Using phone sensors and an artificial neural network to detect gait changes during drinking episodes in the natural environment.
Brian Suffoletto, Pedram Gharani, Tammy Chung and 1 others
PMID 29179052WHAT IT FOUND
Phone sensors and a brief gait task captured drinking-related movement data in 10 young adults.
Sensor features tracked estimated blood alcohol level with small errors, though some readings were predicted less well. This is early feasibility, not a validated clinical test.
Key findings
01Ten participants provided 128 unique data points and 38 drinking episodes, with each participant reporting at least 3 drinking episodes.
02Almost half of gait tasks (n=60, 46.9%) were completed before drinking or on non-drinking evenings, 55 (43.0%) were completed while estimated blood alcohol concentration was rising, and 13 (10.1%) were completed while it was falling.
03The best model predicted eBAC from phone sensor gait features, with more than 95% of errors between −0.012 and +0.012, and very rarely (<20%) actual eBAC values greater than the legal limit (0.08 mg/dl) were classified as below the legal limit (0.08 mg/dl).
STILL TO COME
How it was doneWhat they foundWhat it means for PTs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
Only 10 participants were studied. Participants were recruited from an emergency department and were mostly female and white, limiting generalizability. The app was made only for iOS devices, so results may not apply to other mobile devices. Almost half of the gait tasks were completed before drinking or on non-drinking evenings, and only 13 tasks were completed during the descending eBAC limb. Many gait task data points were missing later in the evenings. Completion rates may have been inflated because participants were paid to complete tasks. eBAC was based on self-reported drink counts, not breath or transdermal alcohol sensors. The model was tested on a small dataset and had outlier errors.
Declared interests
The authors declared no conflicts of interest. The supplied text does not state a funding source.
The easy way to misread this
Do not conclude that a phone app can reliably identify alcohol intoxication for clinical use. This was a 10-person pilot, almost half of the gait tasks were completed before drinking or on non-drinking evenings, and the model still produced outlier errors.