Blood and Breath Alcohol Concentration from Transdermal Alcohol Biosensor Data: Estimation and Uncertainty Quantification via Forward and Inverse Filtering for a Covariate-Dependent, Physics-Informed, Hidden Markov Model

Signal processing and machine learning with transdermal alcohol concentration to predict natural environment alcohol consumption

Wrist-worn alcohol biosensors continuously and discreetly record transdermal alcohol concentration (TAC) and may allow alcohol researchers to monitor alcohol consumption in participants' natural environments. However, the field lacks established methods for signal processing and detecting alcohol events using these devices. We developed software that streamlines analysis of raw data (TAC, temperature, and motion) from a wrist-worn alcohol biosensor (BACtrack Skyn) through a signal processing and machine learning pipeline: biologically implausible skin surface temperature readings (< 28°C) were screened for potential device removal and TAC artifacts were corrected, features that describe TAC (e.g., rise duration) were calculated and used to train models (random forest and logistic regression) that predict self-reported alcohol consumption, and model performances were measured and summarized in autogenerated reports. The software was tested using 60 Skyn data sets recorded during 30 alcohol drinking episodes and 30 nonalcohol drinking episodes. Participants (N = 36; 13 with alcohol use disorder) wore the Skyn during one alcohol drinking episode and one nonalcohol drinking episode in their natural environment. In terms of distinguishing alcohol from nonalcohol drinking, correcting artifacts in the data resulted in 10% improvement in model accuracy relative to using raw data. Random forest and logistic regression models were both accurate, correctly predicting 97% (58/60; AUC-ROCs = 0.98, 0.96) of episodes. Area under TAC curve, rise duration of TAC curve, and peak TAC were the most important features for predictive accuracy. With promising model performance, this protocol will enhance the efficiency and reliability of TAC sensors for future alcohol monitoring research. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

A population model-based linear-quadratic Gaussian compensator for the control of intravenously infused alcohol studies and withdrawal symptom prophylaxis using transdermal sensing

An output feedback LQG compensator (combined controller and state estimator) for the regulation of intravenous-infused alcohol studies and treatment using a noninvasive transdermal alcohol biosensor is developed. The design is based on a population model involving an abstract semi-linear parabolic hybrid reaction-diffusion system involving coupled partial and ordinary differential equations with random parameters known only up to their distributions. The scheme developed is based on a weak formulation of the model equations in an appropriately constructed Gelfand triple of Bochner spaces wherein the unknown random parameters are treated as additional spatial variables. Implementation relies on a Galerkin-based approximation and convergence theory and an abstract formulation involving linear semigroups of operators. The model is fit and validated using laboratory collected human subject data and the method of moments. The results of numerical simulations of controlled intravenous alcohol infusion are presented and discussed.

Computation of nonparametric, mixed effects, maximum likelihood, biosensor data based-estimators for the distributions of random parameters in an abstract parabolic model for the transdermal transport of alcohol

The existence and consistency of a maximum likelihood estimator for the joint probability distribution of random parameters in discrete-time abstract parabolic systems was established by taking a nonparametric approach in the context of a mixed effects statistical model using a Prohorov metric framework on a set of feasible measures. A theoretical convergence result for a finite dimensional approximation scheme for computing the maximum likelihood estimator was also established and the efficacy of the approach was demonstrated by applying the scheme to the transdermal transport of alcohol modeled by a random parabolic partial differential equation (PDE). Numerical studies included show that the maximum likelihood estimator is statistically consistent, demonstrated by the convergence of the estimated distribution to the "true" distribution in an example involving simulated data. The algorithm developed was then applied to two datasets collected using two different transdermal alcohol biosensors. Using the leave-one-out cross-validation (LOOCV) method, we found an estimate for the distribution of the random parameters based on a training set. The input from a test drinking episode was then used to quantify the uncertainty propagated from the random parameters to the output of the model in the form of a 95 error band surrounding the estimated output signal.

An Abstract Parabolic System-Based Physics-Informed Long Short-Term Memory Network for Estimating Breath Alcohol Concentration from Transdermal Alcohol Biosensor Data

The problem of estimating breath alcohol concentration based on transdermal alcohol biosensor data is considered. Transdermal alcohol concentration provides a promising alternative to classical methods such as breathalyzers or drinking diaries. A physics-informed long Short-term memory (LSTM) network with covariates for the solution of the estimation problem is developed. The data-driven nature of an LSTM is augmented with a first principles physics-based population model for the diffusion of ethanol through the epidermal layer of the skin. The population model in an abstract parabolic framework appears as part of a regularization term in the loss function of the LSTM. While learning, the model is encouraged to both fit the data and to produce physically meaningful outputs. To deal with the high variation observed in the data, a mechanism for the uncertainty quantification of the estimates based on a recently discovered relation between Monte-Carlo dropout and Bayesian learning is used. The physics-based population model and the LSTM are trained and tested using controlled laboratory collected breath and transdermal alcohol data collected in four sessions from 40 orally dosed participants (50% female, ages 21 - 33 years, 35% BMI above 25.0) resulting in 256 usable drinking episodes partitioned into training and testing sets. Body measurement (e.g. BMI, hip to waist ratio, etc.), personal (e.g. sex, age, race, etc.), drinking behavior (e.g. frequent, rarely, etc.), and environmental (e.g. temperature, humidity, etc.) covariates were also collected from participants. The importance of various covariates in the estimation is investigated using Shapley values. It is shown that the physics-informed LSTM network can be successfully applied to drinking episodes from both the training and test set, and that the physics-based information leads to better generalization ability on new drinking episodes with the uncertainty quantification yielding credible bands that effectively capture the true signal. Compared to two machine learning models from previous studies, the proposed model reduces relative L 2 error in estimated breath alcohol concentration by 58% and 72%, and relative peak error by 33% and 76%.