Algorithm Predicts Time from Alzheimer's Onset to Nursing Home, Death

Using data from a patient's first visit, an algorithm created by P&S researchers can predict time to full-time care, nursing home residence, or death for patients with Alzheimer's disease.

"Predicting Alzheimer's progression has been a challenge because the disease varies significantly from one person to another. Two Alzheimer's patients may both appear to have mild forms of the disease, yet one may progress rapidly while the other progresses much more slowly," says the algorithm's lead developer Yaakov Stern, PhD, professor of neuropsychology (in neurology, psychiatry, and psychology and in the Taub Institute for Research on Alzheimer's Disease and the Aging Brain and the Gertrude H. Sergievsky Center). "Our method enables clinicians to predict the disease path with great specificity."

The new algorithm differs from previous methods of predicting Alzheimer's progression in that it does not rely on unusual clinical tests. Data for the algorithm can be obtained with routine clinical tests during a patient's first visit. Sixteen types of data are used in the calculation, including scores on cognitive tests, behavioral symptoms, and heart disease risk factors.

"It is more practical for routine use and may become a valuable tool for physicians and patients' families," said Nikolaos Scarmeas, MD, a study co-author and associate professor of neurology in the Taub Institute and the Sergievsky Center.

The algorithm is based on a complex model of Alzheimer's disease progression (Longitudinal Grade of Membership, or L-GoM) that the researchers developed by following a group of 252 Alzheimer's patients every six months for 10 years. The researchers tested the algorithm with data collected from the initial visits of 254 other patients who were also followed for 10 years.

The results, published in the Jan. 1, 2014, issue of the Journal of Alzheimer's Disease, demonstrated that the times predicted by the algorithm closely matched the actual times collected from the second group of patients. "We know of no other prediction algorithm that approaches this level of performance," the authors wrote in the paper.

The advantage of the L-GoM model is that it takes into account the complexity of Alzheimer's disease and does not assume the disease progresses through discrete stages. "Patients don't typically fall neatly into mild, moderate, or severe disease categories," said Dr. Stern, who also directs the Cognitive Neuroscience Division in the Department of Neurology at P&S. "For example, a patient may be able to live independently yet have hallucinations or behavioral outbursts."

Dr. Stern and his team are now developing a computer program that would allow clinicians to input the variables and receive a report. They expect the program to become available within the next two years.

The researchers are testing the algorithm on a third group of patients from the Washington Heights-Inwood Columbia Aging Project, which contains a more diverse population than the primarily white, educated, and relatively wealthy population represented in the initial groups.