aceLong-term prediction of epilepsy following traumatic brain injury among veterans using routine clinical data
Objective Despite elevated risk for epilepsy following traumatic brain injury (TBI), there are limited tools to assess epilepsy risk following TBI using routine clinical data. The objective of this study was to develop and validate a machine learning approach to predict the onset of posttraumatic epilepsy (PTE) over varying time horizons following TBI, using only routine clinical data collected up to the month of TBI documentation. Methods This retrospective longitudinal cohort study included post-9/11 US vetera…
Hospital Militar Doctor Ramon de Lara, F.A.D by Cheposo. CC BY-SA 3.0 · https://creativecommons.org/licenses/by-sa/3.0
Researchers developed and validated machine learning models to predict posttraumatic epilepsy onset at 2, 5, and 10 years after first TBI documentation in 107,987 post-9/11 US veterans, using only routine preinjury clinical data up to the month of injury. An optimized random forest achieved AUCs of 0.75 to 0.73 across horizons on held-out test data and enabled high-risk stratification.
The ability to stratify long-term epilepsy risk from widely available administrative data matters because there are limited tools to assess PTE risk, and enrichment of high-risk individuals could make preventative trials feasible. Uncertainty remains about performance outside the veteran health system, reliance on retrospective documentation, and whether prediction translates into effective prevention.
- Retrospective cohort of 107,987 post-9/11 US veterans with TBI diagnosis between 2008 and 2017, with 4,930 (4.6%) incident epilepsy cases.
- Models used only preinjury information collected up to month of TBI documentation, evaluated on 30% held-out test data.
- SHAP analysis identified TBI severity and cumulative preinjury comorbidity index as important predictors of PTE risk.
- Authors propose use for enrichment of PTE cases in future preventative clinical trials using widely available routine clinical data.
A random forest model using only routine preinjury clinical data predicted posttraumatic epilepsy onset up to 10 years after TBI in veterans, achieving AUC around 0.73-0.75 and identifying 17.5% of 5-year cases at 2.3% false positive rate.
The rundown
The study used routine administrative health data from 107,987 post-9/11 veterans diagnosed with TBI between 2008 and 2017, of whom 4.6% developed incident epilepsy. Prediction used only information available up to the month of first TBI documentation, with no post-injury clinical features.
An optimized random forest was trained to forecast epilepsy at 2, 5, and 10 years and tested on a 30% held-out set, yielding AUCs of .75, .74, and .73 respectively. High-risk stratification captured 17.5% of 5-year cases at low false positive rate, and SHAP values highlighted TBI severity and preinjury comorbidity burden as key drivers.
Sources
- Peer-reviewedEpilepsia2026-08-04
How should this claim be treated?
The debate