Try it: one case
You are the doctor. This is one short, fictional teaching case in the shape of the full simulator: ask, examine, order, treat, then commit. Nothing here is a real patient. Leaving this page discards the case in progress.
ED encounter · Chest pain
Marcus, 58, arrived ten minutes ago complaining of chest pain.
You are scored on: the cues you find, ordering the ECG, your diagnosis, your disposition, and the treatment you start. Diagnosis and disposition credit needs a minimum workup: the ECG, plus at least two questions or the exam. Session limit: 5 minutes. When the limit arrives, you will be told to commit.
Step 1 · Ask and examine
Step 2 · Orders
Step 3 · Treatment
Pick any that you would start now:
Step 4 · Commit
Pick a diagnosis and a disposition to continue. Commit locks both picks; nothing is revealed before you commit.
A short scripted sample, built from the simulator’s own case material. It is not the full simulator: the full simulator runs longer cases with an AI patient, on its own backend. Open the live simulator. The Case Studio is the separate faculty authoring tool; the simulator is the learner product. ECG tracing: a real recording from the PTB-XL database (PhysioNet), record 08075, licensed CC BY 4.0. Leaving this page discards the case in progress.
Clinical reasoning is the one skill medical school never directly tests. Students learn it by lottery, from whichever patients happen to arrive on their shifts. SimED lets a student run a full emergency department encounter, a patient who talks back and deteriorates on a clock, and then grades the reasoning itself.
Before medical school, I spent five years as a charge EMT, leading ambulance crews on 911 calls. The skill that mattered most was never a fact. It was the sequence: what to ask first, what to order before the result in front of you comes back, what to do when the patient gets worse while you are still thinking. Medical school teaches that sequence by lottery, from whichever patients arrive on your shifts. SimED is the simulator I built with Dr. Michael Phipps and Chy Murali so a student can run a complete emergency department encounter, and be scored on the reasoning itself, before the first patient is real.
The setup
Training today tests knowledge with exams and procedures with checklists, and leaves reasoning to chance. The pitch for SimED opens at 2 a.m. with a chest pain patient and a student who has never had to run the encounter end to end.
Presenting problem
Existing products stop short: text cases, 3D physiology, authored cases with static findings. None runs the whole encounter, with a voice, a clock, deterioration, and a disposition, and none returns a reasoning score a curriculum committee can use.
What I built
The full encounter in a browser: triage and ESI, a voice history taken from authored patient facts, 170+ real exam findings with media, orders at real CMS prices, a ranked differential, a SOAP note, and a deterministic, evidence-cited debrief scored against a faculty-authored 100-point rubric.
Next to the simulator sits the Case Studio, where faculty author cases as single files carrying the ECG, labs, and media, and write the 100-point rubrics that grade them.
Workup
Two design decisions carry the product. First, clinical truth is code: a deterministic engine runs the physiology, the clock, and the scoring. Right treatment moves the blood pressure. Waiting lets the septic patient deteriorate on schedule. Labs return on realistic turnaround timers.
Second, the AI patient is fenced in: it can only say what the case author wrote, every case carries a clinician reviewer, and no interpretation is shown before the learner commits their own. Real media only, never AI-generated exam media, and cost feedback stays visible in the debrief, with costs from one cited source.
The interface was redesigned away from a step-by-step quiz on purpose, so the student feels like a real doctor working a case, not a student answering questions in order.
Discharge summary
A working, live simulator with a case library across the presentations that fill an ED: chest pain, dyspnea, abdominal pain, altered mental status, syncope, and sepsis. Abdominal pain was chosen as the first rollout presentation.
The ICTR ATIP application to UMB ICTR, with Dr. Phipps as PI, seeks funding for the 30-case library and the validation study: an IRB-approved study of n=60 across medical students, residents, and attending physicians, two simulation cases four weeks apart, scored on diagnostic accuracy, differential completeness, and ordering efficiency with standardized rubrics. Construct validity is tested by performance differences across training levels, with test-retest reliability and usability (SUS) as secondary outcomes. The study has not run yet, so every result for it is still ahead of us.
SimED is also a preparation tool for Step 3, framed that way by Aaron. That framing is the project's own positioning, not an endorsement by UMB or ICTR.
What I’d do differently
Start the validation design before the feature list: the study's rubric demands shaped the debrief more than any user request did. And keep cases shorter: the full encounter is the point, but the first case a student finishes teaches more than the perfect case they abandon.