AI for Clinicians

Home / Cardiology

Use caseBeginner5 min read

Drafting Heart Failure Discharge Summaries

How to use an AI tool to draft a structured discharge summary after a heart failure admission — turning clinical bullet points into a clear, professional document.

The problem

Discharge summaries after heart failure admissions are time-consuming to write well. The clinical picture is complex — precipitating cause, fluid balance, medication changes, echo findings, follow-up plan. Writing a clear, complete summary at the end of a busy admissions week, when your next clinic is already starting, is not always possible at the standard you would want.

A poor discharge summary creates downstream problems. The GP who receives it does not understand what changed. The heart failure nurse does not know the follow-up plan. The patient is re-admitted because the warning signs to monitor were not communicated.

How AI helps

An AI tool can take a set of clinical bullet points and produce a structured discharge summary draft in under a minute. You provide the clinical facts — the admission reason, the investigations, the treatment changes, the plan. The tool organises them into a readable, professional document. You check the accuracy, add what is missing, and finalise.

This is not the tool making clinical judgements. It is the tool handling the structural work: turning bullet points into sentences, organising content under appropriate headings, and writing in a professional register. The clinical accuracy is entirely yours.

A real example

Dr Sanjay is a consultant cardiologist finishing a weekend on-call. He has a patient in his mid-seventies admitted with decompensated heart failure secondary to dietary non-adherence. The patient is being discharged Monday morning. Dr Sanjay has ten minutes before his outpatient clinic starts.

He opens Claude and types:

Try it yourself
Draft a discharge summary for the following admission. Use professional 
medical language with clear headings.

Admission:
- Man, mid-seventies, background of ischaemic heart disease and 
  heart failure with reduced ejection fraction (HFrEF)
- Admitted with acute decompensation — peripheral oedema and orthopnoea
- Precipitant: dietary indiscretion over Christmas period

Investigations:
- BNP markedly elevated on admission, improved on discharge
- CXR showed pulmonary congestion on admission, clear on discharge
- Echo unchanged from prior — EF 35%

Treatment:
- IV diuresis for 48 hours then converted to oral
- Furosemide dose increased at discharge
- Other HF medications continued unchanged

Discharge plan:
- Follow-up with heart failure nurse specialist in 2 weeks
- GP to check renal function and electrolytes in 1 week
- Patient counselled on fluid restriction and daily weight monitoring
- Return if weight gain greater than [threshold I will insert] or 
  increasing breathlessness

Do not include specific drug doses. I will add those.

The tool produces a structured four-section summary in twenty seconds. Dr Sanjay reads it, adds the specific doses, inserts the weight gain threshold from his unit's protocol, and corrects one sentence that slightly misrepresents the echocardiogram timing. The summary is ready in seven minutes.

Try it yourself

Prompt

Things to watch for

The tool does not know your local pathways. Follow-up intervals, investigation requests, and escalation thresholds vary between institutions. Check every element of the discharge plan against your local protocols.

Drug doses must be added by you. The prompt correctly tells the tool not to include doses. This is intentional. Drug dose errors in discharge summaries cause serious harm. You add doses yourself, from the medication reconciliation you have performed clinically.

It may omit things you did not include in your bullet points. The tool only works with what you give it. If your bullet points are incomplete, the summary will be incomplete. This is why the review step is essential — not a formality.

Professional register can sound generic. Occasionally the tool produces phrases that are correct but impersonal — "the patient was medically optimised" rather than a specific description of what was done. Read for clinical specificity and adjust where the language is too vague to be useful to the receiving clinician.

Remember: AI is a helpful assistant, not a clinician. You make the call.

Was this lesson helpful?

Related lessons