Advances in BAC Estimation

Alcohol abuse results in physical harm and mental malfunction and is responsible for 1 in 10 deaths among adults aged 20-64 years in most Western countries annually, resulting in a grand total of almost 90.000 individuals in the United States alone. Moreover, binge drinking (defined as 4 or more drinks for women on a single occasion, and 5 or more drinks for men on a single occasion) has been steadily on the rise among youngsters. Between 2014 and 2017, over a third of college students aged 18-20 reported binge drinking in the prior month. Even though it’s well known that alcohol impairs driving ability, many people frequently drive when drunk. In 2010, 47.2% of pedestrian fatalities and 39.9% of vehicle occupant fatalities were caused by drunk driving. However, in many driving under the influence (DUI) cases, drivers are unaware that they are over the legal driving limit.

Alcohol consumption raises the blood alcohol content (BAC) of drinkers, impacting their neuromotor and cognitive functions approximately 20 minutes after alcohol consumption. The BAC level measures the amount of alcohol in the blood, therefore traveling through the body to every organ, including the brain. In its simplest form, calculating a person’s BAC level is based on how much alcohol went into what kind of body over a period of how much time.

This article has been thought out to raise awareness on the risks that alcohol abuse still poses in 2020 and why self-monitoring is a crucial step in DUI prevention. First, we will focus on what happens biologically to the human body as soon as we ingest alcohol, trying to highlight which metrics we could use to evaluate its presence, besides BAC. Secondly, we will dive into the physical and psychological consequences that ingesting alcohol can have, especially in DUI scenarios, trying to understand what are the challenges to overcome to encourage people to self-monitor themselves. Thirdly, and lastly, we will discuss the academic contribution to BAC estimation, starting from E. Widmark’s early studies in the 1930s, that paved to way to modelling more and more accurate mathematical models of the alcohol absorption cycle, to recent advances that make extensive use of machine learning in conjunction with biometric data collected via IoT from wearables.

How does alcohol assumption happen?

How does alcohol get into your bloodstream?

How does alcohol get into your breath?

Relationship between BAC and BrAC

DUI prevention

What does DUI mean and how to avoid it?

Why alcohol addiction is still a risk in the 2020s?

Physical and psychological consequences

Self-monitoring is not a luxury

Commercial tools

Academic contribution to BAC estimation: from Widmark to machine learning

Over the years many different academic studies have been conducted in order to build tools that can help people self-monitor themselves. Passive methods to monitor intoxication assume a dual role, since they can be used to prevent mishaps as well as treat heavy drinkers.

• Use in prevention: Soft drinkers who are over the legal driving limit can receive just-in-time (JIT) notifications of excessive alcohol consumption, preventing drunk driving.

• Use in treatment: Hard drinkers’ patterns and the associated social contexts (e.g., time, place, social circle, etc.) can be logged continuously. Alcoholics can reflect on their drinking patterns of abuse and either self-correct or seek treatment in dedicated rehab facilities: Counsellors, for example, could use such records as evidence to prescribe treatment.

Deterministic models

Deterministic models require prior knowledge on what has been drunk, because the amount of alcohol ingested is a variable that plays a huge role in any formulation of theirs. This has historically resulted in the development of tools that ask users to self-report what they have drunk, which by nature is particularly tedious, prone to errors and not suitable to high-risk situations wherein users’ cognitive functions are often too compromised to manipulate a smartphone.

Widmark’s mathematical model

Henry’s extension

Other formulations of the body coefficient r

Machine learning models via IoT

A lot of recent studies have been carried out with the main objective of reducing the intrusiveness of Widmark-derived models. Machine learning models have been developed to do so, by automating BAC prediction through the analysis of drinking patters over a pre-determined span of time or the extraction of relevant features from biometric and spatial sensors installed in IoT wearables.

BAC from breathalysers data

BAC from gait pattern analysis

BAC from transdermal biosensors

Future research directions

Closing thoughts

Appendix A: Personal background

Appendix B: BAC effects

BAC effects on the human body

Trivia