The big case, cardiogenic shock after anterior wall infarction
Everything from modules 1–5 comes together here: one realistic case adapted for teaching, five LLM work assignments and ten multiple-choice questions. You'll summarise, spar on the differential diagnosis, fact-check, hold a treatment plan against the guideline and iterate your prompts, exactly as you'll do it in practice.
Practical: allow 60–90 minutes. Work in ChatGPT or Claude (or both, the differences are instructive). Read the case first, then do the assignments in order.
Before you start, about this case and privacy
Goal: know why you may paste this case into a public chat, and a real one not.
This case has been adapted for teaching
The case below has been edited and pseudonymised for teaching purposes (v3): dates shifted, values adjusted, centres anonymised and identifiable details removed or changed. That's why you may paste it into a public LLM in this module. Keep the core question of module 5 in mind: would this be acceptable with a real case? That's what assignment 5, and the first multiple-choice question below, are about.
Question 1, privacy first
A resident pastes a real, unedited version of this case into a public LLM chatbot to generate a discharge letter. What is the most important objection?
The case (adapted for teaching, v3)
Read the case through once. In the assignments you'll paste it (in whole or in part) into the LLM, copy it to a scratchpad now.
Reason for admission
Cardiogenic shock due to an anterior wall infarction sustained out of hospital.
Relevant history
- Smoking
- Substance abuse with GHB, speed, MDMA preceding the infarction
Laboratory 27-04-26, 08:15
Haemoglobin: 5.3 (L) · Platelets: 402 (H) · Leukocytes: 14.2 (H) · #Neutrophils: 11.9 (H) · Sodium: 150 (H) · Potassium: 3.9 · Glucose: 11.8 (H) · Creatinine: 118 (H) · eGFR (CKD-EPI): 57 (L)
Venous blood gas 27-04-26, 10:53
Hb 5.5 · Temp 36.9 · pH 7.50 · pCO2 4.5 · Bicarbonate 25.4 · BE 2.6 · pO2 5.8 (L) · O2-Hb 79.6 · CO-Hb 1.1 · Met-Hb 0.4 · Reduced Hb 18.8 · Sodium 142 · Potassium 7.1 (H) · Chloride 110 (H) · Ionised Ca 0.98 (L) · Glucose 13.2 (H) · Lactate 2.1
Admission summary, CCU course
7-4-2026: Presentation with an anterior wall infarction sustained out of hospital (>12h old) after drug use (speed, MDMA, GHB), with autonomic symptoms and vomiting ±25 times. Initially conservative management, but with very severe pain, hypotension and AV block still emergency PCI. Temporary pacing wire placed. Afterwards a picture of shock, probably combined hypovolaemic (vomiting, sweating) and cardiogenic with a large infarction and moderate ventricular function. Norepinephrine + milrinone + high-dose furosemide started, minimal diuresis. ICU consulted for CVVH. Impella placed in the right femoral artery, running 3.3 L/min at P9. Cold right leg after Impella placement; transfer to ICU for VA-ECMO and CVVHD.
Investigations
- Lifenet ECG: Sinus rhythm with PVCs or aberrant PACs in bigeminy. Marked first-degree AV block, narrow QRS, normal axis, anterior ST elevation (V1–V5) with Q-wave formation already present and terminally negative T waves, plus inferior ST depression.
- ECG at presentation: ST elevation decreased, still some ST elevation in V4–V5.
- Quick-look TTE 7-4-2026 (cardiologist A): Moderate LV function with anteroseptal akinesia, preserved wall thickness consistent with viable myocardium. RV function visually normal. No pericardial effusion. IVC collapsing.
- Quick-look TTE 8-4-2026 (cardiologist B): Poor LV function, akinetic anterior wall/apex region, good RV function (TAPSE 16 mm). No MR. No free-wall rupture; pericardial effusion up to 1.8 cm near RA and RV (slightly increased), no significant inflow variation. Structure in LV apex, presumably trabecula, beware thrombus.
