Abstract The Neural Craving Signature (NCS), a machine learning derived neuroimaging biomarker, differentiates individuals with from those without substance use disorders (SUDs), but has not been evaluated for predicting clinical outcomes. In a secondary analysis, we applied the NCS to fMRI cue-reactivity data from 39 participants in a published, negative RCT of repetitive transcranial magnetic stimulation (rTMS) for Alcohol Use Disorder (AUD). NCS scores predicted craving Penn Alcohol Craving Scale (PACS), both at the time of fMRI ( R 2 = 0.29, 95%, CI 0.27, 0.73, t(36) = 3.86, p = 0.0005), and during repeated study visits (β = 4.6, SE = 5.3, t(39.15) = 1.17, p < 0.0001). NCS also classified AUD severity (Addiction Severity Index, ASI, alcohol subscale—β = 0.14, SE = 0.04, p = 0.0016, R² = 0.24; Alcohol Use Disorder Identification Test, AUDIT, β = 5.32, SE = 1.46, p < 0.0025, R² = 0.22). Most importantly, the NCS predicted alcohol use, both measured by self-reported percent heavy drinking days (HDD%; β = 10.19, SE = 4.46, t(38.23) = 2.28, p = 0.028) and the biomarker phosphatidyl ethanol (PEth; β = 0.32, SE = 0.15, t(37.10) = 2.15, p = 0.038). Participants with below median NCS scores had a lower likelihood of relapse than those above median (Cox regression—HR = 0.35, 95% CI 0.16–0.80, p = 0.013). NCS identified relapse cases with an area under the curve of 0.79 (SE = 0.077, z = 3.8, p = 0.0001), achieving 66.7% sensitivity and 77.8% specificity at optimal NCS score. These findings provide initial support for the NCS as a predictor of clinical outcomes in AUD.
Löfberg et al. (Wed,) studied this question.