Table 5 Characteristics of the studies used to inform the developed prognostic model
Study | Design | Sample size (EPV) | Outcome | Missing data | Discrimination | Presentation of the model | Model |
|---|---|---|---|---|---|---|---|
Calibration | |||||||
Validation | |||||||
PHeld et al. [9] | Placebo arms of clinical trials Multicentre | n = 821 (NA) | Relapse rate | Complete case analysis | Discrimination: c-statistic NA for continuous outcome Calibration: absent. Internal validation: split-sample | Table | Prognostic model |
Kalincik et al. [35] | Cohort study | n = 8513 | Treatment response for relapse frequency | Mentioned: Values of the principal components can be estimated even for patients with incomplete data | Accuracy and internal validity reported (moderate for relapse rate at 2 years) Internal validation: in a separate, non-overlapping MSBase cohort. External validation: 2945 patients from the Swedish Multiple Sclerosis Registry | Table of principal components | Prediction model |
Liquori et al. [10] | Cohort (R) Single centre | n = 127 (NA) | Relapse rate | Complete case analysis | Overall performance reported R2 Calibration: absent Validation: absent | NA | Prognostic model |
Pellegrini et al. [34] | RCT | n = 2099 | Treatment response to annualized relapse rate | Complete case analysis | Performance measure of area under the AD(c) curve shown in the graph Calibration absent Internal validation: Splitting of the training dataset into two subsets (50%/50%) External validation: Independent RCT | Table | Prediction model |
Signori et al. [36] | All published randomized clinical trials in RRMS reporting a subgroup analysis | Six trials 6693 RRMS patients | Treatment response to annualized relapse rate | Not relevant | Not relevant | Not relevant | Subgroups responsive to treatments |
Sormani et al. [7] | Placebo arm of RCT Multicentre | n = 539 (insufficient data reported) | Number of relapses at 9 months | Complete case analysis | Discrimination: absent Calibration: absent Internal validation: absent | Mathematical formula | Prognostic model |
Stühler et al. [33] | Real-world data | n = 25000 | Treatment response to number of relapses | Complete case analysis | Discrimination: c-statistic (0.65) Calibration: Calibration plot Internal validation: 1)10-fold cross-validation, 2)leave-one-site-out cross-validation, and 3)excluding test set | Table | Prediction model |