- Quick-look TTE 10-4-2026 (cardiologist C): Poor LV function, LVEF 20–25%. Wall motion abnormalities anteroseptal and apical, inferior wall still good. Reasonable RV function. Moderate-to-severe MR (new!). No pericardial effusion.
ICU course, Centre A
- 8-4: ICU takeover for progressive cardiogenic shock despite maximal inotropy. Impella in situ, cold right leg after placement. Uncomplicated VA-ECMO placement with perfusion catheter right leg, followed by improved circulation. ECMO blood flow 3.4 L, Impella 2.4 L.
- 9-4: Falling MAP and flattened Impella curves. Echo: tamponade, for which pericardiocentesis by interventional cardiologist on the ICU. Heparin temporarily stopped, restarted because of the Impella. Impella temporarily P2. Packed cells given per study protocol and Hb below transfusion threshold.
- 10-4: TTE: severe MR, no direct suspicion of chordal rupture, no pericardial effusion. Given the MR, Impella flow increased (P5), ECMO 2.4 L. Over the day 2x packed cells at Hb 5.4/5.7. Persistent blood loss along ECMO cannulas, for which anti-Xa targets temporarily lowered to 0.3–0.5.
- 11-4: Tube withdrawn with desaturations. Contact with Centre B (regarding possible future treatment); heparin stopped with ongoing cannula blood loss.
- 12-4: CVVHD filter expired; trial stop. Dobutamine started, followed by atrial flutter (flutter rate 55/min, ventricular rate 96/min). Impella removed uncomplicated at 14:30.
- 13-4: Successful decannulation, dapagliflozin started. Esketamine started, propofol and fentanyl weaning. Flucloxacillin started for fever and earlier S. aureus in SDD sputum culture.
- 14-4: Start 1 L water via feeding tube for hypernatraemia; acetazolamide for contraction alkalosis. Dobutamine weaning with own atrial flutter, HR 105–115/min.
- 15-4: Water via tube up to 2 L, hydrochlorothiazide started, furosemide stopped. CK values risen and plateaued, no signs of compartment syndrome.
- 16-4: Dobutamine and norepinephrine stopped. Heparin started for suspected left atrial appendage thrombus. Sedation stopped; restless awakening. Decongestion continued with B-lines and pleural fluid on lung ultrasound. Oxygenation problems when weaning PEEP.
- 17-4: TTE: LVEF ~25%, VTI 1.2 cm. Milrinone started. Furosemide continued, target fluid balance −1 to −2 L. Renewed contact with Centre B, to be contacted again if stagnation. Dexmedetomidine started for restlessness. Episode of cold/mottled right leg, spontaneous recovery.
- 18-4: Inadequate coverage of S. aureus pneumonia, flucloxacillin up to 6 g/24h (previously 4 g/24h). Vascular surgery consulted for mottled foot: CTA shows tapering of the SFA at a short occlusion, possible thrombosuction after ICU stay. Decongestion continued.
- 19-4: Ventricular rate 145–175/min, for which amiodarone and then digitalisation. Pain block via popliteal catheter; better perfusion right leg, Doppler checks adequate. Furosemide stopped with A-profile on lung ultrasound.
ICU course, Centre B
- 20-4: Transfer to Centre B for LVAD screening. Because of severe atherosclerotic vascular disease and threatened legs bilaterally (status post ECLS) not an LVAD candidate; moreover no indication with CI 2.9, cardiac function slightly improving. Milrinone at 5.
- Treated with flucloxacillin until 23-04 and ciprofloxacin until 27-04 for pneumonia with S. aureus and Klebsiella. Blood cultures taken (pending), Swan-Ganz removed. Inflammatory parameters fell on 25-04.
- Cold right foot with CTA showing tapering of the distal SFA: 22-04 successful embolectomy. 23-04 cold left foot, CTA: occlusion of the dorsalis pedis artery; given Doppler signals DPA/PTA and distal perfusion managed expectantly, treated with heparin.
- Atrial flutter: digoxin and heparin. AKI improved. Hypernatraemia: 0.5 L water via feeding tube.
- 24-04: Extubated. 25-04: New CVL right, transfer back to Centre A with falling infection parameters. Uncomplicated transport.
