, and 0, respectively, (which is similar but not identical to the utility weights used in the DAWN (Diffusion-Weighted Imaging or Computerized Tomography Perfusion Assessment With Clinical Mismatch in the Triage of Wake Up and Late Presenting Strokes Undergoing Neurointervention With Trevo) trial).
In the simulations, this outcome data was analyzed by 3 different statistical approaches: treatment effect on the ordinal analysis of mRS, dichotomized mRS, and the UW-mRS using linear regression. While all three showed improved functional outcomes in favor of the intervention, the UW-mRS approach was statistically less efficient in detecting treatment effects compared with the ordinal approach (power 85% vs. 87%) and more efficient than the dichotomized approach (power 85% vs. 71%). This would mean a reduction in statistical power to detect a treatment effect when the UW-mRS is used in randomized trials. Further, some authors have analyzed the UW-mRS with a t test (implying a continuous outcome variable); however, the UW-mRS remains a scale with 7 outcome categories. In this case, the assumption of normally distributed residuals can never be met. This leads to underestimation of the standard error resulting in the actual power of the UW-mRS approach dropping even further.
The UW-mRS in this study is limited by the QoL source data. The EQ-5D-3L was not specifically designed with stroke patients and their needs/deficits in mind. 90 days is also a relatively short follow-up period for QoL metrics. The results of this study suggest that the utility-weighted modified Rankin scale should not replace the ordinal analysis of the mRS as the standard outcome in stroke clinical trials. It is less statistically efficient and clinical interpretability is difficult. Perhaps a more effective outcome measure would be a larger ordinal scale that incorporates more domains of neurological function and patient satisfaction. Unfortunately, the simplicity and high inter-rater reliability of the mRS would likely be sacrificed in such a scale. Regardless of what comes next, efforts in this area highlight an important focus for improvement: incorporation of patients’ valuation on their outcome. While subjective, we cannot ignore that one of the most important outcomes is how the patients themselves view their outcome.