Skip to main content
Dryad logo

Data from: Ordinal vs dichotomous analyses of modified Rankin Scale and 5-year outcome and cost of stroke

Citation

Ganesh, Aravind; Luengo-Fernandez, Ramon; Wharton, Rose M.; Rothwell, Peter M. (2019), Data from: Ordinal vs dichotomous analyses of modified Rankin Scale and 5-year outcome and cost of stroke, Dryad, Dataset, https://doi.org/10.5061/dryad.609bp7m

Abstract

Objective – To compare how three common representations (ordinal versus dichotomized as 0-1/2-6 or 0-2/3-6) of the modified Rankin Scale (mRS) – a commonly-used trial outcome measure – relate to long-term outcomes, and quantify trial ineligibility rates based on pre-morbid mRS. Methods – In consecutive ischaemic stroke patients in a population-based, prospective cohort study (Oxford Vascular Study; 2002-2014), we related 3-month mRS to 1-year and 5-year disability and death (logistic regressions), and health/social-care costs (generalized linear model), adjusted for age/sex, and compared goodness-of-fit values (c-statistic, Mean Absolute Error [MAE]). We also calculated the proportion of patients in whom pre-morbid mRS >1 or >2 would result in exclusion from trials using dichotomous analysis. Results – Among 1,607 patients, the ordinal mRS was more strongly related to 5-year mortality than both the 0-1/2-6 and 0-2/3-6 dichotomies (all p<0.0001). Results were similar for 5-year disability, and 5-year care costs were also best captured by the ordinal model (ΔMAE versus age/sex: -3,059 for ordinal, -2,805 for 0-2/3-6, -1,647 for 0-1/2-6). 244 (17.1%) 3-month survivors had pre-morbid mRS>2 and 434 (30.5%) had mRS>1; both proportions increased with female sex, socio-economic deprivation, and age (all p<0.0001). Conclusions – The ordinal form of the 3-month mRS relates better to long-term outcomes and costs in survivors of ischaemic stroke than either dichotomy. This finding favours using ordinal approaches in trials analysing the mRS. Exclusion of patients with higher pre-morbid disability by use of dichotomous primary outcomes will also result in unrepresentative samples.

Usage Notes