The emergency department is the front door of the hospital, and it's not always pretty. EDs are seldom quiet (there, I said the q-word). Physicians, nurses, and techs move patient to patient. Admitted patients board in hallways for hours, hooked up to monitors streaming endless data about the patient no one can use.

I've written plenty about the bottlenecks behind that scene, such as limited inpatient beds, too few scanners, thin staffing. But there's another bottleneck I’ve been thinking about that doesn't present itself on any capacity dashboard. It's the time clinicians spend assembling a coherent story out of what the ED has already collected (e.g., monitor stream, labs, imaging, notes from prior visits). That gap, between when the department has the information and when someone can act on it, is decision latency. It's a throughput constraint, but it’s just difficult to measure it like one.

In this article, I'll break down how decision latency produces ED delays, how Capacity Health is approaching it with its AI Clinical Action Engine for Acute Care, and what it could mean for patients, clinicians, and health systems.

In Why Is the ED So Crowded?, I walked through the boarding problem: admitted patients get stuck in the ED because no inpatient bed is available. Once hospital census passes 85%, patients start stacking up. Above 90%, the department is in trouble.

This is a bed problem, and it's the one most operational dashboards are built to track (beds, wait times, left-without-being-seen rates, door-to-doc).

What those dashboards don't capture, though, is how much clinical time gets spent turning scattered data into a decision.

Consider what a single ED visit generates, and where each piece actually lives.

All of it exists but little-to-none of it arrives assembled. Finding all of this information and consolidating it yourself takes time and delays medical decisions.

I recently wrote about rebuilding the patient's story on consult service. Even when I used AI, I was still choosing which labs, imaging, and notes went into the prompt. In the ED, that reconstruction is not a one-time task: the patient changes, new results arrive, and another team joins the case.

I previously mapped out the process of a patient moving through the emergency department, and you can see below where I starred (red) the area that’s eating up time.

Here's what that costs us, in one concrete example. A study in Annals of Emergency Medicine (2026), found that among 65,000+ monitored ED patients, 3% went home with atrial fibrillation nobody diagnosed (71% of them anticoagulation candidates), with triple the adjusted stroke risk, and only 19% picked up within two years. Those patients skewed toward Medicaid, no primary care physician, and Black or Hispanic/Latino.

The monitor was on and the rhythm was recorded, but nobody assembled it into something a clinician could act on before the patient walked out the door. While this was a single-center, retrospective study that didn't test an intervention, it describes the gap between collecting data and recognizing a finding that may matter for care.

CMS, for its part, is about to start measuring some of this. A new quality measure called Emergency Care Access and Timeliness (ECAT) counts the share of ED visits that hit at least one of four access and timeliness gaps: more than an hour from arrival to a treatment space, boarding longer than four hours, a total length of stay over eight hours, or the patient leaving without being seen. Voluntary reporting starts in 2027, and it eventually carries money with it. More on that later.

ECAT won't measure decision latency directly, but it will put a number on the delays it produces. That turns a clinical annoyance into something a hospital has to report on, and it's the gap the company I'm writing about today was built around.

In my piece on AI-generated ED histories, I explored what changes when a clinician starts with organized context rather than a blank page. Capacity Health asks a broader question: how can that picture stay useful as the encounter evolves and inform what happens next, not just for one patient, but for every patient simultaneously contending for resources and attention?

Capacity Health builds software that pulls in what an ED patient is already generating (monitor waveforms, interventions and responses, lab results, imaging, notes from prior visits) and continuously assembles it into one evolving clinical picture of that patient, along with the evidence behind each finding, possible next steps, and a named owner for each one. They call it the Clinical Action Engine for Acute Care, and position it as a new class of AI for the ED.

