Blood and Breath Alcohol Concentration from Transdermal Alcohol Biosensor Data: Estimation and Uncertainty...

Area of Science:

• Biomedical Engineering
• Data Science
• Pharmacokinetics

Background:

• Current alcohol monitoring relies on breathalyzers and diaries, limiting continuous data collection.
• Accurate conversion of transdermal alcohol concentration (TAC) to blood/breath alcohol concentration (BAC/BrAC) is crucial for practical biosensor application.

Purpose of the Study:

• To develop a novel method for estimating breath alcohol concentration (BrAC) from transdermal alcohol concentration (TAC) using advanced modeling techniques.
• To integrate physical principles of ethanol transport with data-driven learning for improved accuracy.

Main Methods:

• A covariate-dependent, physics-informed hidden Markov model with two emissions was developed.
• The model combines a hidden Markov chain for alcohol dynamics with a bivariate normal distribution for BrAC and TAC, incorporating covariates.
• A hybrid approach regularized the hidden Markov model with a first-principles partial differential equation (PDE) model, trained using the Baum-Welch algorithm.

Main Results:

• The physics-informed Viterbi algorithm was used for forward filtering TAC to estimate BrAC.
• The model achieved good agreement between estimated and actual BrAC, with a median relative peak error of 22% and a median relative area under the curve error of 25% on the test set.
• Non-physical artifacts in BrAC estimates were eliminated by the physics-informed Viterbi algorithm.

Conclusions:

• The developed physics-informed hidden Markov model provides an accurate and reliable method for estimating BrAC from TAC.
• This approach enhances the utility of transdermal alcohol biosensors for continuous alcohol monitoring in research and clinical settings.
• The hybrid modeling strategy effectively combines physical understanding with data-based learning for robust estimations.