This year, OptumLabs began a research project using artificial intelligence to predict Alzheimer's disease.
The long-term project is being done in conjunction with pharmaceutical company and other sponsors to identify clusters in a population at risk of developing Alzheimer's, which could leadi to the development of a drug and clinical trials.
"A lot of researchers are looking at genomics and genetic markers," OptumLabs CEO Paul Bleicher said. "We're approaching this from a different perspective,"
OptumLabs is the research and innovative collaborative within Optum.
WHY THIS IS IMPORTANT
One in 10 people age 65 and older has Alzheimer's dementia, according to the Alzheimer's Association.
The cost of treating the disease is projected to surpass $277 billion this year.
The consensus is that there have not been medications that have knocked it out of the park, Bleicher said. The benefits of current drugs have been modest.
Part of the issue may be that by the time people are identified with Alzheimer's through clinical trials, it's too late for early intervention.
In theory, the earlier individuals are identified, the more effective treatment will be.
OptumLab's goal is to identify patients most at risk for developing Alzheimer's early on, ideally, 10 years prior to onset on the disease.
OptumLabs wanted to have a large enough population for its research using information that was readily available. Genome sequencing didn't offer this because of the limited number of people who have done this.
OptumLabs created two models: a regression model based on best knowledge about Alzheimer's and a machine learning model.
Researchers quickly determined that the machine learning model produced as good, and in some cases, better results by combining information in the electronic health record and claims data.
EHRs and claims have clues to detect Alzheimer's in lab test results, claims for falls and even cognitive impairment, Bleicher said.
But what OptumLabs found even more valuable were the notes in the EHR.
"It's been estimated that 70-80 percent of information in the chart are within notes than is in structured information," Bleicher said. "What we're targeting is, getting information from those notes."
Natural language processing is able to extract the difference in timing when something is mentioned so the progression of the language can be followed over years.
Optum is now creating a model to use a form of unstructured deep machine learning to identify clusters of patients in a de-identified manner. Similar concepts tend to cluster together. Once the clusters are defined, the deep learning techniques can be used to make predictions.
"That's an exciting aspect of this because of the good results we've had in other settings in examining notes," Bleicher said. "What we hope is to be able to get to the ability to identify those clusters of patients that represent Alzheimer's and related dementias. We believe the deep learning can predict this faster."
The next phase of the project started this year and the research is expected to take several years. OptumLabs is not looking at this as a commercial enterprise, Bleicher said, but in 2019 it anticipates a first look at some of the results. Where it goes from there depends on those results.
If AI can predict people's progression in the clusters, it might be possible for a model like this to be used against a patient database to identify those who might be candidates for early trials.
Email the writer: email@example.com