Table 3 Classification performance with real datasets
TwoStep | Latent Gold | SNOB | |
|---|---|---|---|
Number of subgroups detected | |||
MRI1 dataset | 2 | 7 | 10 |
MRI2 dataset | 3 | 11 | 15 |
MRI3 dataset | 2 | 6 | 7 |
SMS dataset | 2 | 10 | 37 |
Clinical dataset | Not available | 8 | 9 |
Certainty (mean classification probability of disc levels or patients) | |||
MRI1 dataset | Not available | 91.2% (SD11.9%) | 91.5% (11.6%) |
MRI2 dataset | Not available | 98.9% (SD3.9%) | 97.1% (SD6.6%) |
MRI3 dataset | Not available | 85.7% (SD19.5%) | 91.0% (SD12.7%) |
SMS dataset | Not available | 96.5% (SD8.8%) | 98.2% (SD4.7%) |
Clinical dataset | Not available | 91.4% (SD12.9%) | 89.9% (SD13.5%) |
Reproducibility (10 iterations of each dataset, with identical results across all datasets) | |||
Number of subgroups | 100% agreement | With fixed seed point = 100% agreement | 100% agreement |
Classification stability (reproducibility of individual disc-levels or people being classified into each subgroup) | 100% agreement | With fixed seed point = 100% agreement | 100% agreement |
Classification certainty (reproducibility of the classification probability of disc levels or patients) | Not available | With fixed seed point = 100% agreement | 100% agreement |