A total of 168 patients with late-life depression (108 from Washington University, 60 from Duke University) were included in the mixed model analyses. Because 10 patients did not complete the 12-week study, only 158 were included in the remission analysis. There were no differences on any variables between the overall sample and the 158 patients who completed the trial. The first analysis compared the depressed and comparison groups to generate regions of interest for testing effects on treatment outcome. As shown in , the two groups were similar in demographic characteristics, although comparison subjects were older and more educated on average. Regions of interest were analyzed using the covariates listed in .
Demographic Variables and Brain Measures in Patients With Late-Life Depression and Comparison Subjectsa
After covarying for demographic factors and adjusting for multiple comparisons, the mean differences between groups in volumes and cortical thickness (regression coefficient for group) were computed. The volumes of the hippocampus and amygdala and the thicknesses of the frontal pole, superior frontal gyrus, middle frontal gyrus, orbitofrontal gyrus, and anterior cingulate gyrus differed between the depressed and comparison groups. The parahippocampus and caudate did not differ between groups and were not included in our treatment outcome analyses. Total cortical gray matter volume did not differ significantly between groups.
Using the significant regions of interest generated from the group comparisons, we evaluated the effects of regions of interest on prospective treatment outcome using a mixed model. presents the results of a model that adjusted for the covariates age, education, age at onset, gender, scanner, and baseline depression severity but not for the interaction with time. Using this model, smaller amygdala and hippocampal volumes significantly predicted worse depression course as measured by MADRS scores. Using a model that adjusted for these covariates and that also adjusted for the interaction of the particular region of interest with time, only smaller hippocampal volume predicted slower rate of response to antidepressant treatment (hippocampal volume-by-time interaction, p=0.03). We then entered white matter hyperintensity severity, which did not significantly add to the prediction, and neuropsychological factor scores (episodic memory, cognitive processing speed, executive function, language, and working memory), one at a time. The Akaike information criterion (AIC), a goodness-of-fit test, was used to determine whether any neuropsychological factor scores improved the prediction of treatment response. While episodic memory, cognitive processing speed, executive function, and language, in combination with the hippocampal volume-by-time interaction, did predict the rate over time of recovery, cognitive processing speed in combination with the hippocampal volume-by-time interaction resulted in the smallest AIC and therefore was chosen in the model. We repeated this process adding in the other neuropsychological factor scores, but no other variable improved the model when added to the combination of hippocampal volume, hippocampal volume-by-time interaction, and processing speed. Thus, our working model for predicting rate of treatment response over time is presented in , with hippocampal volume and cognitive processing speed as the predictors.
Prediction of Depression Severity Over Time, Adjusting for Covariates, in Patients With Late-Life Depression and Comparison Subjectsa
Model of Treatment Response in Late-Life Depressiona
We previously showed (21
) that vascular risk factor scores predicted worse treatment outcome. As shown in , using Spearman’s correlation, the vascular risk factor scores (minus age) were highly correlated with all of the predictor variables except frontal pole thickness and anterior cingulate gyrus thickness. Since we wanted to ascertain whether the hippocampal volume predictor was significant independent of vascular risk factor score, we entered the latter post hoc into the analysis and found that hippocampal volume remained a significant predictor.
Correlation Between Brain Region of Interest and Framingham Vascular Risk Factor Score in Patients With Late-Life Depression and Comparison Subjects
Finally, in a confirmatory analysis, we examined which baseline volume and thickness variables differed between depressed patients who achieved remission (MADRS score ≤7) and those who did not. shows that patients who did not achieve remission had significantly smaller hippocampal volumes and thinner frontal poles than those who did achieve remission. There were no significant differences in demographic characteristics between those who achieved remission and those who did not. The box plots in for left and right hippocampal volumes and cognitive processing speed show that the median for both measures was highest in depressed patients who achieved remission, next highest in comparison subjects, and lowest in depressed patients who did not achieve remission.
Comparison of Demographic and Brain Variables Between Patients With Late-Life Depression Who Achieved Remission and Those Who Did Not
Box Plots of Hippocampal Volume and Cognitive Processing Speed in Patients With Late-Life Depression Who Did or Did Not Achieve Remission With Antidepressant Treatment and in Comparison Subjectsa