Sponsored

How point-of-care AI can protect case mix index (CMI): 5 takeaways

Advertisement

Hospitals lose case mix index, quality scores and revenue to a problem that rarely shows on a dashboard: diagnoses that sit in the chart but never reach the physician’s note. 

Regard is a New York-based clinical AI company whose platform analyzes chart data to recommend diagnoses and draft documentation. David Kirk, MD, a 25-year ICU physician and former hospital executive at WakeMed., said he joined after its AI recommended a diagnosis in his own ICU that he would otherwise have missed.

In a session sponsored by Regard at Becker’s 14th Annual CEO + CFO Roundtable on Nov. 3, the panelists discussed why documentation fails, what point-of-care AI changes, and why human judgment must stay in the loop. The panelists were Dr. Kirk; Faisal Chawla, MD, medical director of clinical documentation integrity and physician advisor for utilization management at Cedars-Sinai in Los Angeles; and Gurmeet Sran, MD, a hospitalist at Sutter Health in Berkeley, Calif.

Here are five takeaways from their conversation.

1. Documentation is a diagnosis problem before it is a revenue problem.

Dr. Kirk was once tasked with fixing a  cardiovascular service line that had excellent care, but  poor revenue and quality scores. The reason boiled down to documentation, and he explained,  “If you have the right diagnosis, you can have the right care plan.” Denials become easier to defend because the diagnosis is supported throughout the stay rather than reconstructed afterward, explaining, “Ultimately what AI should do is do the bulk of the clerical work for the docs, make the diagnoses, make the supporting evidence, where the docs can spend their time reviewing and getting back to the bedside.”

2. Start with the problem, and measure quality alongside CMI.

“One of the most important things when you’re considering a tool is to consider what the problem is you’re trying to solve,” Dr. Chawla said. Cedars-Sinai set quality metrics as KPIs alongside CMI and matched its analysis within DRG families, for example, pneumonia against pneumonia and heart failure against heart failure. He said expected mortality and expected length of stay both improved, lifting the Vizient index measures the organization watches. 

Physicians accept about 50% of the AI’s suggestions, about 60% in recent months, which is proof that they have final say over what enters the note. Working with Regard, queries fell 6%, with Dr. Chawla clarifying that “the more we can get documentation right upstream, the less work there is downstream,” he said. Denials, the most lagging KPI, are still being measured.

3. The story is usually in the chart; it’s just buried.

Dr. Chawla said his team overturns 70% to 75% of peer-to-peer reviews, mostly by piecing together why the patient needed hospital care. 

“In 75% of the cases, the story’s there. It’s just buried in the chart,” he said. Clinical validation denials, he added, systematically target CCs and MCCs, so contemporaneous, concise documentation in one place is the defense.

4. Guard against clinical surrender.

Dr. Sran said reasoning models have improved so fast in the past 16 months that deskilling is a governance concern for attendings and trainees alike. “This is the whole art of the deskilling, or what they often may call the clinical surrender to AI,” he said. 

He has seen hallucinated antibiotics in discharge summaries and drew a line between non-deterministic large language model summaries and the deterministic logic Regard uses to identify disease states. 

“You work at the speed of trust,” he said. Dr. Kirk compared it to an EKG’s automated read: “Even if you have autopilot on your plane, you still have to continue to know how to manually land that plane.”

5. Ask vendors about data, platform and trust.

Dr. Sran offered three questions. First, what happens to the organization’s data: “The risk of what’s called data exfiltration, or that IP effectively being used down the road, is something which we always, especially as larger health systems, need to be very mindful about.” 

Second, whether the vendor is building a platform or another pilot. 

Third, he pointed to trust: “Do you trust the vendor that you’re talking to? Do you believe in the theory and the thesis? Are they well funded?”

What CFOs should weigh next

Dr. Chawla said protecting the record and the revenue belong near the top of every list given reimbursement headwinds. Under-documentation leaves money on the table, but integrity matters as much, and the CC/MCC capture rate “without a balancing measure like denials, I think it’s overrated.” 

Dr. Sran said agentic frameworks that handle asynchronous work without a human signing off on everything are where health systems will rethink operations next.

Advertisement

Next Up in HR

Advertisement