"We will have clinicians in the loop for a period of time until we can actually trust that AI and those technology solutions are taking care of the patient in the way that is best."
I’ve been thinking about this quote a lot. In part, it horrifies me. My primary care provider is a true care partner, and I trust her judgment, knowing she will refer me to a specialist when an issue requires it. I’m not comfortable with the idea that she might be replaced by an AI chatbot.
But then, I never thought I’d be comfortable with my car making decisions on its own either, yet that has become my new normal.
On the other hand, I’m a lot more familiar with the details of health information quality and usability than I am with automotive issues. And the healthcare data on which we are relying for AI-based chatbots leaves a lot to be desired. In fact, a recent survey conducted by Sage Growth Partners found that only 20% of the healthcare executives they interviewed fully trust their data. Less than 8% of them judged their analytics capabilities as very mature, and less than a quarter have a data model that can ingest unstructured data – the bedrock of healthcare information.
The Tangible Benefits of Investing in Healthcare Data Quality
Sage just published another report, The Hidden Costs of Bad Healthcare Data, in which they discuss the costs associated with poor healthcare data quality. Looking at a variety of peer-reviewed and other credible data sources, they estimated the savings that could be realized by addressing a set of six issues, for example, the implications of duplicate patient records. They came up with a conservative estimate of more than $40 million in savings for a large US integrated delivery network over the space of three years, based on an investment in better healthcare data quality.
Frankly, I’m equally interested in the implications for better decision making, better care, and better AI. After all, that survey I mentioned earlier also found that 53% of respondents say poor healthcare data quality reduces their ability to make decisions. It also impacts their ability to identify gaps in care, meet quality metrics, and optimize the revenue cycle. I must assume it would also affect the results of any exercise in machine learning conducted on such data – garbage in, garbage out – right?
My car shows me that what I once thought the stuff of fantasy is entirely possible – with better quality health information, the same can be true for care, probably sooner than I expect.
Looking Towards a Future with AI Chatbots in Healthcare
Russ Leftwich, says that when he was in med school, the dean told them “Three-fourths of the patients you see will get better no matter what you do. Try not to screw up!” So if, with a healthy data foundation, we can produce virtual chatbots that can correctly distinguish the ¾, prescribe chicken soup and Tylenol, and do a reasonable job of triaging the other 25%, then maybe, one day, we’ll have enough healthcare resources to go around.