Single-subject anxiety treatment outcome prediction using functional neuroimaging

Abstract

The possibility of individualized treatment prediction has profound implications for the development of personalized interventions for patients with anxiety disorders. Here we utilize random forest classification and pre-treatment functional magnetic resonance imaging (fMRI) data from individuals with generalized anxiety disorder (GAD) and panic disorder (PD) to generate individual subject treatment outcome predictions. Before cognitive behavioral therapy (CBT), 48 adults (25 GAD and 23 PD) reduced (via cognitive reappraisal) or maintained their emotional responses to negative images during fMRI scanning. CBT responder status was predicted using activations from 70 anatomically defined regions. The final random forest model included 10 predictors contributing most to classification accuracy. A similar analysis was conducted using the clinical and demographic variables. Activations in the hippocampus during maintenance and anterior insula, superior temporal, supramarginal, and superior frontal gyri during reappraisal were among the best predictors, with greater activation in responders than non-responders. The final fMRI-based model yielded 79% accuracy, with good sensitivity (0.86), specificity (0.68), and positive and negative likelihood ratios (2.73, 0.20). Clinical and demographic variables yielded poorer accuracy (69%), sensitivity (0.79), specificity (0.53), and likelihood ratios (1.67, 0.39). This is the first use of random forest models to predict treatment outcome from pre-treatment neuroimaging data in psychiatry. Together, random forest models and fMRI can provide single-subject predictions with good test characteristics. Moreover, activation patterns are consistent with the notion that greater activation in cortico-limbic circuitry predicts better CBT response in GAD and PD.

Figure 1

Random forest procedure. Step 1a is to build a decision tree based on a bootstrapped sample of participants (filled circles represent responders and open circles represent non-responders) and a random sample of predictor variables (eg, average activation in anatomically defined brain regions, arbitrary data shown). The random forest algorithm determines the optimal split point for each variable in order to correctly classify this subset of participants. Step 1b is to repeat this process hundreds or thousands of times to generate a forest of trees. In step 2, each tree classifies the participants that were not used in its original construction; each tree then ‘votes' for the classification of these participants, and these votes are aggregated to provide the predicted status of each participant and thereby determine accuracy. Figures marked with an asterisk (*) indicate inaccurately classified participants in this example. Step 3 is the identification of the most important variables for prediction. Brain regions are ranked in terms of their variable importance scores: only those with greater importance than the absolute value of the most negative importance rating are selected for the final model (arbitrary data shown). The variables selected for inclusion are then used as the sole input variables for another iteration of steps 1 and 2, generating the final model.

Figure 2

Model comparison of positive and negative likelihood ratios and posterior probabilities for (a) the clinical and demographic predictive model, (b) the fMRI predictive model, and (c) the combined model. Brackets indicate 95% confidence intervals. Upper lines indicate positive test result (ie, predicted responder) and lower lines indicate negative test result (ie, predicted non-responder).

Figure 3

Average activation in responders and non-responder in regions selected for the final fMRI model. Error bars=SEM. Y axis is the perentage of signal change. Hippocampus and uncus activations were from the maintain condition, all other regions were from the reappraise condition. HIPP, hippocampus; INS, anterior insula; L, left; PRCEN, precentral gyrus; R, right; SFG, superior frontal gyrus; SN, substantia nigra; SPMG, supramarginal gyrus; STG, superior temporal gyrus; TTG, transverse temporal gyrus.