A new study in Neuropsychopharmacology shows that a machine learning brain tool can support alcohol relapse prediction in patients with alcohol use disorder (AUD) before they take a single drink. Researchers from Linköping University, Sweden, and the University of California, Berkeley, tested a neuroimaging biomarker called the Neurobiological Craving Signature (NCS). This is the first time scientists have used the NCS for predicting alcohol relapse in a real clinical setting.

What Is the NCS and How Does Alcohol Relapse Prediction Work?

The NCS uses functional magnetic resonance imaging (fMRI) scans alongside machine learning. Instead of examining one brain region, it reads patterns of activity across the whole brain. Researchers show patients images of alcoholic drinks inside the scanner. The NCS then produces a score that reflects how intensely the brain signals craving.

Earlier methods focused on isolated areas like the ventral striatum or orbitofrontal cortex. The NCS captures a wider neural picture, which means it picks up signals that narrow, single-region tools routinely miss.

The scale of the problem makes this kind of tool urgent. Around 10% of people in Western countries meet the clinical criteria for AUD. Fewer than 10% of those people ever receive evidence-based treatment. Among those who do, relapse is still the most common outcome.

Inside the Study: How Researchers Tested Alcohol Relapse Prediction

Researchers applied the NCS to fMRI data from 39 participants in a randomised controlled trial of repetitive transcranial magnetic stimulation (rTMS) for AUD. The original treatment, targeting the insula on both sides of the brain, produced no significant effect on craving or drinking. That neutral result was actually helpful. The team could pool all participants together while still controlling for which group they were in.

Inside the scanner, participants completed a beverage-matching task. They viewed images of alcoholic and non-alcoholic drinks and pressed a button to match a target image. This kept their attention focused actively on the cues rather than letting them drift. Researchers then tracked NCS scores against clinical outcomes across a 15-week period covering treatment and follow-up.

Cravings, Diagnoses and Outcomes: What the Data Showed

The results across multiple measures pointed in the same direction.

Craving. NCS scores predicted self-reported craving on the day of the scan (R² = 0.29, p = 0.0005). They also predicted craving at every follow-up visit. Each one-unit rise in NCS score pushed Penn Alcohol Craving Scale ratings up by roughly 4.7 points.

Severity. Higher NCS scores linked strongly to greater disorder severity on the Addiction Severity Index alcohol subscale (R² = 0.24, p = 0.0016) and the Alcohol Use Disorder Identification Test (R² = 0.22, p = 0.0025). The brain signature does not just detect alcohol use. It reflects how serious that use has become.

Drinking behaviour. The NCS predicted both self-reported heavy drinking days and blood levels of phosphatidylethanol (PEth), an objective biomarker for recent alcohol consumption. PEth levels rose by around 0.32 units for each one-unit increase in NCS score. Notably, self-reported craving scores alone did not significantly predict relapse. The NCS did. That gap matters.

Alcohol Relapse Prediction: The Survival Analysis

Researchers split participants into high and low NCS groups at the median score. The high NCS group relapsed to heavy drinking significantly faster. Cox regression confirmed that participants with below-median NCS scores carried a 65% lower hazard of relapse (HR = 0.35, 95% CI [0.16 to 0.80], p = 0.013). On average, the low NCS group stayed relapse-free for nearly four extra weeks compared to the high group.

As a tool for predicting alcohol relapse, the NCS reached an area under the curve (AUC) of 0.79, with sensitivity of 66.7% and specificity of 77.8%. A Bayesian analysis estimated the median AUC at 0.91, which the authors describe as excellent discrimination.

Why Self-Reports Cannot Replace Alcohol Relapse Prediction Tools

Asking patients how much they crave alcohol is not enough to forecast whether they will relapse. The NCS, however, can do this. Questionnaires carry known limitations. Patients underreport consumption. Many lack full insight into how severe their condition is. Social pressure can shape what they say during clinical appointments.

A brain-based measure removes these barriers. It records what the brain does when it encounters alcohol-related cues. It does not rely on memory or self-awareness. That objectivity is exactly what makes alcohol relapse prediction through neuroimaging so valuable.

What Predicting Alcohol Relapse Could Mean for Treatment

Researchers describe the NCS as a potential “theragnostic” biomarker. It could work both as a predictor of risk and as a live monitor of whether treatment is working. That combination could support genuinely personalised care for AUD patients, identifying those who need more intensive help before they have already lapsed.

Limitations Worth Noting

The sample of 39 participants is small. Female participants were underrepresented. The study also used data from a trial not originally designed for this purpose. The authors call for replication in larger, prospectively designed research. Still, the NCS produced consistent findings across craving scores, drinking days, and a blood biomarker. That convergence strengthens the case.

It also generalised to a cue-reactivity method quite different from the one used to build it. That suggests the brain signature captures something real and transferable, not a quirk of one particular experimental setup.

A Practical Step Towards Earlier Alcohol Relapse Prediction

Alcohol use disorder is one of the most undertreated conditions in medicine. Tools that identify vulnerability early, without depending solely on what a patient reports in a clinical room, carry real practical value. Predicting alcohol relapse through brain imaging is no longer an abstract idea. This study shows it is becoming a workable clinical approach.

Future research needs to test whether the NCS works at the start of treatment, whether it can anticipate responses to specific therapies, and whether it performs consistently across broader, more diverse populations. The science is early. The direction, though, is clear.

Source: dbrecoveryresources