Same Patients, Different Diagnoses: Mapping Six Classifications of Multiple Sclerosis and Recovering the 2×2 Tables Between McDonald Revisions
Before two cohorts are pooled, or a treatment effect is carried from a trial population to a clinic population, something has to fix which patients count alike. Six programs do that in multiple sclerosis, and they sort different things: the diagnostic criteria sort a patient’s accumulated evidence at a moment, the course phenotypes a retrospective course, imaging subtypes an inferred trajectory, the pathological patterns a sampled lesion, drug labels a patient under a product license, and a 2023 position paper describes a continuum with no categories. Two diagnostic rules applied to one cohort of 309 patients moved the disability rate inside the diagnosed group from 46% to 27%, with no patient’s course changed. Of eight full texts bearing on two revisions of the McDonald criteria, none reports an agreement coefficient, one prints the cross-classification, and for one the direction of the change cannot be recovered.