Routinely collected patient information may help predict Alzheimer’s risk
Click Here to Manage Email Alerts
A patient’s risk for developing Alzheimer’s disease and related dementias can be predicted using information collected during routine doctor’s visits, according to findings from a pair of studies.
“Existing data/passive data captured by the patient's electronic health record from previous encounters with the health care system can be employed to detect future risk for developing Alzheimer’s disease,” Malaz Boustani, MD, MPH, research scientist at Regenstrief Institute in Indiana, told Healio Psychiatry. “This low-cost approach can identify patients with Alzheimer’s disease and thus improve the care for their other medical conditions.”
In studies published in Journal of the American Geriatrics Society and Artificial Intelligence In Medicine, Boustani and colleagues published findings of two separate machine learning approaches. In the first study, they analyzed data from a natural language processing algorithm. In the second, they reported the results of a random forest model. They trained the algorithms using data gathered from patients included in the Indiana Network for Patient Care — a regional health information exchange in Indiana. To predict the onset of dementia, the researchers included information on diagnoses and prescriptions, as well as medical notes, in the models.
The researchers reported that models used in the Journal of the American Geriatrics Society study had 1- to 10-year, 3- to 10-year and 5- to 10-year sensitivity that ranged from 51% to 62% and specificity that ranged from 80% to 89%. In the Artificial Intelligence In Medicine study, they noted that a combined model is generalizable across multiple institutions and can predict dementia within 1 year of its onset with approximately 80% accuracy, despite the fact that it was trained using routine care data. Further, free text medical notes were the most valuable identifier for those at risk for developing Alzheimer’s and related dementias. The researchers also noted that these methods may provide significant cost savings for health systems and patients by replacing the need for expensive tests and allowing clinicians to screen entire populations to identify patients who are most at risk.
“I was very surprised by the high level of accuracy of the passive digital markers of detecting patients with Alzheimer’s disease and related dementia 1 to 3 years before the diagnosis of the disease,” Boustani said. – by Joe Gramigna
Disclosures: Boustani reports being the founding director of the Sandra Eskenazi Center for Brain Care Innovation. Please see the studies for all other authors’ relevant financial disclosures.