Does a Wrist Sensor’s BAC Display Validate Its Blood-Alcohol Estimate? An Eight-Volunteer Modeling Study

A displayed “BAC” value is not, by itself, evidence that a wrist sensor accurately measured blood alcohol. An eight-volunteer pilot study developed a skin-diffusion model from a prototype sensor’s readings. It demonstrated how a model can fit those readings, while leaving important questions about independent blood validation unresolved.

What the device actually sensed

Jalal and colleagues built a wrist-worn device using a fuel-cell sensor. Ethanol reaching the sensor produced an electrical signal, which was processed through calibration and conversion steps to generate estimated alcohol values.

The underlying sample was alcohol diffusing outward through skin and across an air gap. This was not a blood specimen, a breath specimen or a urine ethanol test. The paper’s use of “BAC” for derived values should not obscure that measurement chain.

Eight volunteers, including two women, wore the prototype under controlled temperature and humidity. The protocol described six 50-milliliter servings of 35% alcohol at ten-minute intervals. Skin-signal collection was used to characterize the resulting time profiles.

Fitting an average curve is a different test from blood validation

The researchers aligned and averaged the eight volunteers’ sensor profiles, then compared three assumed input shapes in a computational diffusion model. An exponential-linear input produced a lower cumulative fitting error than piecewise-linear or Hoerl inputs.

That result identifies which of the tested shapes reproduced the averaged sensor data more closely. It does not establish that the inferred input was the volunteers’ true blood-alcohol curve.

The reported protocol did not provide a simultaneous venous-blood reference series for the eight volunteers. A roughly 14.15% deviation discussed in the paper referred to earlier calibration work. It should not be recast as a demonstrated error bound for this study’s person-specific blood estimates.

Skin thickness and the air gap were model assumptions

The model represented several epidermal layers plus a 0.1-centimeter air gap. Skin-thickness information came from a separate ultrasound study on upper arms, rather than individualized wrist measurements for every alcohol trial.

When modeled layer thicknesses were reduced by 20%, the predicted output became more compact and faster. Increasing thickness broadened and lowered the predicted curve. These were sensitivity simulations, not experimental proof that each volunteer’s skin had changed thickness.

This distinction matters when interpreting an individual’s trace. A good fit can depend on assumed thickness, diffusivity, sensor placement and the selected mathematical input. Those parameters require evidence rather than being established by the smooth appearance of a fitted curve.

A modeled peak delay is not a universal clock

The model predicted a delay between the input peak and the outer-skin concentration peak. Its size changed with the assumed input concentration and absorption duration. The approximately 60-minute skin-peak timing discussed for one analysis was not a universal delay measured directly between blood and wrist peaks in every participant.

Consequently, this pilot cannot justify subtracting a fixed number of minutes from any wearable peak to determine when a driver reached a particular blood concentration. It also did not measure impairment, driving performance or collisions.

What a forensic opinion would need

The paper provides an engineering starting point, with a small sample, controlled conditions and averaged data. An individual reconstruction would additionally need appropriate reference measurements, individual prediction errors, validation under relevant conditions and uncertainty around the estimate.

The site’s Bayesian skin-to-breath reconstruction analysis addresses a separate modeling approach and the role of uncertainty bands. Both topics require keeping the measured skin signal separate from the concentration estimated in another specimen.

Primary source

Jalal AH, Arbabi S, Ahad MA, Alam F, Ahmed MA. Wearable Alcohol Monitoring Device for the Data-Driven Transcutaneous Alcohol Diffusion Model. Sensors. 2024;24:4233. https://doi.org/10.3390/s24134233.