Table 3 Verification of the absence of residual confounder effects following confounder regression

From: High-dimensional detection of imaging response to treatment in multiple sclerosis

Confounder

Before confounder regression

After confounder regression

Agea,b

0.645 [0.629–0.661]

0.483 [0.472–0.494]

Gender

0.723 [0.711–0.736]

0.488 [0.475–0.501]

Scanner manufacturerc

0.768 [0.756–0.780]

0.503 [0.487–0.519]

Field strengthd

0.803 [0.790–0.815]

0.503 [0.489–0.516]

Disease durationa,b

0.506 [0.493–0.519]

0.488 [0.477–0.499]

EDSSa,b

0.533 [0.511–0.555]

0.502 [0.483–0.521]

T1 slice thicknesse

0.522 [0.488–0.556]

0.442 [0.408–0.477]

T1 voxel resolutionb

0.789 [0.775–0.803]

0.497 [0.483–0.511]

FLAIR slice thicknesse

0.627 [0.591–0.662]

0.498 [0.474–0.523]

FLAIR voxel resolutionb

0.717 [0.699–0.736]

0.504 [0.491–0.517]

  1. None of the confounders could reliably be predicted using the imaging data by a support vector machine classifier following confounder regression. All figures are given as mean AUC [95% confidence interval]
  2. aAt the start of treatment with natalizumab
  3. bPredicting whether the confounder value is less than the mean for the sample population or otherwise
  4. cThe most common manufacturer in the sample population vs. the rest
  5. d1.5 T vs. 3.0 T
  6. ePredicting whether the slice thickness is less than 6 mm or otherwise
Back to article page