Becker's Health IT: Turn ambient into action at the bedside

The following is an excerpt from a feature with Andrew Napier, MD, board-certified emergency physician and co-founder of Sayvant, in Becker's Health IT. You can find the full post on the Becker's Hospital Review website here.

‍

The emergency department rewards clear thinking under pressure. The mistakes I fear don’t come from ignorance. They come when attention splits at the wrong second. A result posts, a phone rings, a code announcement overhead, a family wants an update, and then suddenly the moment to act passes.

‍

Ambient AI promised relief from charting overhead to focus more on patient care, but there’s been limited progress bringing clinical decision support into the 21st century. Most ambient solutions are only focused on generating outputs, and traditional CDS systems rely on brittle algorithms based on EMR data, making it difficult to access real-time assistance while treating patients.

What shows up in the room: the mental checklist

Imagine: A man in his forties arrives with pressure in his chest that reaches into his jaw. Every ER doc is trained to run through the same mental checklist. Moderate concern for ischemia. Consider acute coronary syndrome, pulmonary embolism, and aortic disease. Get an ECG now and high-sensitivity troponins with HEART to guide the plan. If numbers break bad, escalate and admit. If HEART is low with two negative troponins, discharge with real return precautions. Don’t forget to compare arm pressures and pulses and ask about stimulant use.

‍

That’s how we’re trained – to distill thousands of pages of research and guidelines into simple heuristics that minimize risk, improve patient outcomes, and reduce medical malpractice risk.

Clinical AI: Assembling actionable heuristics

What actually helps clinicians on shift? We don’t need a giant rule book or AI-driven diagnosis that displaces our clinical decision-making. Imagine a clinical AI system that mirrors how we as EM doctors process differential diagnoses: trigger words based on chief complaint, results and risk scores that dismiss landmines, and recommended courses of action to address remaining risk.

‍

Chest pain wakes up on rest pain with radiation, diaphoresis, presyncope, or concerning history. It shuts off acute coronary syndrome only when three facts hold at once: a normal ECG, two negative high-sensitivity troponins, and a low HEART score. It shuts off pulmonary embolism when Wells is low with PERC satisfied or a high sensitivity D-dimer is negative. It quiets aortic dissection if there is no pulse or pressure gap, no high risk features, or a CTA is already negative. If none of those suppressors are present, the AI system names the three landmine diagnoses and pushes the next action that moves care, an ECG now and the troponin pathway, Wells then D-dimer in the right patients, a pulse and pressure check before you forget it.

‍

Thunderclap headache wakes up on sudden onset at maximal intensity. It shuts off subarachnoid hemorrhage only after an early non-contrast CT is negative and the local pathway that follows a negative early CT has been addressed. It watches for cerebral venous thrombosis in postpartum patients and those on estrogen, and asks for venous imaging when the story fits and the initial workup is clean. If the story still reads like blood after the early CT, it points to lumbar puncture or CTA per site policy and makes sure blood pressure, pain control, and clear return instructions are not forgotten.

‍

This is how AI can add value in our clinical workflow: by mirroring our mental heuristics in clearly explainable ways, and assessing the patient’s full story dozens of times throughout the encounter. By constantly analyzing the spoken history, exam, vitals, and orders placed, clinical AI can help keep your judgement for every case.

‍

This pattern works because it respects how clinicians think. It shows up on time, speaks once, and leaves proof. Ambient documentation becomes more than a cleaner note. It becomes a light touch form of decision support that protects judgment and improves care.