Fall risk classification with posturographic parameters in community-dwelling older adults: a machine learning and explainable artificial intelligence approach.
Huey-Wen Liang, Rasoul Ameri, Shahab Band and 5 others
PMID 38287415WHAT IT FOUND
Trunk sway measured by a tracker and a computer model sorted older adults into fall-risk groups better when risk was defined by a Timed Up and Go time of 10 seconds or more than by a fall in the past year.
Key findings
01Models classified risk better when the risk label was a Timed Up and Go time of 10 seconds or more than when it was a fall in the past year.
02The best reported discrimination came from the TUG-defined groups, the Slime Mould feature selection method, and the Easy Ensemble classifier.
03For the TUG-defined groups, the model explanations pointed to sway during feet-together eyes-closed standing and to personal metrics such as age.
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
The study was cross-sectional, so it classified fall-risk groups at one time and did not follow participants to see who fell later. The sample was relatively small and had more low-risk than at-risk participants, creating an imbalanced data set. Participants had to be able to walk independently for at least 10 m and were excluded for significant cognitive impairment, severe visual impairment, major neurological conditions, or major musculoskeletal disorders, so results may not apply to frailer or more impaired older adults. The mobility-based risk label used a Timed Up and Go cutoff of 10 seconds, which the paper notes is lower than some previously proposed cutoffs. Feature selection improved accuracy only for the mobility-based label, not for the fall-history label.
Declared interests
The study was supported by National Taiwan University Hospital Yulin Branch. The supplied text gives no further conflict-of-interest declaration.
The easy way to misread this
Do not conclude that sway-based machine learning predicts falls. The models classified groups defined by TUG time or past-year fall history, and the study did not follow participants for future falls.