Towards a Wearable Alcohol Biosensor: Examining the Accuracy of BAC Estimates from New-Generation Transdermal Technology using Large-Scale Human Testing and Machine Learning Algorithms

A wearable alcohol biosensor could represent a tremendous advance towards helping people make informed decisions about their drinking and, ultimately, towards curbing alcohol-related morbidity and mortality. Transdermal sensors, which measure alcohol consumption by assessing the alcohol content of insensible perspiration, offer a uniquely non-invasive, passive, and low-cost method for the continuous assessment of drinking likely to be attractive to a range of populations. But the relationship between transdermal alcohol concentration (TAC) and blood alcohol concentration (BAC) is highly complex, varying across individuals and contexts and involving some degree of lag time. Prior research, which has featured extremely small participant samples and examined old-generation transdermal devices, has been poorly suited to modeling this complexity. Thus, scientists are left with little sense for how to translate data produced by transdermal sensors into estimates of BAC. Importantly, the past decade has seen remarkable technological and analytic developments, offering the potential to tackle the challenge of TAC-BAC translation. In particular, in recent years, machine learning approaches have been developed that are particularly well suited to modeling highly complex and time-lagged relationships within larger datasets. Also during this time period, a new generation of transdermal device has come under development, featuring sleek/compact designs, smartphone integration, and capabilities for sampling TAC at approximately 90 times the rate of older-generation devices. These sensors thus provide a rich source of data for machine learning models and also, for the first time, the potential to produce transdermal BAC estimates in real time. The proposed research leverages machine learning, novel transdermal technology, and large-scale multimodal human testing to translate transdermal sensor data into estimates of BAC. Transdermal sensors will be examined in the context of multimodal research featuring a large and diverse participant sample (N=240) examined both inside and outside the laboratory. The ambulatory arm of the proposed project is aimed at capturing the TAC-BAC relationship across individuals in varying real-world drinking contexts, examining regular drinkers wearing new-generation transdermal sensors in everyday settings while providing prompted breathalyzer readings. This ambulatory research will be complemented by a laboratory study arm, aimed at examining the TAC-BAC relationship among individuals drinking in a controlled setting while alcohol dose and rate of consumption are systematically manipulated. Machine learning algorithms, including deep neural network models, will be used to create estimates of BAC from transdermal sensor data. These estimates will be examined in terms of their accuracy, temporal specificity, and also context-dependence. Thus, results will carry significance for addiction science by translating transdermal sensor data and clarifying the place of these sensors in our arsenal of techniques for assessing, preventing, and treating problem drinking.