Use of smartphone-enabled mood assessments and sensor data to predict alcohol craving in patients with alcohol-associated liver disease and alcohol use disorder

April 13, 2024

Alcohol-associated liver disease (ALD) is one of the main causes of chronic liver disease worldwide and accounts for nearly half of cirrhosis-related deaths in the United States. Treatment of alcohol use disorder (AUD) is the most effective strategy to prevent progression and complications of liver disease. However, current practice in treating patients with ALD and AUD relies on a one-size-fits-all approach, and numerous barriers, including program availability, limit patient access to these resources. Furthermore, individuals with ALD and AUD are heterogeneous and have individualized needs, leading to varied outcomes from current pharmacotherapies and behavioral interventions.

Recognizing these challenges, Mayo Clinic researchers looked to the evolving world of digital technologies such as smartphones for possible solutions. In an article published in Hepatology Communications in 2023, a team of Mayo Clinic researchers published results from a study that examined the use of smartphone-enabled sensor data and mood assessments to predict alcohol craving in a group of patients with ALD and AUD.

"Digital technologies have enabled the development of scalable, effective and individualized solutions for improving access to care," explains Tiffany Wu, M.D. Dr. Wu is currently a transplant hepatology fellow at Mayo Clinic in Rochester, Minnesota, and first author on the study publication.

"Digital tools like smartphones may facilitate collection of real-time data, the use of artificial intelligence to diagnose and predict outcomes, and the on-demand delivery of targeted treatment interventions through these devices," says Dr. Wu. "This emerging intervention design, called the just-in-time adaptive intervention, holds immense potential for promoting health and behavior change."

Study methods

Dr. Wu and colleagues sought to establish proof of concept for digital phenotyping as a means to measure alcohol craving in patients with ALD and AUD. They established a framework for using smartphone sensors to quantify markers of behavior in patients with AUD.

"Using that framework, we evaluated the relationship between smartphone sensor data and alcohol craving, which is an important target in prevention of relapse and management of alcohol-associated liver disease," explains Dr. Wu.

The researchers enrolled 24 adult individuals with ALD and AUD who agreed to download the AWARE app and complete daily ecological momentary assessments (EMAs) on alcohol craving, alcohol or substance use, and mood for up to 30 days. The AWARE application collected continuous data from 11 sensor streams, such as accelerometer, location and phone use. No personal identifiers were collected or stored within the app or server. From the different sensor streams, there were 127 sensor features characterized.

"Overall, we found that smartphone sensor data may be utilized as novel surrogate markers of alcohol craving and mood. These findings may classify new behavioral phenotypes of disease activity, while holding potential for predicting disease-related outcomes and opportunities for treatment."

— Tiffany Wu, M.D.

During an initial visit with study staff, study participants shared demographic information and medical history related to liver disease and alcohol use. They also completed a series of assessments for behavioral and psychological characteristics, and a Chronic Liver Disease Questionnaire (CLDQ) to provide a quality-of-life measurement specific to individuals with liver disease.

The daily EMA questions used the three-item Brief Alcohol Craving Scale measure of alcohol craving. After 30 days, participants completed a follow-up visit and a second CLDQ assessment to determine longitudinal change and a timeline follow-back for protocolized assessment of alcohol use during the study period.

The researchers used repeated measures correlation for longitudinal data to evaluate associations between sensors and EMAs and standard Pearson correlation to evaluate within-individual relationships between sensors and craving.

Among 24 participants who enrolled in the study, 12 participants (50%) had adequate AWARE sensor and EMA data. The median number of days during which AWARE data were transmitted was 34.5 per participant, and the most common sensor types available were accelerometer, screen and Wi-Fi data.


According to Dr. Wu and co-authors, analysis of the data yielded multiple interesting findings, including the following.

  • Alcohol craving was significantly correlated with mood obtained from EMAs among participants who experienced at least two days of change in their daily craving score.
  • Across all sensors, features associated with alcohol craving also were significantly correlated with all moods, except for boredom. This shows in aggregate that sensors may serve as a surrogate measure for mood as related to craving.

Individual-level analysis revealed significant relationships between a participant's craving and certain smartphone sensor data streams.

  • Location entropy: Some individuals experienced an increase in craving scores on the days when their time was primarily concentrated in a smaller number of locations.
  • Accelerometer magnitude: An increase in the average accelerometer magnitude, which is a measure of relative movement over time, was associated with an increase in craving score in some individuals.

"Overall, we found that smartphone sensor data may be utilized as novel surrogate markers of alcohol craving and mood," says Dr. Wu. "These findings may classify new behavioral phenotypes of disease activity, while holding potential for predicting disease-related outcomes and opportunities for treatment."

Dr. Wu and co-authors note that larger studies are needed to validate smartphone sensor relationships and predict long-term outcomes including alcohol relapse, liver decompensation and death in this patient population.

"We also need to advance digital technology to incorporate additional measures of human activity recognition that may allow for more fine-grained, precision measurement to understand distinct digital phenotypes and opportunities for individualized treatment," says Dr. Wu.

For more information

Wu, T, et al. Smartphone sensor data estimate alcohol craving in a cohort of patients with alcohol-associated liver disease and alcohol use disorder. Hepatology Communications. 2023;7:e0329.

Refer a patient to Mayo Clinic.