Inter and intra-hemispheric structural imaging markers predict depression relapse after electroconvulsive therapy: a multisite study - PubMed (original) (raw)

Inter and intra-hemispheric structural imaging markers predict depression relapse after electroconvulsive therapy: a multisite study

Benjamin S C Wade et al. Transl Psychiatry. 2017.

Abstract

Relapse of depression following treatment is high. Biomarkers predictive of an individual's relapse risk could provide earlier opportunities for prevention. Since electroconvulsive therapy (ECT) elicits robust and rapidly acting antidepressant effects, but has a >50% relapse rate, ECT presents a valuable model for determining predictors of relapse-risk. Although previous studies have associated ECT-induced changes in brain morphometry with clinical response, longer-term outcomes have not been addressed. Using structural imaging data from 42 ECT-responsive patients obtained prior to and directly following an ECT treatment index series at two independent sites (UCLA: n = 17, age = 45.41±12.34 years; UNM: n = 25; age = 65.00±8.44), here we test relapse prediction within 6-months post-ECT. Random forests were used to predict subsequent relapse using singular and ratios of intra and inter-hemispheric structural imaging measures and clinical variables from pre-, post-, and pre-to-post ECT. Relapse risk was determined as a function of feature variation. Relapse was well-predicted both within site and when cohorts were pooled where top-performing models yielded balanced accuracies of 71-78%. Top predictors included cingulate isthmus asymmetry, pallidal asymmetry, the ratio of the paracentral to precentral cortical thickness and the ratio of lateral occipital to pericalcarine cortical thickness. Pooling cohorts and predicting relapse from post-treatment measures provided the best classification performances. However, classifiers trained on each age-disparate cohort were less informative for prediction in the held-out cohort. Post-treatment structural neuroimaging measures and the ratios of connected regions commonly implicated in depression pathophysiology are informative of relapse risk. Structural imaging measures may have utility for devising more personalized preventative medicine approaches.

PubMed Disclaimer

Conflict of interest statement

The authors declare that they have no competing interests.

Figures

Fig. 1. Flowchart of classification process

At each leave-one-out cross-validation fold, training subjects are randomly assigned to each of ten folds and feature selection based on collinearity filtering and recursive feature elimination (RFE) is performed on nine of the folds. Features selected by this process above a given threshold are used to train a model on the whole training set and the parameters from this model are used to predict the originally held-out observation

Fig. 2. Regions predictive of relapse and their relationship to the predicted probability of relapse

Top row illustrates anatomical locations of cortical and subcortical regions most important to relapse prediction. The middle row indicates the posterior probability of individual relapse over an observed range (minimum to maximum in 20 even increments) of region ratios locally averaged across 10 bootstrapped resamples of the data set and refitted to the derived classifier. A non-parametric LOESS model was fit to the predicted responses. Points about each line indicate predicted probabilities from each resample while rugs of each plot indicate the density of observed values in the whole sample. The bottom row illustrates corresponding distributions of random forest decision points (black) for these regions across underlying 1000 classification trees in the determination of relapse status. These are compared to distributions of these regions for relapsing (red) and non-relapsing (blue) patients

Fig. 3. Distributions of balanced accuracies across all classifier parameterizations

a Performances of classifiers trained and tested within a cohort. Horizontal red lines indicate the site-specific baseline detection rate. b Distributions of balanced accuracies achieved in cross-site predictions. All models exceeding their respective baseline detection rates within site for UCLA at pre-treatment (b, left) and UNM post-treatment (b, right) were used to predict relapse in the independent site. These performances are compared to the independent site’s respective baseline detection rate shown in the horizontal red line

References

1. Kessler RC, Bromet EJ. The epidemiology of depression across cultures. Annu. Rev. Public Health. 2013;34:119–138. doi: 10.1146/annurev-publhealth-031912-114409. -DOI -PMC -PubMed
1. Trivedi MH, et al. Evaluation of outcomes with citalopram for depression using measurement-based care in STAR*D: implications for clinical practice. Am. J. Psychiatry. 2006;163:28–40. doi: 10.1176/appi.ajp.163.1.28. -DOI -PubMed
1. McGrath PJ, et al. Tranylcypromine versus venlafaxine plus mirtazapine following three failed antidepressant medication trials for depression: a STAR*D report. Am. J. Psychiatry. 2006;163:1531–1541. doi: 10.1176/ajp.2006.163.9.1531. -DOI -PubMed
1. Eaton WW, et al. Population-based study of first onset and chronicity in major depressive disorder. Arch. Gen. Psychiatry. 2008;65:513–520. doi: 10.1001/archpsyc.65.5.513. -DOI -PMC -PubMed
1. Fava M, et al. A comparison of mirtazapine and nortriptyline following two consecutive failed medication treatments for depressed outpatients: a STAR*D report. Am. J. Psychiatry. 2006;163:1161–1172. doi: 10.1176/ajp.2006.163.7.1161. -DOI -PubMed

Inter and intra-hemispheric structural imaging markers predict depression relapse after electroconvulsive therapy: a multisite study - PubMed (original) (raw)