Machine Learning and Clinical Outcomes in Health Information Technology

The Disease Management Care Blog is guilty of not paying that much attention to the flow of news about information internet techie killer ap stuff. It’s hard to keep it all straight, much of it seems either impermanent or futuristic, plus the 2.0 jargon is opaque. But when the New York Times Business Section and the Wall Street Journal Health Blog both mention a new healthcare computer thingy, the DMCB perks up.

Welcome to Keas (pronounced KEE’_ahs). As the DMCB understands it, this is a web site (or to the cognoscenti, an ‘application’) in which you can record your medical and family history and input other medical information, including labs from Quest. The site’s ‘machine learning’ (MLing) can apparently fashion a health profile, interpret (‘red,’ ‘yellow,’ or ‘green’) lab test results, generate a personalized care plan (based on input from Healthwise or anyone of a number of high powered academics), issue prompts or offer specific suggestions (and even quizzes) that help the user-patient improve their health or manage conditions such as diabetes, high blood cholesterol or being overweight.

The DMCB views ‘MLing’ as a learning opportunity, so it dove right in. Keas’ co-founder Adam Bosworth is of Google-engineer pedigree and he specifically mentioned the concept in his blog. Assuming that’s what makes up Keas’ insides, MLing is computers that deploy algorithms to search for and ‘learn’ known and unknown patterns and make associations. Presumably this technology can tap into what is known about patient data and various diseases, much like credit card companies can spot unusual transactions and issue fraud alerts or Amazon can prompt customers with purchase recommendations.

What luster hm? This is part cool, part patient empowerment, part information tech, part meaningful EHR use and part venture capital.

Then the DMCB dug a little deeper. While a response from Keas to an email inquiry and a phone call is still pending at the time of this posting, the DMCB did what it always does when its curiosity is piqued. It took a look at some of the pertinent medical literature, favoring randomized prospective trials from reputable peer review journals. From a clinical standpoint that was curiously missing in the New York Times, two well done studies may lend some insight as to what Keas can, and cannot, do:

Check out this Annals of Internal Medicine study on self management in asthma in which the authors compared the outcomes of a group of patients randomly assigned to an ‘internet-based self management program’ with monitoring, advice, education and web-based communication versus usual care. Instead of MLing, a questionnaire was used to discern how patients were doing and, if things got bad, a live nurse intervened. At twelve months, the internet group had modestly better improvement in their quality of life and measures of lung function but there were not differences in severe asthma attacks.

Or how about this Archives of Internal Medicine study on the use of a ‘practice linked online personal health record’ for patients with diabetes. Once again, patients were randomly assigned to usual care versus a web application that listed medications, asked questions about the diabetes and other labs and then generated a care plan. At the end of one year, the intervention group had experienced greater changes in their medications, but there were no meaningful differences in blood sugar control, blood pressure and blood cholesterol levels.

Based on this information, the DMCB suspects that organizations that may consider paying for this service on behalf of their enrollees may be skeptical about the ability of Keas to lower claims expense enough to justify the investment. However, as pointed out before, Keas-like applications' greatest potential is when it's combined with other population based care interventions that synergistically add up to more than the sum of their parts.

Since the literature above may prompt some skepticism, Keas’ may wish to conduct some of its own studies to better define what it can and cannot do and how it best fits with the patient centered medical home, disease management, benefit-based insurance incentives, physician patient reform, accountable care organizations, registries and traditional electronic records.

Addendum: In looking around the web site, the DMCB also found these terms of service (bolding from the DMCB) that speak for themselves:

'The Content and Services may link you to other web sites or information, software, data, or other contents on or off the Internet, including linked click-through or other advertising, or through featured or sponsored sites. We have not reviewed the contents that may be reached by such links and we are not responsible for such content. Your linking to any other pages on other sites is at your own risk. The information, software, data, or other contents (including opinions, claims, comments) contained in linked references are those of the companies responsible for such sites and should not be attributed to us. We have not attempted to verify the truth or accuracy of any such opinion, claim, or comment, nor do we endorse or support them. We do not warrant, nor are we in any way responsible for, information, software, data, privacy policies, or other content that is outside of our control.'