Can Artificial Intelligence Solve Our $125 Billion a Year Medical Records Problem?
By Carm Huntress, CEO, Credo.
It’s the 21st century, and 78% of hospitals still “often or sometimes” receive their medical records via mail or fax, according to a 2021 report from the Office of the National Coordinator for Health Information Technology.
This isn’t just an inconvenience, it’s a massive expense and time commitment for providers. It’s also dangerous for patients; when a provider doesn’t have an up-to-date, readable medical history, it can delay vital diagnoses and prevent patients from receiving timely, potentially life-saving care.
If there is one problem we can solve in healthcare that will positively impact patients, providers and payers alike, this is it. And thanks to some important regulatory and technological advances — including artificial intelligence — we’re starting to see a large group of leaders and innovators come together to tackle this challenge head on.
Has our investment in electronic health records paid off?
The U.S. government has spent more than $30 billion on incentivizing the use of EHRs since the passage of the HITECH Act of 2009. Unfortunately, focusing only on implementation and not on standardization created a new problem to solve.
Without a standard format or structure for records, we ended up with hundreds of thousands of systems exchanging unstructured data — it wasn’t until 2014 that the first official Fast Healthcare Interoperability Resources standards were published.
In addition to the issues around interoperability, we’ve also lacked a standard process for inputting the physician progress note into the EHR. An enormous amount of clinical value is found in that unstructured note, and without a readable format, that information goes to waste.
Because of these problems — along with slow adoption of new technologies among clinicians — healthcare is still faxing billions of pages of medical records every year.
New regulations and standards can help solve the interoperability problem
Recent advances in regulation and standardization are laying the groundwork for important progress in the coming years. Over the past decade, the implementation of standards like FHIR has created a standardized format for medical record data.
And with the 21st Century Cures Act, we’ll soon see the emergence of a regulated, mandated, and interoperable national network under the Trusted Exchange Framework and Common Agreement.
It will take an enormous amount of effort and collaboration to implement these changes at scale. But with the lessons learned since 2009 in mind, these regulations provide a vital foundation that was missing from previous legislation.
Artificial intelligence and machine learning have a vital role to play
Advancements in artificial intelligence and machine learning offer an unprecedented opportunity to synthesize medical record data into a readable, structured form.
Consider the volume of care a patient receives throughout their entire lifetime. Especially for an individual with one or more chronic conditions, one patient’s entire history is often spread across hundreds of pages of documents, with no means to connect them into a structured narrative.
Today, when we pull data on a patient digitally, we get an average of 43 clinical documents per patient. These can be in any format — HL7, FHIR, CCDA, or even a JPEG or TIFF. And then within each file, there’s an enormous amount of unstructured clinical narrative.
Artificial intelligence can play a key role in synthesizing these records — extracting diagnoses, lab results, medications, procedure history, and more — into a finely-tuned, digital record that’s fully searchable and comparable.
These records will reduce both the cost of treatment and time to treatment, making it easier for clinicians to provide the care patients need, when they need it.
Using existing technologies to build a scalable, user-friendly solution for medical records is not just a useful innovation. It is a vital step we must take if we want to solve the $125 billion problem of medical record retrieval and analysis.