Their read on the market is that healthcare AI's first wave helped us document, retrieve, and summarize. We had ambient scribes, search tools, risk models, alerts, and operational dashboards each handle important parts of that work. Some tools already assemble patient information into a clinical view. Capacity Health's focus is connecting that view to supporting evidence and clearly owned next actions for clinician review, then using the same patient model to inform hospital and system operations. Capacity Health's argument is that clinical decision support and capacity management should not be disconnected, because they're describing the same patient from two different chairs.

The architecture is one patient model supporting three levels of action. The ordering is key, because the system is built from the patient up rather than starting with the operational view and drilling down.

As I wrote about trust at the point of care, 'the clinician stays in control' has to be practical. I would want to see the findings, supporting evidence, and uncertainties quickly enough to check a possible next step before acting on it.

For anyone who has sat through a health system procurement review: Capacity Health reports that the platform is HIPAA-compliant, SOC 2 Type 2 audited with an unqualified opinion in 2026, and independently penetration tested.

Capacity Health was founded by two emergency physicians who also happen to be fellow Huddlers. William Dixon, MD, MSEd, is an emergency physician and Stanford clinical assistant professor who previously co-founded Signos. David Kim, MD, PhD, is an emergency physician and NIH-funded medical AI researcher whose lab pioneered using continuous physiologic signals for clinical prediction in acute and post-acute care.

This explains why the research in this piece traces back to Stanford. Their research explores the gap between information an ED already collects and the clinical and operational decisions it can inform:

These are the founders’ own research papers, which help explain the research foundation behind Capacity Health's approach. In an early-stage company, this is an incredible strength.

The hardest problem with prediction tools is delivery: can a prediction reach a clinician or operations team in a form they'll actually act on, at a moment when acting is still possible? This is what Capacity Health is betting on. And when they succeed, it’ll have a profound impact on key stakeholders.

For patients, Capacity Health could help deliver the right care at the right time in the right place by cutting the time clinicians spend hunting for fragmented data, stitching it into a coherent picture, and then deciding what to do next. The aim is to make relevant findings and possible next steps easier to review while there is still time to act.

For clinicians like me, the value is closer to home. A “busy shift” is carrying twelve incomplete patient stories in your head at once, worried the thing you’re missing is the thing that matters most. A tool that reduces what you have to track—by consolidating information and flagging the sickest patients and what each needs—has clinically significant value.

For health systems, the opportunity is: better flow, shorter boarding times, lower length of stay, forecasted capacity, and load balancing across facilities. Boarding is already expensive. Patients who leave without being seen are revenue that walked out, diversion sends volume to a competitor, and an ED that can't move admitted patients upstairs can't take new ones. The economic opportunity is to make better use of existing staffed capacity. ECAT, which I mentioned earlier, adds pressure on top of this. Hospitals that fail the reporting requirements face up to a 2 percentage-point reduction in their Medicare outpatient payment update, beginning with the 2030 payment determination. ECAT adds accountability for access and timeliness, without telling hospitals which intervention will work.

Here's what I'll be watching from Capacity Health as they grow:

I've argued that making a broken process faster is not the same as changing it. Capacity Health's more interesting bet is that the same patient picture can inform both clinical decisions and the work surrounding them. The test is whether that changes how care gets coordinated, not simply how quickly we can describe a patient who is still waiting.

In summary, the ED's capacity problem is usually described in physical terms, but a meaningful share of it is decision latency: the time between when the department has the information and when someone can act on it. Capacity Health is betting that one patient-level model can shorten that gap at the bedside while giving hospital and system teams a shared basis for action. The research is rigorous and the framing matches what I see day to day in the hospital.

About this post: This is a partnered deep dive with Capacity Health. The framework and conclusions, however, are my own. As a reminder, I only partner with companies working on problems I’d write about anyway, and Capacity Health’s focus on reducing decision latency by turning fragmented acute-care data into a usable clinical picture fits that bar. As always, my goal is to give you a transparent breakdown of what works, what still needs scrutiny, and why it matters for patients, physicians, and the health system. If you or your company would be interested in a partnership like this, click here.