Footnotes

• Competing interests None declared.

• Provenance and peer review Commissioned; externally peer reviewed. This paper was reviewed by Colin Mumford, Edinburgh, UK, and Jamie Stankiewicz, Boston, USA.

• ↵i Both methods look to predict the behaviour of a cohort over time based on their entry data alone, using figures derived from a reference population followed longitudinally (in our case from the British Columbia multiple sclerosis (BCMS) data set). For full details see refs. 4 and 5.

  Markov model: Markov modelling is based on the chance of an individual, at reassessment, staying in the same EDSS state or moving to a different one. Probabilities were calculated for each transition from the BCMS data set and then applied to the risk-sharing scheme baseline data. The result of the modelling is expressed as the predicted percentage of individuals in each EDSS state at any time point.

  Multilevel model: projected curves of EDSS progression with time from an entry EDSS were derived from analysis of the BCMS database. Unlike the Markov model which considers each transition independent from what has gone before, multilevel modelling factors in the length already spent in each EDSS state. The model permits non-linear trajectories for individuals and allows for decreasing variability over time (reflecting the observed natural history of multiple sclerosis with slower progression through the higher grades).