Table 4 The results of the logistic regression (LR) classifier in the leave-one-site-out analyses.

From: Prediction of lithium response using genomic data

Centre (LR+/LR−; %)

AUC

Accuracy

F1

NPV

PPV

Sensitivity

Specificity

Kappa

ALL (29/71)

0.57 (0.55, 0.6)

0.7 (0.7, 0.71)

0.08 (0.05, 0.1)

0.72 (0.71, 0.72)

0.36 (0.27, 0.46)

0.04 (0.03, 0.06)

0.97 (0.97, 0.98)

0.02 (− 0.01, 0.04)

Barcelona (27/73)

0.59 (0.58, 0.6)

0.72 (0.71, 0.72)

0.09 (0.07, 0.12)

0.72 (0.72, 0.72)

0.58 (0.53, 0.63)

0.05 (0.04, 0.07)

0.98 (0.98, 0.99)

0.05 (0.04, 0.06)a

Cagliari (28/72)

0.59 (0.57, 0.61)

0.72 (0.71, 0.73)

0.08 (0.03, 0.14)

0.72 (0.71, 0.73)

0.51 (0.22, 0.8)

0.04 (0.01, 0.08)

0.99 (0.98, 0.99)

0.05 (0.01, 0.09)

Geneva (23/77)

0.59 (0.57, 0.6)

0.71 (0.71, 0.72)

0.08 (0.07, 0.1)

0.72 (0.72, 0.72)

0.55 (0.47, 0.63)

0.05 (0.04, 0.05)

0.98 (0.98, 0.99)

0.04 (0.03, 0.05)

Halifax (45/55)

0.59 (0.56, 0.62)

0.75 (0.74, 0.75)

0.03 (0.01, 0.05)

0.75 (0.74, 0.75)

0.7 (0.31, 1)

0.01 (0, 0.03)

1 (1, 1)

0.02 (0, 0.04)

Japan (24/76)

0.59 (0.56, 0.62)

0.71 (0.71, 0.72)

0.09 (0.06, 0.12)

0.72 (0.71, 0.72)

0.54 (0.44, 0.63)

0.05 (0.03, 0.07)

0.98 (0.98, 0.99)

0.04 (0.02, 0.07)

Mayo (23/77)

0.58 (0.57, 0.6)

0.71 (0.71, 0.72)

0.1 (0.08, 0.11)

0.72 (0.72, 0.72)

0.51 (0.44, 0.59)

0.05 (0.04, 0.06)

0.98 (0.97, 0.99)

0.04 (0.03, 0.05)

Paris (18/82)

0.59 (0.57, 0.61)

0.7 (0.7, 0.71)

0.08 (0.06, 0.1)

0.71 (0.71, 0.71)

0.52 (0.43, 0.61)

0.04 (0.03, 0.05)

0.98 (0.98, 0.99)

0.03 (0.02, 0.05)

Poznan (48/52)

0.58 (0.57, 0.59)

0.73 (0.72, 0.73)

0.07 (0.04, 0.1)

0.73 (0.73, 0.73)

0.61 (0.51, 0.71)

0.04 (0.02, 0.05)

0.99 (0.99, 1)

0.04 (0.02, 0.06)

Romania (21/79)

0.6 (0.58, 0.62)

0.71 (0.7, 0.72)

0.1 (0.06, 0.14)

0.72 (0.71, 0.72)

0.59 (0.39, 0.8)

0.05 (0.03, 0.08)

0.99 (0.98, 0.99)

0.05 (0.02, 0.09)a

San Diego (11/89)

0.6 (0.59, 0.61)

0.7 (0.69, 0.7)

0.13 (0.1, 0.16)

0.7 (0.7, 0.71)

0.54 (0.46, 0.62)

0.07 (0.06, 0.09)

0.97 (0.96, 0.98)

0.06 (0.04, 0.08)a

Sweden (25/55)

0.54 (0.51, 0.57)

0.74 (0.74, 0.75)

0.05 (0.02, 0.08)

0.74 (0.74, 0.75)

0.55 (0.27, 0.82)

0.03 (0.01, 0.04)

1 (0.99, 1)

0.03 (0.01, 0.05)

Sydney (20/80)

0.59 (0.57, 0.61)

0.71 (0.71, 0.72)

0.08 (0.07, 0.09)

0.72 (0.71, 0.72)

0.52 (0.41, 0.63)

0.04 (0.04, 0.05)

0.98 (0.98, 0.99)

0.04 (0.02, 0.05)

Taiwan (14/86)

0.59 (0.58, 0.59)

0.71 (0.7, 0.71)

0.08 (0.06, 0.11)

0.71 (0.71, 0.72)

0.48 (0.39, 0.57)

0.05 (0.03, 0.06)

0.98 (0.97, 0.98)

0.04 (0.01, 0.06)

Würzburg (17/83)

0.6 (0.57, 0.62)

0.71 (0.7, 0.71)

0.11 (0.08, 0.14)

0.71 (0.71, 0.72)

0.55 (0.48, 0.63)

0.06 (0.04, 0.08)

0.98 (0.97, 0.99)

0.05 (0.04, 0.07)a

  1. Table columns represent different classification statistics and values represent the mean of each statistic over five folds along with an empirical 95% confidence interval. The center column shows the center that was left out for each row. We have included the results of the aggregate analysis in the top row for ease of comparison. In the column listing respective centres, we have included the percentage of lithium responders (LR+) and non-responders (LR−) in the format %LR+/%LR− to provide context for the classification results. Asterisks signify that the given metric was found to have a p value less than 0.01 compared to the simulated null classifier. Abbreviations: area under the receiver operating characteristic curve (AUC), positive predictive value (PPV), negative predictive value (NPV), F-1 score (F1).
  2. aKappa value was found to have a p value less than 0.01 in comparison to the null classifier.
Back to article page