Table 4 Classification performance of XGBoost using clinical features, hematologic biomarkers and their combination, based on samples of disease duration ≤ 3 years.

From: Optimizing multi-domain hematologic biomarkers and clinical features for the differential diagnosis of unipolar depression and bipolar depression

Feature

Dataset

Accuracy

Sensitivity

Specificity

AUC

Clinical

10fold_CV*

0.930

0.785

0.732

0.795

Open test*

0.940

0.800

0.725

0.817

CBC

10fold_CV**

0.910

0.665

0.608

0.702

Open test**

0.920

0.678

0.658

0.719

BCM

10fold_CV**

0.890

0.608

0.499

0.625

Open test**

0.900

0.615

0.502

0.627

Clinical + CBC

10fold_CV**

0.930

0. 796

0.741

0.812

Open test**

0.940

0. 815

0.765

0.847

Clinical + CBC + BCM

10fold_CV

0.940

0. 800

0.754

0.829

Open test

0.950

0. 820

0.775

0.859

  1. Comparisons of AUCs between the combined features and other feature groups were performed using T-tests based on 1000 permutations. A statistically significant difference is noted: *P-value < 0.05; **P-value < 0.001. CBC Complete blood count, BCM Blood biochemical marker.
Back to article page