A cognitive-emotional biomarker for predicting remission with antidepressant medications: a report from the iSPOT-D trial

Abstract

Depression involves impairments in a range of cognitive and emotional capacities. It is unknown whether these functions can inform medication choice when considered as a composite predictive biomarker. We tested whether behavioral tests, grounded in the neurobiology of cognitive and emotional functions, predict outcome with common antidepressants. Medication-free outpatients with nonpsychotic major depressive disorder (N=1008; 665 completers) were assessed before treatment using 13 computerized tests of psychomotor, executive, memory-attention, processing speed, inhibitory, and emotional functions. Matched healthy controls (N=336) provided a normative reference sample for test performance. Depressed participants were then randomized to escitalopram, sertraline, or venlafaxine-extended release, and were assessed using the 16-item Quick Inventory of Depressive Symptomatology (QIDS-SR16) and the 17-item Hamilton Rating Scale for Depression. Given the heterogeneity of depression, analyses were furthermore stratified by pretreatment performance. We then used pattern classification with cross-validation to determine individual patient-level composite predictive biomarkers of antidepressant outcome based on test performance. A subgroup of depressed participants (approximately one-quarter of patients) were found to be impaired across most cognitive tests relative to the healthy norm, from which they could be discriminated with 91% accuracy. These patients with generally impaired cognitive task performance had poorer treatment outcomes. For this impaired subgroup, task performance furthermore predicted remission on the QIDS-SR16 at 72% accuracy specifically following treatment with escitalopram but not the other medications. Therefore, tests of cognitive and emotional functions can form a clinically meaningful composite biomarker that may help drive general treatment outcome prediction for optimal treatment selection in depression, particularly for escitalopram.

Figures

Figure 1

QIDS-SR16 remission rates based on predictions generated by the escitalopram QIDS-SR16 remission classifier, separated by medication arm. Plotted are remission rates when classifier outcome is not considered (gray bars, ie, current clinical practice), and response or remission rates when the QIDS-SR16 remission escitalopram classifier predicts that a participant will remit (black bars) or will not remit (white bars). *Significant difference between participants predicted to remit vs predicted not to remit on escitalopram (logistic regression odds ratio: 7.5; p=0.001). ESC, escitalopram; QIDS-SR16, 16-item Quick Inventory of Depressive Symptomatology; SER, sertraline; VEN, venlafaxine–extended release.

Figure 2

Relationships between individual tests and classifier performance. (a) Weights on each variable from the linear discriminant analysis. Shown are the means across all rounds of the classifier (not considering the majority vote ensemble step in the classifier). Together, these comprise the average classification ‘equation' that can be applied to future data. (b) Impact of individual behavioral tests on classification accuracy for the QIDS-SR16 escitalopram remission classifier. Plotted is the change in accuracy after removal of each test. ID, identification; LDA, linear discriminant analysis; QIDS-SR16, 16-item Quick Inventory of Depressive Symptomatology; RT, reaction time.