Using machine learning for real-time BAC estimation from a new-generation transdermal biosensor in the laboratory

Abstract

Background: Transdermal biosensors offer a noninvasive, low-cost technology for the assessment of alcohol consumption with broad potential applications in addiction science. Older-generation transdermal devices feature bulky designs and sparse sampling intervals, limiting potential applications for transdermal technology. Recently a new-generation of transdermal device has become available, featuring smartphone connectivity, compact designs, and rapid sampling. Here we present initial laboratory research examining the validity of a new-generation transdermal sensor prototype.

Methods: Participants were young drinkers administered alcohol (target BAC = .08 %) or no-alcohol in the laboratory. Participants wore transdermal sensors while providing repeated breathalyzer (BrAC) readings. We assessed the association between BrAC (measured BrAC for a specific time point) and eBrAC (BrAC estimated based only on transdermal readings collected in the immediately preceding time interval). Extra-Trees machine learning algorithms, incorporating transdermal time series features as predictors, were used to create eBrAC.

Results: Failure rates for the new-generation prototype sensor were high (16 %-34 %). Among participants with useable new-generation sensor data, models demonstrated strong capabilities for separating drinking from non-drinking episodes, and significant (moderate) ability to differentiate BrAC levels within intoxicated participants. Differences between eBrAC and BrAC were 60 % higher for models based on data from old-generation vs new-generation devices. Model comparisons indicated that both time series analysis and machine learning contributed significantly to final model accuracy.

Conclusions: Results provide favorable preliminary evidence for the accuracy of real-time BAC estimates from a new-generation sensor. Future research featuring variable alcohol doses and real-world contexts will be required to further validate these devices.

Keywords: Alcohol; Biosensor; Blood alcohol concentration; Machine learning; Real-time; Transdermal.

Copyright © 2020 Elsevier B.V. All rights reserved.

Figures

Figure 1.

AMS SCRAM ankle bracelet (left) and BACtrack Skyn wrist monitor (right) displayed side-by side. The approximate weights of the devices are 6oz (SCRAM) and 1oz (Skyn prototype).

Figure 2.

A visual representation of the data analysis plan employed in the current project. Data analysis involved the extraction of multiple time series features (e.g., mean, trends, periodicity) from the 30 minutes of raw TAC data that preceded each breathalyzer (BrAC) reading. These time series features were then entered as predictors into Extra-Trees machine learning algorithms to create estimates of BrAC from transdermal data (eBrAC). The top panel provides a broad visual depiction of the entire analysis process, the bottom left panel provides examples of a subset of time series features extracted (see Table 1 for additional features), and the bottom right panel provides a flow chart of machine learning modeling procedures.

Figure 3.

Graphs for participants with the “best” (minimum MAE), “worst” (maximum MAE), and average (Median MAE) prediction accuracy from both alcohol and no-alcohol (control) conditions in the current study. Precise average MAEs for alcohol condition graphs shown above are as follows: best case MAE=0.006; worst case MAE=0.028; median case MAE= 0.013. Precise average MAEs for the no-alcohol (control) condition graphs are as follows: best case MAE=0.000; worst case MAE=0.011; median case MAE=0.001. Baseline standardization procedures were applied to all Skyn data, as described in the Data Analysis Plan. For the purposes of graphs displayed here, data from Skyn was transformed (divided by 20,000) such that it could be visualized on approximately the same scale as eBrAC and BrAC.