Table 2. Prediction Accuracy of ΔHDRS and ECT Remitters for Three Sites.
| UNM | UCLA>50 years | LIJ>50 years | ||||
|---|---|---|---|---|---|---|
| No. of subjects | 38 | 10 | 7 | |||
| Correlation r/p | 0.91/2e−15 | 0.75/0.01 | 0.74/0.05 | |||
| RMSE | 6.0 | 5.8 | 7.0 | |||
| Predicted remitter | Predicted non-remitter | Predicted remitter | Predicted non-remitter | Predicted remitter | Predicted non-remitter | |
| True remitter | 24 | 3 | 2 | 0 | 2 | 1 |
| True non-remitter | 1 | 10 | 1 | 7 | 0 | 4 |
| Sensitivitya | 88.9% | 100% | 66.7% | |||
| Specificitya | 90.9% | 87.5% | 100% | |||
| PPVa | 96.0% | 66.7% | 100% | |||
| NPVa | 76.9% | 100% | 80.0% | |||
| Accuracya | 89.5% | 90.0% | 85.7% | |||
Abbreviations: ECT, electroconvulsive therapy; HDRS, Hamilton Depression Rating Scale; LIJ, Long Island Jewish Health System; MDD major depressive disorder; NPV, negative predictive value; PPV, positive predictive value; RMSE, root mean squared prediction error; UCLA, University of California at Los Angeles; UNM, University of New Mexico.
Based on the predicted ΔHDRS and the pre-ECT HDRS, the post-ECT HDRS could be estimated and each MDD patient can further be classified as remitter or non-remitter according to (Heijnen et al, 2010). Here, sensitivity is defined as number of correctly predicted remitters divided by number of true remitters; specificity is defined as number of correctly predicted non-remitters divided by number of true non-remitters; PPV is defined as number of correctly predicted remitters divided by number of predicted remitters; NPV is defined as number of correctly predicted non-remitters divided by number of predicted remitters; accuracy is defined as number of correctly predicted remitters and non-remitters divided by number of all MDD.