Table 4 Cross-validated performance for predicting comorbidities, non-motor outcomes, and progression rate subgroups

From: Integrating digital gait data with metabolomics and clinical data to predict outcomes in Parkinson’s disease

Outcome/median AUC (mad)

Gait data

Clinical data

Metabolomics data

Cognitive impairment (MoCA)

0.664 (0.14)

0.759 (0.14)

a0.788 (0.08)

Dopamine dysregulation syndrome

0.688 (0.19)

a0.714 (0.32)

0.676 (0.34)

Depression (BDI)

a0.783 (0.14)

0.764 (0.17)

0.647 (0.11)

Hallucinations

0.673 (0.17)

0.750 (0.31)

a0.785 (0.08)

Dyskinesias

0.637 (0.20)

a0.917 (0.12)

0.667 (0.17)

Apathy (Starkstein scale)

0.598 (0.19)

a0.616 (0.15)

0.524 (0.13)

Quality of life (PDQ-39)

0.647 (0.17)

0.600 (0.08)

a0.676 (0.21)

Progression rate

0.633 (0.13)

a0.728 (0.13)

0.717 (0.22)

  1. Cross-validated performance for predicting comorbidities, non-motor outcomes, and progression rate subgroups (left column) in Parkinson’s disease patients using either the time series features derived from the gait data (column 2), clinical variables (using only the non-motor features from the “Clinical data” section in the Methods part; column 3) or blood metabolomics data (column 4) as input. Quantitative outcome scores were binarized using a median threshold to obtain comparable AUC scores across different types of outcomes (for the progression rate outcome only, the fast and slow progressor subgroups were defined as the upper and lower quartiles, respectively, of the average annual change in the MDS-UPDRS III motor score, consistent with previous studies27). The extreme gradient boosting (XGB) algorithm was used for prediction and 10-fold cross-validation was applied. The presented scores represent the median area under the Receiver Operating Characteristic Curve (AUC) across the cross-validation cycles and their median deviation (mad).
  2. aThe highest median AUC achieved for each outcome.
Back to article page