Multiplatform Biomarker Discovery for Bladder Cancer Recurrence Diagnosis
Table 4
Multivariate regression models.
Model
Strategy
Description
Included parameters
AUC
AUC (LOOCV)
Model 1
Manual selection
The model comprises clinical parameters exhibiting on the individual level some association with the outcome parameter and the clinically relevant age at time of sample
no.past.recurrences, BCG.therapy, no.past. TURBTs, and age.sample
0.78
0.65
Model 2
Automatic selection
The model comprises clinical parameters with a selection probability greater than 50%
no.past.recurrences, BCG.therapy, and stage.diagnosis
0.80
0.72
Model 3
Manual selection
The model comprises biomarker candidates exhibiting on the individual level some association with the outcome parameter
, , , and
0.72
0.51
Model 4
Automatic selection
The model comprises biomarker candidates with a selection probability greater than 50%
, , , , , and
0.78
0.61
Model 5
Union of the parameters in Model 1 and Model 3
0.82
0.64
Model 6
Union of the parameters in Model 2 and Model 4
0.91
0.70
(a) Included parameters: stage.diagnosis: stage of the tumor at time of diagnosis. The other clinical parameters are defined in the Specimen and Data Collection. (b) Markers ending with chip were measured with the BCa chip and markers ending with AP were measured with the automated platform for 96-well plate ELISA analysis. (c) LOOCV: leave-one-out cross-validation. (d) Biomarker candidates chosen during manual selection for Model 3 are a subset of Model 4.
