Table 2 Summary of classification results from meta-analysis of sample-level classifiers, leave-one-site-out and aggregate subject-level analyses

From: Using structural MRI to identify bipolar disorders – 13 site machine learning study in 3020 individuals from the ENIGMA Bipolar Disorders Working Group

Statistic

Meta-analysis

Leave-one-site-out

Aggregate subject-level

Accuracy (%)

-

58.67 (56.70–60.63)

65.23 (63.47–67.00)

ROC-AUC

-

60.92 (58.18–63.67)

71.49 (69.39–73.59)

Sensitivity (%)

42.60 (13.40–71.57)

51.99 (48.20–55.78)

66.02 (62.71–69.33)

Specificity (%)

59.14 (30.59–87.94)

64.85 (61.91–67.79)

64.90 (62.86–66.93)

PPV (%)

-

47.25 (37.67–56.84)

44.45 (42.04–46.86)

NPV (%)

-

67.67 (60.36–74.98)

83.73 (82.21–85.26)

  1. Note that meta-analytic results of the HSROC package include only sensitivity and specificity of the overall meta-analytic classification. Results for meta-analytic summary are the posterior predictive value of the performance metric, reported as mean (95% credible interval; the Bayesian analog of 95% confidence intervals). Results for the aggregate subject-level and leave-one-site-out analyses are reported as mean and 95% confidence interval
  2. NPV negative predictive value, PPV positive predictive value, ROC-AUC area under receiver operating characteristic curve
Back to article page