Evidence-based interconversion of the Glasgow Outcome and modified Rankin scales: pitfalls and best practices

Objective: the aim of this study was to provide the evidence base to guide interconversion of the modified Rankin Scale (mRS) and Glasgow Outcome Scale (GOS) in neurological research. Methods: a retrospective analysis of paired mRS and GOS recordings was conducted using datasets with the following selection criteria: (1) patients had haemorrhagic stroke, (2) simultaneous mRS and GOS measurements were available, and (3) data sharing was possible. The relationship between mRS and GOS was assessed using correlation analysis. The optimum dichotomisation thresholds for agreement between the mRS and GOS were identified using Cohen's kappa coefficient. Two-way conversion tables between mRS and GOS were developed based on the highest agreement between scores. Finally, to identify which direction of conversion (mRS to GOS or vice versa) was better, the Kolmogorov-Smirnov D statistic was calculated. Results: using 3474 paired recordings the mRS and GOS were shown to be highly correlated (ρ = 0.90, p < 0.0001). The greatest agreement between the two scoring systems occurred when mRS=0-2 and GOS=4-5 was used to define good outcome (κ=0.83, 95% confidence interval: 0.81–0.85). Converting from mRS to GOS was better than the reverse direction as evidenced by a lower Kolmogorov-Smirnov statistic (D=0.054 compared to D=0.157). Conclusions: this study demonstrates that the mRS and GOS are highly correlated, establishes the optimum dichotomisation threshold for agreement, provides a method for interconversion and shows that mRS to GOS conversion is superior to the reverse direction if a choice is available.