In partnership with

Become An AI Expert In Just 5 Minutes

If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.

This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.

Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.

AI Is About to Force Healthcare to Confront Patient Harm at Scale

And the regulators are already moving in the same direction.

I’m calling this now. One of AI's most valuable use cases in healthcare will be exposing how often patients are harmed in the settings we trust most. And that will be the forcing function for the kind of systemic change that’s desperately needed.

We talk constantly about AI lightening the documentation load. Absolutely! It’s important, low-hanging fruit. Burnout is real. The clerical burden is crushing. Healthcare desperately needs workflow (and often care) redesign. But the most transformative use of AI in healthcare may be something less comfortable to talk about. That’s AI identifying patients who were harmed by the care they did or did not receive.

That was my biggest takeaway from the KFF Business of Health podcast episode featuring Dr. David Bates, "Can AI Break the Measurement Paradigm?" The conversation named something healthcare still struggles to confront and prioritize in a standardized way.

We talk about "quality" constantly.

But much of what we measure is process metrics, compliance activity, documentation requirements, and lagging indicators. The actual patient experience inside hospitals can look very different. The research is blunt: roughly 1 in 4 hospitalized patients experiences harm.

Twenty-five percent.

25%!

If one in four patients are harmed during a hospitalization, improving patient safety needs to be the operational challenge.

And most of that harm is still invisible.

What a High Reliability Organization (HRO) journey actually teaches you

Earlier in my career, I was fortunate to be part of a hospital system on an HRO journey while we were simultaneously operating in value-based care arrangements. One of the first things you learn on a real HRO journey is deeply counterintuitive:

Patient safety events initially go up. That’s a reason to keep going on the journey. Why does it go up? Because people finally feel safe enough to report what was already happening. That is one of the hardest truths in healthcare. Harm is often all around us, but it’s underreported because people fear retribution, blame, embarrassment, litigation, peer judgment, leadership fallout, or being labeled "the problem."

In weak “safety cultures,” silence becomes normalized and sometimes even expected. “Near misses” disappear. Patterns stay hidden. Small failures compound. And systems never learn. True HRO transformation requires the opposite:

• Daily reporting

• Constant surfacing of near misses

• Psychological safety

• Structured root cause analysis

• Relentless transparency

• A shared understanding that reporting is not betrayal. It is protection.

That is why we call it a journey. It takes years.

Years to build trust.

Years to normalize transparency.

Years to redesign workflows.

Years to find the structural weaknesses buried deep inside operational processes.

But when organizations stay committed long enough, the root causes start to surface. And many of them are structural, not individual. They include things like:

• Communication failures

• Unsafe handoffs

• Poor workflow design

• Medication management gaps

• Fragmented transitions

• Data buried in disconnected systems

• Staffing strain

• Alert fatigue

• No real-time visibility into deterioration

• Excessive top-down hierarchy

Healthcare treats safety events as isolated human mistakes when so many are predictable system failures. That is exactly where AI becomes transformational.

Not faster notes. Visible harm.

AI may finally let health systems identify patterns of harm at scale and in real time.

Right now, organizations sit on enormous amounts of data trapped inside the EHR. For years, most of it has been operationally unusable.

Structured fields. Unstructured notes. Radiology reports. Medication histories. Nursing documentation. Discharge summaries. Lab trends. Consult notes. Follow-up recommendations.

We digitized healthcare and, in many ways, simply turned paper chaos into digital chaos.

The EHR era built the foundation. AI may finally operationalize it.

Here is the connection I keep coming back to.

Remember the counterintuitive truth from the HRO journey? Events only surface when people feel safe enough to report them. AI offers a second path to that same visibility. It doesn't wait for a clinician to feel safe enough to file a report. It reads the record itself.

It can ingest fragmented information and detect subtle patterns humans cannot consistently catch across thousands or millions of encounters. This includes things like medication-related issues, missed follow-up, delayed interventions, early deterioration, diagnostic drift, unsafe transitions, and the compounding low-level failures that eventually become catastrophic outcomes.

And it can do this continuously.

Healthcare has historically reviewed safety retrospectively. When it’s too late to prevent a serious safety event from happening in the first place. After discharge. After readmission. After litigation. After mortality review. After the preventable harm already occurred.

AI creates the possibility of catching those signals while there is still time to intervene.

