When CMS Asks AI to Read the Chart, What Does It Find?
Here is a question every CDI specialist should be able to answer this week. If an AI tool read your last 20 charts tomorrow, what would it flag?
That is no longer a hypothetical. CMS has narrowed more than 300 industry submissions to eight finalists for QualTech, an event run through its Center for Clinical Standards and Quality. The finalists present at CMS headquarters in Baltimore on November 9.
What CMS actually asked for
QualTech named four opportunity areas:
AI-assisted quality and patient safety review
Annual wellness visit support
A next-generation digital quality measure calculator
A national quality hospital dashboard
Source: CMS Center for Clinical Standards and Quality, QualTech (https://www.cms.gov/cms-ccsq-qualtech-0)
The first one is the one that should get our attention. CMS calls it using "Super Intelligence," also called artificial intelligence, to protect patients and strengthen healthcare quality. The idea is AI-assisted medical record review that identifies emerging quality and patient safety risks and surfaces potential documentation gaps for closer clinical review. It would help surveyors before they arrive onsite and help providers fix issues before a survey ever begins.
The focus is the survey and certification side: hospitals, nursing homes, hospice and dialysis facilities. Still, read that again. Documentation gaps. Patient safety risks. In the medical record. That overlaps with our work.
What this does and does not mean
CMS states that QualTech is not associated with a funding opportunity or an RFP, and that being selected does not affect eligibility for future ones. So this is not a rule and it is not a mandate.
It is a signal. CMS also says it may pursue demonstrations, pilots and collaboration with select organizations after the event. The agency that sets the rules for quality reporting and payment is actively looking for technology that reads medical records at scale and surfaces gaps before a survey does.
The part AI cannot do for you
An AI tool can flag a missing diagnosis, an unclear timeline or a conflict between two notes. What it cannot do is decide whether the clinical picture supports the diagnosis, or whether a flagged event is a true safety event or a documentation and coding issue.
That judgment still belongs to a person who understands the clinical story. Take a PSI flag. This is my example, not CMS's. A tool can tell you the flag fired. It takes a trained reviewer to tell you whether the patient was actually harmed or whether the record just failed to say what happened.
If anything, a faster flagging tool raises the bar. More flags will surface sooner. The teams that can sort them with confidence will be ahead. The teams that cannot will spend their time defending records instead of improving them.
What to do now
Pull 10 recent charts that carried a quality flag and sort each one: true quality event or documentation and coding issue.
Write down why for each one. If you cannot explain it in two sentences, that is your gap.
Check your queries. Would each one hold up if a reviewer, human or AI, read it cold?
Ask your team who owns the call when a flag is ambiguous. Someone should.
AI will help us see more of the chart. Knowing what we are looking at is still the skill.
Where to go from here
If your team gets a PSI flag and cannot say with confidence whether it is a true quality event or a documentation and coding issue, that is the gap PSI Without Panic is built to close.
Learn more at https://www.thequerycorner.com/psi-without-panic
Comments