Cerner AI helps Memorial Hermann document its diagnoses
Houston-based Memorial Hermann Health System had a problem: Its providers often were not documenting all the diagnoses managed in the acute care setting.
Coders and clinical documentation-improvement specialists spend several hours each week educating physicians on the implications of missing or nonspecific diagnosis. Over the last several years, the informatics division partnered with CDI specialists in the acute care venues to design and implement interruptive EHR alerts that prompt physicians to enter diagnoses when specific criteria are met.
Clinical documentation improvement queries
"The predominant method for capturing the missing or incomplete diagnoses in our organization is CDI queries that often are several days after the condition was managed; the CDI query process is not within the physicians' workflow, and the newly entered diagnosis rarely gets propagated to future physician documentation," explained Dr. Nnaemeka Okafor, vice president and chief analytics and informatics officer at Memorial Hermann.
So the provider organization looked to its EHR vendor for a solution. Chart Assist with Cerner.AI abstracts data from a patient's chart, concurrent with their hospital visit, and identifies opportunities to help ensure that the documentation accurately reflects the care provided.
Cerner's proposal was to implement Chart Assist with Cerner.AI as the EHR prompts are as near real time as possible and within the providers' EHR MPage workflow. The entered diagnoses would be part of the providers' current note and future provider documentation as well.
"Using Cerner's natural language processing engine, along with clinical rules and algorithms, documentation is parsed and chart data is collected, looking for these opportunities."
Dr. Nnaemeka Okafor, Memorial Hermann
"Using Cerner's natural language processing engine, along with clinical rules and algorithms, documentation is parsed and chart data is collected, looking for these opportunities," Okafor explained. "The content leverages Cerner's Ontology to label and classify structured data from the EHR and Cerner's Knowledge Library to group related medical concepts and diagnoses together, thus supporting semantic interoperability."
A natural part of chart review
These opportunities then are presented to the provider within their workflow, making it a natural part of their chart review process and supporting actions they can quickly take to update their documentation.
The AI portion of Chart Assist parses the health record to identify clinical evidence that supports missing diagnoses and diagnoses that lack specificity. It can also identify documented diagnoses that lack appropriate clinical evidence.
Memorial Hermann implemented the tool as a pilot with its hospitalist providers as they utilized the view of the EHR that incorporated the Chart Assist module. The tool did not need to be integrated in any other system.
Chart-assisted recommendations accepted
The results were clear.
"The ratio of acceptance to rejection of the chart-assisted recommendations are 6:1 for the six diagnoses included in the pilot," Okafor reported. "It's been well accepted by many of the hospitalist providers. A few diagnoses have higher acceptance ratios than others."
Okafor advised peers using similar technology to include the end users in the design of the tool, solicit feedback as early as possible and ensure that the IT staff is responsive to user feedback.
The Diverse Roles of AI & ML
Machine learning applications are proliferating and improving, and this month we'll explore the many ways they're being put to work.