Analysis of selected QEEG predictors of response to transcranial magnetic stimulation in major depression

Abstract

We used a logistic diskriminant analysis (LDA) to determine a QEEG predictive model of response to rTMS therapy in patients with pharmacoresistant depression (N=25; 16 nonresponders). Out of the set of variables generated by the Neuroguide software (N=836), we selected 12 variables (predictors) using a t-test.

This number has been then reduced within LDA by means of a stepwise method on 5 (in the brackets are P values for significancy of the hypothesis of predictors' null contribution to the model)- alpha assymetry (ASAL) C3-Cz (0,4177), ASAL C4-Cz (0,3170), theta asymmetry (TAS) C4 -T4 (0,3475), TAS T3 - T4 (0,3332) and theta peak (TP) C4 (0,2061). The LDA model containing these five variables demonstrated correct classification on 88%.

Furthermore, it shows a good overall model fit based on -2 Log Likelihood (Chi-squred = 24,03; df = 5; P = 0ased on ,0002). Hosmer & Lemeshow's test also showed a significant model fit, however, requirements of this test on filling particular decils with the corresponding number of cases were not met.

The model demonstrates a solid prediction rate, although its failure in case of externalization might be expected due to the small sample size. Progressive and eliminating employment of bootstrapping (10000 opakování) on the original 12 variables confirmed though, that the variables of asymmetry alpha 1 (ASAL 1) C4-Cz and TAS C4 -T4 demonstrate robust resistancy (p = 0,001 resp. 0,02) against bias while keeping a satisfactory level of classification.