CMS is already moving

Here is the part most people missed in the noise of the 2026 OPPS final rule.

For years, the Overall Hospital Quality Star Rating let strong performance in one area paper over weakness in another. In terms of patient safety, that meant a hospital could land in the lowest-performing quartile on the Safety of Care measure, which means they are in the bottom 25% on healthcare-associated infections and patient harm, and still earn an overall rating of five stars. Carried by good numbers everywhere else, it was easy for a consumer to overlook the low rating in Safety of Care.

CMS called out the problem and closed the gap. In the CY 2026 final rule, the agency finalized a two-stage change:

• 2026 ratings: any hospital in the lowest Safety of Care quartile (with at least three safety measures) is capped at four stars regardless of how strong the hospital looks everywhere else.

• 2027 and beyond: that cap becomes a blanket one-star reduction. Bottom quartile on safety, and your overall rating drops a full star, with one star as the floor.

CMS estimates the second stage alone will pull down roughly 459 hospitals. THAT MANY HOSPITALS score in the bottom quartile of safety! Read the language the agency used and the intent is hard to miss. CMS tied the change directly to its "commitment to improving health outcomes and advancing patient safety."

This also represents a philosophical shift in what "quality" is allowed to mean.

And I don't think the timing is an accident.

Look at what's converging. The research says one in four patients is harmed. The tools to see that harm continuously, at scale, and in real time, are arriving right now. And CMS just built a rule that turns safety into the thing that caps your reputation.

This reads like a rule written in anticipation of what is about to become much easier to measure.

For most of the EHR era, harm was hard to surface and easy to externalize. When AI makes harm patterns visible across millions of encounters, "we didn't know" stops being a defense. I read this as CMS effectively putting hospitals on notice. The direction is set. The incentive is public and reputational.

No health system will be able to say it wasn't told to look hard at its own safety practices.

The economics finally line up

Patient safety journeys are, in many ways, de facto value-based care exercises. Hear me out.

There is no billable CPT code for becoming a High Reliability Organization. No fee-for-service reward for reducing near misses. No reimbursement line for building psychological safety across clinical teams.

Historically, healthcare economics did not align cleanly with harm reduction. Preventable medical errors cost the system tens of billions of dollars a year — but many of those costs were shifted outward, to payers, employers, patients, families, and society. That weakened the direct financial incentive for hospitals to aggressively invest in safety.

Value-based care flips that.

In risk-bearing arrangements, preventable harm destroys margin. Unsafe care raises total cost of care. Complications drive utilization. Readmissions rise. Length of stay expands. Downstream costs compound.

The organizations furthest along in patient safety and HRO principles are already positioning themselves better for the future of value-based reimbursement.

Safer care is lower-cost care. But more importantly, safer care saves lives and reduces suffering.

Now stack the three forces together. AI is making harm visible. CMS is making safety the gate on your public rating. Value-based care is making harm a direct hit to your margin. The moral case, the regulatory case, and the financial case for patient safety are pointing the same direction.

What this means right now

For decades, organizations leaned on manual reporting to find safety risk. HRO work is an investment in time and money, and requires an all-of-health-system-approach. But humans still miss things. They get overwhelmed, they normalize dysfunction, and adapt to unsafe systems because they have no choice.

AI does not replace culture, leadership, or accountability when applied to patient safety. However, it should dramatically increase visibility into hidden risk of harm.

This is the moment for owners, operators, executives, and policymakers to lean in. Early movers may redesign care delivery around safety intelligence itself:

• Real-time harm surveillance

• Predictive deterioration monitoring

• Closed-loop follow-up systems

• AI-assisted medication management

• Continuous risk detection

• Dynamic safety workflows

This calls for a celebration! This is where healthcare is headed, and the organizations that embrace it earliest may become the highest performers in both quality and financial sustainability as the market increasingly rewards better outcomes at lower total cost. And if the prevention of harm is not one of the outcomes we care about most, then what are we even doing?

We should want hospitals and inpatient settings to be obsessed with patient safety.

Especially for the frail, the medically complex, and the vulnerable, who suffer some of the worst consequences when systems fail.

One of the most important eras of AI in healthcare will be exposing where care delivery is failing patients. CMS is already writing the rules for the world that exposure creates. After that comes the transformation many of us have hoped for all along.

