Skip to main content

Table 4 Information gain of multivariable logistic regression models compared to univariate logistic regression including only amyloid modalities

From: PET and CSF amyloid-β status are differently predicted by patient features: information from discordant cases

   AIC   AIC  
Predictor PET ~
CSF
PET ~
CSF + predictor
AIC difference CSF ~
PET
CSF ~
PET + predictor
AIC difference
Total APOE E4 positivity 580 533 47 573 531 42
CSF tau 580 508 71 573 563 10
CSF p-tau 580 481 99 573 555 17
SCD APOE E4 positivity 142 138 4 167 152 15
CSF tau 142 132 10 167 162 5
CSF p-tau 142 132 10 167 163 4
MCI APOE E4 positivity 104 86 18 104 100 4
CSF tau 104 74 30 104 105 − 1
CSF p-tau 104 83 21 104 106 − 2
Dementia APOE E4 positivity 317 295 22 286 267 19
CSF tau 317 294 23 286 285 2
CSF p-tau 317 263 54 286 276 10
  1. This table illustrates the change in Akaike Information Criterion (AIC) from the bivariate models including only amyloid modalities (PET ~ CSF and CSF ~ PET) to multivariable models including also an additional predictor. AIC measures model fit and penalizes adding additional predictors. A decrease in AIC between models shows some (0–2), considerable (4–7), or strong (> 10) evidence for gain in model fit for the second model