Day 03 Shift Reports

Day 3: Shift-Report Summarization with Claude

Hand off cleaner reports by treating Claude as a thought-organizer, not an oracle. Voice in, structured SBAR out, your clinical intuition still in charge.

~45 min Hands-on By Bo Peng

Today's Goal

Use voice-to-text plus Claude to turn 60-90 seconds of unstructured shift fragments per patient into a clean, SBAR-structured handoff. Eliminate the wandering "and another thing..." pattern that drops information at change-of-shift.

What you'll learn

The handoff problem

The Joint Commission's Sentinel Event database has identified communication failure as a root cause in roughly 60-70 percent of sentinel events for the last two decades. Most of those communication failures happen at handoff — change of shift, transfer between units, return from procedure. SBAR (Situation, Background, Assessment, Recommendation) and I-PASS (Illness severity, Patient summary, Action list, Situation awareness, Synthesis by receiver) were both developed specifically to standardize handoffs and have measurably reduced harm in published studies.

The problem is that under time pressure, even nurses trained on SBAR drift into chronological storytelling: "So she came up from the ED at 1400, and her blood pressure was fine, and then around 1600 she said her chest hurt, and we did an EKG..." The story is accurate but the structure is gone, and the receiving nurse has to extract the actionable items in real time.

Claude is excellent at restructuring. Given the same chronological story, it will produce a SBAR with action items at the top. You spend 60 seconds talking; you get back a 30-second handoff that the next nurse can act on.

The voice-to-structure workflow

  1. Capture by voice. Use your phone's native voice memo or your facility-approved transcription tool. Talk for 60-90 seconds per patient. Do not try to be organized — just dump everything.
  2. Transcribe. iOS Notes, Android Recorder, Otter.ai (with caution — check BAA), or your hospital's enterprise tool will produce a text transcript.
  3. De-identify. Same Day-1 rule. No names, no MRN, no specific dates. Use "the patient" or pseudonyms ("Bed 1," "Bed 2"). The structuring is generic; the identifiers go back in your written report.
  4. Run the SBAR-structuring prompt. Output is structured for handoff.
  5. Review and add what only you know. Your gut feeling about a patient does not survive transcription — you must add it back.
  6. Use it as a script, not a substitute for the bedside conversation. Hand it to the receiving nurse as a written aid; tell the story too.

The SBAR-structuring prompt

sbar-structure.txt
PROMPT
You are an experienced charge nurse organizing end-of-shift report.
The text below is a voice transcript from the off-going nurse. It
is unstructured. Reorganize it into SBAR for handoff:

S (Situation): one sentence on who the patient is, why admitted,
  current acuity. Use generic identifiers (Bed 3, the patient).

B (Background): pertinent history, code status, allergies, isolation
  precautions, lines/tubes/drains in place.

A (Assessment): current systems-based status. Highlight changes
  from previous shift. Bold any abnormal findings.

R (Recommendation / Action items): MUST be a numbered list. The
  most time-sensitive items go first. Include:
  - Pending labs/orders/consults to follow up
  - Scheduled tasks (med due, dressing change, ambulation)
  - Specific things to call the provider about
  - Family/social concerns to be aware of

After SBAR, output a "GAPS" section listing anything that should
have been mentioned but wasn't (no I&O, no pain reassess in last
2 hours, no last bowel movement noted, etc.). Be specific.

Voice transcript:
[paste de-identified transcript]

The "what changed" prompt for high-acuity units

what-changed.txt
PROMPT
For ICU / step-down handoff. Read the transcript below and produce
a one-page report with these three sections only:

1. WHAT CHANGED IN THE LAST 4 HOURS
   - Vitals trends (improving/worsening/stable)
   - Drips titrated up or down
   - New orders, new abnormal labs
   - Any clinical deterioration or improvement

2. ACTIVE PROBLEMS WITH PARAMETERS
   For each problem, list:
   - The problem
   - The number to watch (HR > 120, MAP < 65, etc.)
   - What to do if it crosses

3. NEXT 4 HOURS
   - Scheduled labs/imaging/procedures
   - Med doses due
   - Things to follow up on

[paste de-identified transcript]

What AI absolutely cannot replace

This is the most important section of this lesson. The structuring tool gets your report cleaner. It does not get your report smarter. Several categories of information do not come through transcription and cannot be produced by any model:

Add these manually after the AI does the structuring. They are exactly the kind of information that distinguishes an experienced nurse's handoff from a textbook one.

