Banner researchers using machine learning to predict dementia

Researchers at Banner Alzheimer’s Institute within Banner Health System have been working to apply machine learning technology to predicting Alzheimer’s disease.

They are working on a very large administrative claims data set of 125 million patients, seeking to train a machine learning model to predict new cases of Alzheimer’s disease four or five years in advance of symptoms.

“We hypothesized that machine learning algorithms using administrative claims data may represent a novel approach to predicting Alzheimer’s disease and related dementias (ADRD)” the scientists note. “Machine learning algorithms can rapidly develop predictive models for ADRD with massive datasets, without requiring hypothesis-driven feature engineering.”

For the most part, however, Banner served as a member of an expert panel and did not have a direct role in the design, execution of the study, or development of a research paper. Optum Lab designed the study, collected the data, performed the analysis and wrote the article.



Also See: Machine learning helps Northside track insurance payments

Worldwide, 50 million people have dementia, with about 10 million new cases annually, according to the World Health Organization. Alzheimer’s disease is the most common form of cognitive impairment, contributing as many as 70 percent of new cases.

As many as 35 million people lived with dementia worldwide in 2010, and those numbers were expected to double every 20 years, reaching 115 million by 2050. The costs in the United States were projected to reach nearly $250 billion by 2018.



Machine learning algorithms have been used previously for developing predictive models on large administrative claims datasets. In a recently published paper, other researchers developed models using machine learning algorithms on electronic health records data to predict tasks such as a patient’s final discharge diagnosis, or 30-day unplanned readmission. Still, more researchers used machine learning algorithms with medical and pharmacy claims for 473,049 people to identify those at risk for type 2 diabetes.

“The model for using machine learning to apply large datasets to a cohort of patients becomes increasingly useful as potential disease-modifying treatments for dementia move toward a stage for clinical testing,” according to the Banner scientists.

“Thus, the ability to achieve a lift of 6.4 means that a patient identified by the model will be 6.4 times more likely to be diagnosed in the near-term with dementia. An identified cohort with such enhanced prior probability could be much more cost-effectively screened for clinical research than an unselected population.”

The complete report, “Identifying incident dementia by applying machine learning to a very large administrative claims dataset,” is available in the July issue of the PLOS ONE journal.