ICU course, Centre A (continued)
25-4: Arrival Centre A. New arterial line left radial artery. Milrinone and furosemide continued, water via tube increased.
CCU course, Centre A
- 26-4: Transfer to CCU for careful introduction of HF medication. Attempt to wean norepinephrine/milrinone, target MAP 65–70. Sliding-scale insulin given glucose.
- 27-4: Norepinephrine weaned, milrinone at 5. Aspirin stopped, furosemide 40 bolus. Dapagliflozin started. Still therapeutic nadroparin; if swallow test good, switch to apixaban.
Conclusion
60-year-old man, known substance abuse, admitted with:
- Anterior wall infarction sustained out of hospital, treated with PCI of the LAD (07-04), complicated by cardiogenic shock with poor LV function, reasonable RV function and AV conduction disturbances, for which temporary pacing wire, Impella and VA-ECMO (08-04 until 12/13-04)
- Temporary takeover by Centre B for LVAD screening: not a candidate (severe atherosclerotic vascular disease, threatened legs bilaterally after ECLS; moreover CI 2.9); 25-04 return to Centre A
- Not yet recompensated
- Moderate-to-severe ischaemic mitral regurgitation (due to tethering with LV remodelling)
- De-novo atrial flutter with adequate rate control, CHA₂DS₂-VA = 2 (CHF, coronary disease), suspected left atrial appendage thrombus, on therapeutic nadroparin
- Recovering AKI (eGFR 57, creatinine 118)
- HAP with S. aureus and Klebsiella, CRP falling (121 → 39), antibiotics until 27-04
- ICU-acquired weakness
- Tube-feeding dependent with hyperglycaemias
- Heel pressure ulcers
Status post:
- Embolectomy SFA; DPA occlusion managed expectantly, heparin
- Hypernatraemia
Plan
- Re 1: Milrinone weaning, target MAP 65. Expand HF medication: 27-04 SGLT-2 started, in due course BB, ARNI, MRA. Furosemide 100 mg/24h. Clopidogrel + therapeutic nadroparin (in due course DOAC once swallowing is safe).
- Re 5: Digoxin. Therapeutic nadroparin.
- Re 7: Ciprofloxacin 400 mg three times daily until 27-04 (flucloxacillin until 20-04). Blood cultures Centre B already taken.
- Re 9: Sliding-scale insulin. Dietitian consulted.
- Re 10: Pressure ulcer protocol. Wound-care nurse consulted.
- PM: Physiotherapy, speech therapy.
- Weekend shift: Wean milrinone cautiously step by step. Continue furosemide (not recompensated). Expand HF medication if possible.
Assignment 1, Structured summarising with role instruction
Learning goal: experience the effect of role and audience instructions, and assess which clinical nuances get lost per format.
You'll have the LLM summarise the same case in three formats for three different audiences. Pay particular attention to what disappears: does the LVAD rejection survive? The suspected appendage thrombus? The anticoagulation trade-off?
Live assignment 6.1, Three formats, one case
Step 1. Paste this prompt (with the case) into ChatGPT or Claude:
Step 2. In the same chat, ask for the two other formats:
Assess per format: which clinical nuances disappeared (LVAD rejection, appendage thrombus, bleeding problems, DPA occlusion)? And is that loss acceptable for that audience, or dangerous? Note at least one nuance per format that you would manually put back.
Question 2
What best describes “few-shot prompting”?
Assignment 2, Sparring on the differential diagnosis
Learning goal: use the LLM as a sparring partner for Bayesian reasoning, and recognise where the model anchors, or misses or overweights rare causes.
You give the LLM only the day-1 presentation, without the clinical course. Then you “release” information step by step and watch how the differential shifts.
Live assignment 6.2, Building the DD step by step
Step 1. Start a new chat (important: the model must not know the rest of the case yet) and paste:
Step 2. Now release step by step, after each step: “How does your differential shift, and why?”
- First the quick-look TTE of 7-4 (anteroseptal akinesia, preserved wall thickness, normal RV function, collapsing IVC);
- then the laboratory results and the venous blood gas;
- then the further course (shock, Impella, ECMO).
