Prevention of irreversible disability is currently the most important goal of disease modifying therapy for multiple sclerosis. The disability outcomes used in most clinical trials rely on progression of Expanded Disability Status Scale score confirmed over 3 or 6 months. However, sensitivity and stability of this metric has not been extensively evaluated. Using the global MSBase cohort study, we evaluated 48 criteria of disability progression, testing three definitions of baseline disability, two definitions of progression magnitude, two definitions of long-term irreversibility and four definitions of event confirmation period. The study outcomes comprised the rates of detected progression events per 10 years and the proportions of the recorded events persistent at later time points. To evaluate the ratio of progression frequency and stability for each criterion, we calculated the proportion of events persistent over the five subsequent years once progression was achieved. Finally, we evaluated the clinical and demographic determinants characterising progression events and, for those that regressed back to baseline, determinants of their subsequent regression. The study population consisted of 16 636 patients with the minimum of three recorded disability scores, totalling 112 584 patient-years. The progression rates varied between 0.41 and 1.14 events per 10 years, with the length of required confirmation interval as the most important determinant of the observed variance. The concordance among all tested progression criteria was only 17.3%. Regression of disability occurred in 11–34% of the progression events over the five subsequent years. The most important determinant of progression stability was the length of the confirmation period. For the most accurate set of the progression criteria, the proportions of 3-, 6-, 12- or 24-month confirmed events persistent over 5 years reached 70%, 74%, 80% and 89%, respectively. Regression post progression was more common in younger patients, relapsing-remitting disease course, and after a smaller change in disability, and was inflated by higher visit frequency. These results suggest that the disability outcomes based on 3–6-month confirmed disability progression overestimate the accumulation of permanent disability by up to 30%. This could lead to spurious results in short-term clinical trials, and the issue may be magnified further in cohorts consisting predominantly of younger patients and patients with relapsing-remitting disease. Extension of the required confirmation period increases the persistence of progression events.