Using machine learning to discover traumatic brain injury patient phenotypes: national concussion surveillance system Pilot.
Dana Waltzman, Jill Daugherty, Alexis Peterson and 1 others
PMID 38722037WHAT IT FOUND
Machine learning sorted self-reported concussion symptoms into five distinct groups.
One group had many symptoms and worse outcomes, including more social and work impairment. Another had headache alone and resolved quickly. These profiles may help target care, but the study did not test any treatment.
Key findings
01Unsupervised machine learning grouped respondents into five distinct TBI phenotypes based solely on self-reported signs and symptoms.
02Phenotype C, characterized by high symptom prevalence and cardinal symptoms like loss of consciousness, was associated with significantly worse outcomes, including higher odds of medical evaluation and greater impact on social and work functioning compared to Phenotype A.
03Phenotype A, characterized by headache as the only reported symptom, was associated with less severe outcomes and a high prevalence of symptoms resolving within one day.
STILL TO COME
How it was doneWhat they foundWhat it means for PTsWhat it means for OTsWhat it means for SLPs
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 relies entirely on self-reported symptoms from a telephone survey, which is subject to recall bias and lacks clinical validation or objective diagnostic confirmation. The data comes from a pilot survey designed for case ascertainment, not for producing nationally representative estimates of TBI, limiting the generalizability of the prevalence figures. The clustering was based only on signs and symptoms; it did not include biological biomarkers, neuroimaging, or clinical exam findings, which might refine the phenotypes further. The study is cross-sectional and observational; it identifies associations between symptom clusters and outcomes but does not establish causality or test the efficacy of phenotype-directed treatments.
Declared interests
No potential conflict of interest was reported by the authors. The study was supported by the National Institute of Health (Extramural) and the CDC (Non-U.S. Gov't).
The easy way to misread this
Do not assume these machine-learning-derived phenotypes are currently validated clinical diagnostic categories. This study describes a method for grouping patients based on self-report data; it does not prove that treating patients according to these specific clusters leads to better outcomes, nor does it replace standard clinical assessment.