Neuroimaging and Clinical Features Models May Predict Response to Antidepressants in Adults With MDD
Neuroimaging and clinical features models predicting response to antidepressant treatment in adults with major depressive disorder (MDD) showed significant generalizability to other randomized clinical trials (RCTs), according to study findings published in JAMA Network Open.
“Although several studies have identified promising markers of antidepressant response, it is unclear whether findings generalize across trials and, thus, to future patients,” wrote lead author Peter Zhukovsky, PhD, Center for Depression, Anxiety and Stress Research, Department of Psychiatry, McLean Hospital, Harvard Medical School, Belmont, Massachusetts, and study coauthors. “The findings of this prognostic study identified promising biological and clinical markers of antidepressant treatment response.”
READ>>Phase II Trial of AL001 for MDD Set to Begin Later this Year
The study utilized structural and functional resting-state magnetic resonance imaging and clinical and demographic data from 2 distinct trials investigating the efficacy of antidepressants—the Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care (EMBARC) RCT and the Canadian Biomarker Integration Network in Depression (CANBIND-1) RCT. The trials administered sertraline and escitalopram, respectively. The researchers then used elastic net logistic regressions with regularization to predict treatment outcomes in the data sets from the RCTs, which included a total of 363 adult patients with MDD (225 from EMBARC and 138 from CANBIND-1; mean [SD] age, 36.6 [13.1] years; 235 women [64.7%]). The primary treatment outcome was treatment response, defined as a 50% or greater reduction in depression severity.
The researchers trained and tested the prediction performance of 5 sets of models on baseline depression severity data across the 2 trials; a clinical model including age, sex, employment, baseline HRSD, SHAPS, and BMI; a clinical plus global functional connectivity (FC) model; a clinical plus dorsal anterior cingulate cortex (dACC) FC model; a clinical plus rostral aterior cingulate cortex (rACC) FC model; and a clinical plus cortical thickness model. Researchers then tested 5 analogous models that included the week 2 data instead of the baseline depression severity scores. The performance response was assessed using balanced classification accuracy and area under the curve (AUC).
The best-performing models were the clinical data model (trained on CANBIND-1 and tested on EMBARC, AUC = 0.62 for stage 1 and AUC = 0.67 for stage 2) and the model using dACC-to-cortex connectivity alongside clinical data (trained on EMBARC stage 1 and tested on CANBIND-1, AUC = 0.66), which both showed moderate cross-trial generalizability.
The researchers also found that using neuroimaging features in addition to clinical data improved prediction performance, and that the models had greater generalizability for week 2 depression severity scores than baseline scores.
While the study was limited by its small sample size, lack of preregistration of its analytic approach, and the challenge of data harmonization, the authors emphasized that the findings may improve the process of prescribing effective treatment for adults with MDD.
“Leveraging data to identify robust biomarkers that generalize across patient populations in different geographic locations will allow us to test such biomarkers in prospective randomized clinical trials and hopefully help connect patients with treatments that work best for them,” the researchers concluded.
