Table 2.

‘Goodness of fit’ statistics for the 10-state† disability (EDSS) Markov models

Description of each 10-state† disability model	−2 Log likelihood‡	Prediction errors (years 1–10)§
	×1000	Cells	EDSS	Utility
No covariates	17.152	2.20	0.24	0.022
One covariate models
Age at onset, binary	17.458	1.39	0.09	0.009
Age at onset, continuous	17.599	1.58	0.13	0.007
MSSS at baseline, binary	17.460	1.41	0.10	0.008
MSSS at baseline, continuous	17.457
Disease duration, binary	17.462	1.33	0.10	0.009
Disease duration, continuous	17.557
Sex	17.470	1.32	0.10	0.008
Two covariates models
Sex and age at onset, binary	17.603	1.51	0.14	0.007
Sex and age at onset, continuous	17.618
Age at onset and MSSS, binary	17.609	1.53	0.14	0.007
Age at onset and MSSS, continuous	17.618
Age at onset and disease duration, binary	17.603	1.52	0.14	0.007
Age at onset and disease duration, continuous	17.618

Primary goodness of fit statistic is −2 log likelihood; prediction errors have only been calculated for the binary versions of the individual models except for the ‘final’ model with age at onset as covariate where prediction errors have been calculated for both versions.

†The 10 disability states refer to EDSS 0–9, that is, EDSS 0 is ‘state 1’, EDSS 1 is ‘state 2’, etc.

‡Log likelihood statistic as calculated by ‘msm’ module, see Jackson¹⁹ for details; lower values implying a better model (to be compared within each class of models, eg, one-covariate and two-covariate models).

§Prediction errors, averaged over years 1–10, for (a) the EDSS distribution in individual cells, (b) average EDSS, (c) average utility (see definitions in the online supplementary appendix 3, comparing the values predicted by the model with the ‘observed’ values using the method of midpoint interpolation (see online supplementary appendix 2).

EDSS, Expanded Disability Status Score; MSSS, Multiple Sclerosis Status Score.