Photo: Hero Images Inc./Getty Images
Suki, an ambient AI healthcare company, is expanding its assisted revenue cycle capabilities to provide more specific automation and generate additional codes, including CPT and E/M codes.
Suki, which recently refreshed its brand to highlight its various offerings, provides an AI-enabled ambient clinical intelligence tool that generates notes, allows dictation, recommends ICD-10 and HCC codes and answers clinicians' questions. The offering can be used with EMRs to copy clinical notes into the EHR.
The company is building on its platform's existing ability to generate ICD-10 and HCC codes from ambient listening to automate the detection of additional codes, including CPT and E/M codes.
"This sounds a little weird to say, in some ways, the primary goal is not to get ICD-10 or any sort of risk adjustment model accurate. The primary intention I have is focusing on the output of the note–the words, the ontology, the diction, and that is really hard," Dr. Kevin Wong, chief medical officer of Suki, told MobiHealthNews.
"Can we get the note so clinically accurate that it then stays ahead of, upstream of, other kinds of revenue cycle coding-based accuracy enhancements?"
Wong says the platform can pick up on accents, speech nuances, determine who said what and provide other quality-based enhancements.
"Part of what we do is we have an intent extractor. When we layer on speech and the language model, can we also determine the intent? And we fine-tune that with a lot of scoring," Wong said.
"How can we automate intelligently by extracting the intent, then outputting the quality of the note to capture that?"
He said Suki has evolved from scoring a note across many domains that had clinical significance to realizing it is not the whole note that needs to be scored, but rather different sections of a note.
"It's the history of present illness, it's the physical exam, it's plan, and then we further fine-tuned this and said maybe it is not just the sections, but maybe the sentences, which sounds crazy to think about looking at the sentences, but that is the only way that you can score and train the difference between what you're saying," Wong said.
Suki's platform identifies how a word itself can match an ICD code with more than 90% accuracy in some situations, Wong says. Still, the accuracy of E/M code detection is different.
"With the launch of this, we're going to do a lot more testing, but it has passed our internal testing evals to get that high," Wong said. "Now, E/M, the number is a little bit different. Why? There is an inherent difference in the medical decision-making of that exam."
Wong says the way the healthcare system is structured is that if a doctor sees a patient for longer than a certain number of minutes, or depending on the severity of the visit, different codes are used.
"There's an inherent amount of, kind of, gray or variability. And so the numbers for E/M are a little bit lower, but the point behind why we are excited about this is we have accuracy that at least is above a human evaluator," Wong said.
Not only can the machine already detect better than a human coder, Wong said, it has a high pass rate among providers who review the codes and determine their accuracy.
"What we are starting to explore is what does this mean when there's an audit, a RADV audit, or other types of audit? Those are things we are monitoring because, as we all know, even a physician today, a provider today, that drops all these codes, it is unbeknownst to them that many of them get denied behind the scenes or get queried," Wong said.
"Our belief at Suki is that our notes will be so accurate, it will reduce the need for so much to be scrubbed and audited after the fact. That is for us to prove and to show after this launch, and I very much think we will get there because it has passed our internal tests."
