Table 4 Feature component accuracy of 12-month MDS-UPDRS total meta-prediction.

From: Genetically-informed prediction of short-term Parkinson’s disease progression

 

PPMI train-test split

PDBP independent test

 

ROC AUC

PR AUC

ROC AUC

PR AUC

Full meta-prediction

077 ± 0.04

0.76 ± 0.04

0.74 ± 0.05

0.73 ± 0.05

No genetics

0.66 ± 0.04

0.68 ± 0.04

0.66 ± 0.05

0.66 ± 0.05

No physician exam

0.67 ± 0.04

0.71 ± 0.04

0.69 ± 0.05

0.71 ± 0.05

No surveys

0.72 ± 0.04

0.72 ± 0.04

0.72 ± 0.05

0.72 ± 0.05

No imaging

0.73 ± 0.04

0.73 ± 0.04

-

-

  1. ROC AUC and PR AUC 12-month values for 12-month MDS-UPDRS Total progression prediction overall (full meta-prediction) and with feature class removal. PPMI is used for training in all cases. Train-test split is 75% training and 25% testing with stratified sampling. Note that removal of genetic features results in the greatest decline in predictive accuracy. ROC and PR curves for meta-prediction are provided in Fig. 4.