FIGURE 3. Estimated Survival Risk Functions and Receiver Operating Characteristic Curves for Gray Matter Volumes in Specific Significant Brain Regions in Alcohol-Dependent Patients (N=44)^a.

^aPanels A and B present estimated survival risk functions (with mean age, IQ, and baseline total amount of alcohol consumed held constant) for mean gray matter volumes as well as for volumes one and two standard deviations above and below the mean for the medial frontal cluster (panel A; cluster χ²=6.7, p<0.009; hazard ratio=0.52, 95% CI=0.31–0.85) and the parietal-occipital cluster (panel B; χ²=9.28, p<0.002; hazard ratio=0.52, 95% CI=0.34-0.79). Although the survival function was a 90-day analysis, the graphs are cut off at day 60 because all alcohol-dependent patients with gray matter volumes two standard deviations below the mean for each of the two regions relapsed by day 60. For patients with volumes two standard deviations above the mean in the medial frontal cluster, the estimated survival function at day 60 spans a 0.68 (68%) proportion of surviving relapse, and for those with volumes two standard deviations above the mean in the parietal-occipital cluster, a 0.66 (66%) proportion, whereas for patients with volumes two standard deviations below the mean, the estimated survival function at day 60 for both regions spans only a 0.02% chance of surviving relapse. Panel C shows receiver operating characteristic (ROC) curves for the medial frontal (area under the curve=0.83; 95% CI=0.65–0.96), lateral frontal (area under the curve=0.75, 95% CI=0.55–0.89), and parietal-occipital clusters (area under the curve=0.70, 95% CI=0.49–0.86). The optimal gray matter volume cutoff values are circled in the ROC curves for each of the three regions that best differentiated between those who relapsed and those who did not (medial frontal cluster: χ²=7.04, df=1, 44, p<0.008; odds ratio=0.25, 95% CI=0.09–0.70; correctly classified 80% of relapsers at a 93.3% sensitivity and 50% specificity; optimal cutoff value=8.41; lateral frontal cluster: χ²=4.23, df=1, 44, p<0.04; odds ratio=0.56, 95% CI=0.32–0.97; correctly classified 71% of relapsers at a 83.3% sensitivity and 42.9% specificity; optimal cutoff value=10.6; parietal-occipital cluster: χ²=3.86, df=1, 44, p<0.05; odds ratio=0.47, 95% CI=0.22–0.99; correctly classified 66% of relapsers at a 90% sensitivity and 14.3% specificity; optimal cutoff value=10.4).