If you work in a hospital, here's how to bring this up now

Maybe you're a bedside nurse. A hospitalist. A quality director. A service-line leader. An executive.

Wherever you sit, you can move this conversation today. You don't need a budget or a mandate to start.

• Look up your own numbers. Pull your hospital's current Overall Star Rating and, specifically, its Safety of Care performance on Care Compare. Know whether you're near the bottom quartile before CMS tells the public in 2027.

• Name the stakes in the room's language. With leadership, frame safety as what it now is: the gate on your public rating and a direct hit to margin under risk. It’s a reputational and financial project,

• Report the thing you've been sitting on. The near miss. The workaround. The handoff that almost went wrong. Surfacing it is the raw material every improvement depends on. Model that.

• Ask the AI question differently. Your organization is already talking about AI for documentation and efficiency. Ask the harder one: what are we doing to surface harm with the data we already have?

• Find the process, not the person. When something goes wrong, push the conversation past "who" toward "what in the system made this likely." That single reframe is the start of every HRO journey.

You don't have to fix the culture this week. You just have to refuse to let the silence stay normal.

Start one conversation. Then another.

That is how every safety transformation actually begins.

Sources & further reading

• The podcast that started this. "Can AI Break the 'Measurement Paradigm?'" — Chip Kahn in conversation with Dr. David Bates, KFF's The Business of Health (AI series, Episode 7). kff.org/other-health/can-ai-break-the-measurement-paradigm

• Where the 1-in-4 number comes from. Bates DW, et al. "The Safety of Inpatient Health Care." New England Journal of Medicine, 2023;388(2):142–153. The study that found roughly a quarter of hospital admissions involved an adverse event. nejm.org/doi/full/10.1056/NEJMsa2206117

• The CMS rule, in CMS's own words. "CMS Empowers Patients and Boosts Transparency by Modernizing Hospital Payments" — the agency's announcement of the CY 2026 OPPS final rule, including the Star Rating change. cms.gov/newsroom/press-releases/cms-empowers-patients-boosts-transparency-modernizing-hospital-payments

• The technical details of the two-stage change. CMS fact sheet on the CY 2026 OPPS/ASC rule, laying out the four-star cap (2026) and the one-star reduction (2027 and beyond). cms.gov/newsroom/fact-sheets/calendar-year-2026-hospital-outpatient-prospective-payment-system-opps-ambulatory-surgical-center-0

• Where the idea of a High Reliability Organization comes from. The concept was coined by a research group at UC Berkeley — Todd LaPorte, Gene Rochlin, and Karlene Roberts — who studied error-free operations in aircraft carriers, air traffic control, and nuclear power starting in 1984. Foundational articulation: Roberts KH. "Some Characteristics of One Type of High Reliability Organization." Organization Science, 1990;1(2):160–176. A plain-language overview is here: AHRQ PSNet, "High Reliability." psnet.ahrq.gov/primer/high-reliability

• HRO translated for hospitals. Chassin MR, Loeb JM. "High-Reliability Health Care: Getting There from Here." The Milbank Quarterly, 2013;91(3):459–490. The paper that brought HRO thinking — leadership commitment to zero harm, a true safety culture, robust process improvement — into mainstream hospital practice. pmc.ncbi.nlm.nih.gov/articles/PMC3790522

• How many hospitals we're talking about. American Hospital Association, Fast Facts on U.S. Hospitals, 2026: 6,093 hospitals in the U.S., of which 5,121 are community hospitals (the acute-care group most affected by the Star Rating change). Useful for putting the ~459 hospitals hit by Stage 2 in context. aha.org/statistics/fast-facts-us-hospitals

• What the Safety of Care measures actually are. The Star Rating's Safety of Care group is built mostly from outcome measures: the CDC/NHSN healthcare-associated infections (CLABSI, CAUTI, surgical site infection, MRSA bacteremia, and C. diff), the AHRQ PSI-90 patient-safety composite, and hip/knee complication rates. CMS overview of the measure groups: data.cms.gov/provider-data/topics/hospitals/overall-hospital-quality-star-rating · Measure definitions: cms.gov/medicare/quality/value-based-programs/hospital-acquired-conditions

Read more from me about the HRO journey in the context of hospital mobility here:

*Disclaimer: All opinions and ideas expressed in this article are solely mine and none represent a recommendation or should be viewed as advisement of any kind to anyone to do anything.*