Step 3. Close with this check question:
Watch for anchoring: does the model cling to its first lead diagnosis, even when new information points elsewhere? And the reverse: does it overweight rare causes as soon as you mention them? Both are instructive, the model mirrors your line of questioning.
Question 3
While sparring on the differential, the LLM gives a plausible-sounding but incomplete DD. What is the best next step?
Assignment 3, Optimising and fact-checking the conclusion/problem list
Learning goal: never adopt LLM output unchecked; verify systematically against the source.
You'll have the LLM rewrite the existing conclusion and then classify every change: improvement, neutral, or error/hallucination. This is the core skill of safe LLM use in the medical record.
Live assignment 6.3, Rewrite, then mark strictly
Paste the conclusion and the plan from the case into the LLM with this prompt:
Now classify every change yourself: improvement / neutral / error or hallucination. Look specifically for: an invented date, a wrongly extrapolated LVEF, a medication that suddenly appears, a CHA₂DS₂-VA score that quietly changed. Found no errors? Then ask: “Which of your changes cannot be traced literally to the source text?”, and see what happens.
Questions 4–6
The LLM writes in the summary: “Patient received a MitraClip on 10-4 for severe MR.” This appears nowhere in the case. This is an example of:
Which prompt strategy reduces the risk of invented details the most?
The resident wants the LLM to reprioritise the problem list. Which prompt yields the most usable output?
Assignment 4, Holding a treatment plan against the guideline
Learning goal: recognise knowledge cut-off and guideline drift; complex trade-offs stay with the doctor.
This patient is a trade-off nightmare: introducing HF medication in a not-yet-recompensated patient, plus anticoagulation with flutter + suspected appendage thrombus + recent PCI (LAD) + recent bleeding along the ECMO cannulas. Exactly the kind of case an LLM will advise on confidently, and possibly outdated or context-blind.
Live assignment 6.4, LLM proposal vs. ESC guideline
Paste this prompt (with the case, or at least the conclusion + the plan):
Test the proposal against the ESC guidelines HF 2023 and AF 2024: (1) what's correct? (2) what's outdated, which guideline version does the model actually “know”? (3) what's context-blind, for example a standard DAPT duration that ignores the bleeding problems and the OAC indication? This is where the doctor makes the trade-off, not the model.
Questions 7–8
The resident asks the LLM about the anticoagulation strategy per “the most recent ESC guideline”. Why is caution warranted?
The LLM advises “standard 12 months DAPT” for this patient. What is the core problem?
Assignment 5, Prompt iteration and privacy
Learning goal: prompt engineering and GDPR-aware working, the two skills you combine on every case.
Finally, you compare a “lazy” prompt with a structured prompt on the same case, and close with the privacy question that precedes every LLM use.
Live assignment 6.5, Lazy vs. structured
Step 1. New chat. The lazy prompt:
Step 2. Another new chat. The structured prompt:
Compare the two outputs on completeness, prioritisation and usability for the weekend shift. Then answer for yourself: which elements of the original case should have been removed or changed before entering it into a public LLM, and why is date shifting alone insufficient? (Think: the combination of a rare course, two centres, an exact sequence of interventions.)
Questions 9–10
Why is date shifting and slightly changing lab values alone insufficient as anonymisation?
Which statement about the doctor's role in clinical LLM use is correct?
Take-home of the case module
1. Format determines what disappears
SBAR, GP letter and plain-language explanation each drop different nuances. The LLM chooses what gets cut, unless you explicitly state what must survive (appendage thrombus, LVAD rejection, bleeding problems).
2. Sparring partner, not decision-maker
Building a DD and drafting therapy proposals are fine practice terrain, provided you probe for what's missing, anchor to the source and hold every proposal against the current guideline. Knowledge cut-off and context blindness are predictable weaknesses.
3. Privacy before the prompt
This case was allowed in a public chat because it was edited for teaching. For a real case: date shifting alone is insufficient, the combination of a rare course and context remains identifiable. When in doubt: hospital-approved environment or don't.