A Multivariable Model to Predict Risk of Dementia or Death in Older Veterans With Traumatic Brain Injury

Authors:
Publisher website
Google Scholar
Abstract
Traumatic brain injury (TBI) is associated with increased risk of both death and dementia in older adults, but no prognostic models currently exist to help identify which individuals are at greatest risk of long-term adverse outcomes. Our primary objective was to develop a single prognostic model that uses electronic health record (EHR) data to predict 5-year risk of dementia or death in older veterans who have experienced TBI. We performed a retrospective cohort study using EHR data from the US Department of Veterans Affairs Health Care System. Participants were US veterans aged 55 years or older without dementia who had a TBI diagnosis and received care between 2002 and 2019. Potential predictor variables included demographics, medical comorbidities, mental health conditions, and health care utilization factors. We used multinomial logistic regression with least absolute shrinkage and selection operator to develop a prognostic model for 5-year risk of dementia or death. Cohort members (N = 113,779) had a mean ± SD age of 68 ± 10 years; 6% were female, and 20% were non-White or Hispanic. The key predictors of dementia or death included older age; male sex; number of inpatient visits; and comorbidities such as Parkinson disease, diabetes, weight loss, epilepsy, substance use disorder, and psychosis. The c-statistics were 0.756 (95% CI 0.751-0.761) for dementia and 0.783 (0.779-0.786) for death. Among those in the lowest predicted decile of dementia risk, 3% developed dementia within 5 years, compared with 43% of those in the highest predicted decile of risk. Similarly, the proportion of participants who died without dementia was 4% among those in the lowest predicted decile of risk compared with 64% in the highest predicted decile of risk. A single prognostic model that includes demographics, comorbidities, and utilization variables can predict risk of dementia or death in older veterans who have experienced TBI with good accuracy. Additional studies are needed to determine how this information might be best used to support better care.

This publication has 33 references indexed in Scilit: