Combing through insurance claims and other health data on more than 72 million U.S. residents, a machine learning algorithm was able to quite accurately identify more than 222,000 individuals who have very early stage Alzheimer’s disease.
And most of the records were from primary care, the ideal setting in which to begin intervention or at least monitoring progression of the disease.
A study report describing the work is running in the Journal of Prevention of Alzheimer’s Disease.
Corresponding author Sam Khinda and colleagues at IQVIA, the multinational contract research organization headquartered in Durham, N.C., trained their algorithm to recognize patterns predictive of Alzheimer’s in combinations of drug treatments, doctor visits, diagnostic tests, therapeutic procedures and confirmed clinical diagnoses.
The patients and healthy subjects ranged in age from 50 to 85, and around 667,000 of them had at least one record of Alzheimer’s diagnosis or treatment. Around 3.7 million were free of the disease and so served as the control cohort.
The team used several supervised machine-learning techniques to predict and classify the occurrence of prodromal, or initial-stage, Alzheimer’s.
One of the classifiers, a “gradient boosted tree” tool, performed the best, identifying 222,721 subjects in the prodromal Alzheimer’s stage with 80% precision.
Of these, some 76% were in the primary care setting.
This latter finding “could drive major advances in Alzheimer’s disease research by enabling more accurate and earlier prodromal Alzheimer’s disease diagnosis at the primary care physician level,” the authors concluded, “which would facilitate timely referral to expert sites for in-depth assessment and potential enrolment in clinical trials.”
Expounding on the relevance of the project, the authors commented on the difficulty of recruiting patients for clinical trials of potential therapies to treat Alzheimer’s. In the United States alone, they noted, 150 clinical trials are seeking 70,000 participants.
“Most people with early cognitive impairment consult primary care providers, who may lack time, diagnostic skills and awareness of local clinical trials,” Khinda et al. wrote. “Machine learning and predictive analytics offer promise to boost enrollment by predicting which patients have prodromal Alzheimer’s disease” and which will go on to develop the full condition.
Radiology Business Journal recently looked at the gap between surging diagnostics and stalled therapeutics in the race to counter Alzheimer’s as the population ages and cases multiply by the millions.