@HandSurgeonMD Agreed. Training and expertise has to take precedence over PARs and ordering patterns of the system. Difficult to show that retrospectively.
@MaieuticMD@EricTopol Fair. And it's fixable in the tools. If the model knew the patient was paying cash and saw the $4 list, the plan it suggests would actually get filled. You're building clinical AI. Are you feeding it coverage data?
@obienemuo Exactly. A single vital sign or lab is a snapshot. The trend is what tells you whether the patient at 2 a.m. is getting better or quietly getting worse.
@DrNeilStone The wait is the dangerous part. The ER patient who worries me most is the one in the waiting room at hour 9 whose heart rate has crept up since triage and nobody has rechecked. Are repeat vitals required during waits that long?
@DrDiGiorgio Most ER docs are working within metric worlds now, and you feel it in the dashboards: door-to-doc times, RVUs per hour, patients per shift. None of them count the 2 a.m. call to admit someone who looked fine on paper. Who should own that metric?
@EricTopol Matches residency. A resident with a good attending looks great on shift, then struggles alone on nights. The skill sticks only when they have to do it unassisted. Did the study test performance after the AI was removed?
@NEJM@myESMO From the ER side, the visit that matters comes weeks in: a patient on targeted therapy with new swelling or shortness of breath, and the question is disease versus drug. What toxicities should ER docs watch for with this combo?
@DrDiGiorgio Same pull in the ER chart: query messages asking if "AKI" or "acute respiratory failure" fits a creatinine bump or a sat of 91. The diagnosis drifts toward what pays. How do researchers adjust for that drift?
@chrissyfarr Same logic as walking the department instead of charting from a workstation. In the ER, I hear more on a lap past the hallway beds than I'd ever get from a screen. Do your calls run shorter than the video ones?
@kevinmd Small things do the most here. In the ER, an NPO order written at 3 a.m. for a scan can ride along for days after the scan is done. Someone has to ask whether it still applies. Who owns rechecking those orders?
@RVarmaMD@molsjames Fair. "Rule out acute process" tells radiology almost nothing. A line like "RLQ pain, fever, WBC 15" gets a better read. What indication detail helps you most?
@HandSurgeonMD Exactly. A one-line note that the requested implant wasn't available and what was used instead protects the surgeon and the patient. Did the hospital ever put the reason in writing?
@molsjames Mostly, yes. The gray zone is the obs patient: chest pain with a negative first troponin, or the elderly fall who can't walk safely yet. That middle lane is where trajectory matters most.
@simplydt Same rule as an ER shift: treat what's killing the patient first and park the interesting finding for follow-up. Building between shifts helps here, because there's no time for the cool feature anyway.
@VALW Brachial plexus cases often come down to the delivery note written minutes after the shoulder dystocia. From the ER side, the record made in the moment does most of the defending. Curious what the timeline showed in this one.
@lawheroezV2 Hospitals hit this too. The general model may read a chart fine, but if the vendor won't sign the privacy terms you need, the vertical tool wins on paperwork, not on the model. Curious whether firms are pricing that in yet.
@jakesendar Hospitals see the same thing: two vendors, same health system, same press release week. The question nobody asks is which one the people doing the work actually open at 4 p.m. on a Friday.
@mo__nyc Publishing your own misses is the part I trust. In the ER the scary scribe error isn't a typo, it's a med I never said sitting in a note I signed. That one's mine now. Would love to see the traps broken out by drug class.
@jasonlk From the ER side: triage is qualifying inbound, and most of it is easy. The miss that hurts is the quiet 70-year-old with "just indigestion" who looks low priority. How does the agent flag the one it isn't sure about?