Voice transcription and PHI. Most consumer voice-to-text services (Apple Dictation, Google Voice typing) process audio in the cloud and are not HIPAA-covered without a BAA. Apple Dictation has on-device processing for newer iPhones; check Settings > Keyboard. Otter.ai offers a HIPAA-covered Enterprise plan. Your hospital may already have a BAA-covered tool (Nuance Dragon, Microsoft Speech, Epic's Hey Epic). Use that. Until then, treat voice memos like written PHI and de-identify at the transcript stage before any AI processing.

Worked example: chronological dump → SBAR

Below is a 70-second voice-memo transcript (de-identified) and the SBAR Claude produced from it. Reading both side by side is the fastest way to internalize the workflow.

voice-dump.txt
RAW INPUT
Bed 4 came up from the ED late last night for a CHF exacerbation,
about 80 years old, history of afib on Eliquis, also has CKD
stage 3. So overnight was getting Lasix IV, 40 milligrams, and
diuresed about 1.8 liters. This morning lung sounds were clearer,
crackles only at the bases now, sats came up to 94 on 2 liters
nasal cannula from 88 on 4 liters. BP runs around 105 over 60.
Last K was 3.4 this morning, hasn't been replaced yet, I think
the team wanted to recheck. Patient is alert and oriented, asked
about going home tomorrow. Daughter is the contact, she's anxious
and calls a lot, that's normal apparently. Has a saline lock in
the right AC, IV team had a hard time yesterday. Foley still in,
draining clear yellow. No bowel movement since admission, hasn't
been a complaint yet. Cardiology is supposed to round this morning
but I haven't seen them. The patient looks better than yesterday
honestly, but I'd still keep an eye on the potassium.
sbar-output.txt
CLAUDE OUTPUT
S: Bed 4. ~80 y/o admitted overnight for acute CHF exacerbation;
   currently improving on diuresis, awaiting cardiology rounds.

B: Hx afib on Eliquis, CKD stage 3. Code/allergies not in transcript
   [VERIFY]. Saline lock R AC (difficult IV stick history).
   Foley in place, draining clear yellow. Daughter is contact;
   anxious by baseline.

A: Respiratory: improving. Crackles bases only (down from diffuse).
   SpO2 94% on 2L NC (down from 4L overnight).
   Cardiac: HR/rhythm not stated [VERIFY]. BP ~105/60.
   Renal/Fluid: net negative ~1.8L overnight on Lasix 40 mg IV.
   K 3.4 this AM — NOT YET REPLACED.
   GI: no BM since admission.
   Mental status: A&Ox, asking about discharge tomorrow.

R (action items, time-sensitive first):
   1. Recheck/replace K — team wanted recheck this morning.
   2. Cardiology rounding today — flag for plan / disposition.
   3. Reassess respiratory after next Lasix dose.
   4. Last BM since admission — consider bowel regimen.
   5. Difficult IV history — if new access needed, ultrasound /
      IV team.
   6. Daughter likely to call — expect frequent updates.

GAPS to address before next shift:
   - Code status not stated.
   - Allergies not stated.
   - Pain assessment not in transcript.
   - HR / rhythm not in transcript (afib history — verify
     controlled).
   - Last bowel movement / bowel regimen status.

HIPAA / nursing-board-compliance pitfalls

Homework before Day 4

  1. For one shift, voice-record your patient summaries before report. Run them through the SBAR prompt.
  2. Compare the AI structure to the report you actually gave. What did the AI catch that you missed? What did you say in person that the AI missed?
  3. Identify three "tacit knowledge" items per patient that the AI cannot produce. Write them down.
  4. Time the workflow. Most nurses get to under 90 seconds per patient by the third shift.

Day 3 Checkpoint

Disclaimer. AI-generated SBARs are drafts. Clinical intuition, family-context, and provider-preference notes must be added by the nurse before report. Always check with your facility's compliance officer, IT/security team, and state board of nursing before integrating any voice-or-AI workflow into clinical handoff.

← Day 2: Patient Education Course index Day 4: NCLEX Prep →