Prompt Chain: Build a Comprehensive CDI Case Analysis Workflow

What This Builds

A multi-step prompt chain that walks a de-identified inpatient case through a complete CDI analysis — from initial chart review to prioritized query list to draft queries to physician education notes — in a single structured conversation. Instead of running multiple one-off requests, you chain prompts together so each step builds on the previous one, producing a complete CDI case workup in under 15 minutes.

This is most valuable for complex cases with multiple documentation gaps (multiple comorbidities, potential CC/MCC capture across several diagnoses, concurrent clinical validation concerns) where you'd normally spend 45–60 minutes on analysis before writing a single query.

Prerequisites

• Claude {{tool:Claude.plan}} subscription ({{tool:Claude.price}}) — longer context handling required for multi-step analysis
• Comfortable using Claude for CDI reference work (Level 3)
• Your CDI Project set up with reference documents (see Level 4 guide: "Claude Project: Your Personal CDI Intelligence System") — optional but recommended

The Concept

Prompt chaining is using the output of one AI request as the input to the next. Instead of asking a single question, you run a series of connected questions where each response builds context for the next. For CDI case analysis, this mirrors the way an experienced CDI specialist thinks through a complex case: first identify what's documented, then identify the gaps, then prioritize by DRG impact, then draft the queries.

The power of chaining is that by step 4, Claude has built up a complete understanding of the case context — so the query drafts it produces are more targeted and clinically coherent than if you'd just asked for queries directly.

Build It Step by Step

Part 1: Prepare Your De-identified Case Summary

Before starting the chain, prepare a de-identified case summary. Remove ALL PHI. Create a clinical facts document with:

• Reason for admission / chief complaint
• Key diagnoses documented (use the documented terms, not your CDI interpretation)
• Relevant clinical indicators (labs, vitals, imaging findings — no dates, no identifiers)
• Treatments ordered and administered
• Current clinical status
• Any existing specialist consult findings

Part 2: Start the Chain — Step 1: Clinical Summary

Open Claude (or your CDI Project if you have one set up). Start with:

STEP 1 - CDI CASE REVIEW

I'm going to walk a de-identified inpatient case through a complete CDI analysis. Please respond to each step as I provide it. Start by acknowledging you're ready.

De-identified case summary:
[Paste your prepared case summary]

Step 1: Identify all diagnoses and conditions documented in this case. List each one and note whether the documentation is:
- Specific and complete (no query needed)
- Documented but with insufficient specificity (query opportunity)
- Present clinically but not documented at all (potential query opportunity)
- Uncertain/conditional (may need clinical validation)

Part 3: Step 2 — DRG Impact Analysis

After Claude's response, continue the chain:

STEP 2 - DRG IMPACT ANALYSIS

Based on the documentation gaps you identified, which of these represent the highest DRG impact opportunities? Please rank the top 5 query opportunities by:
1. Estimated DRG weight impact (high/medium/low)
2. Likelihood the physician will confirm the queried diagnosis based on clinical evidence
3. Clinical documentation strength (how well does the evidence support the query?)

Format as a ranked list with brief rationale for each.

Part 4: Step 3 — Query Priority List

STEP 3 - QUERY PLAN

Based on your analysis, create a query plan for this case:
1. List queries in recommended priority order (highest DRG impact first)
2. For each query, specify: query type (specificity/clinical validation/etiology/POA), the documentation gap, and the relevant clinical evidence
3. Flag any queries that have clinical validation risk (where confirming the diagnosis may not be clinically supportable)
4. Note any diagnoses I should NOT query due to insufficient clinical evidence

Part 5: Step 4 — Draft the Top Priority Query

STEP 4 - QUERY DRAFT

Draft a compliant, non-leading AHIMA physician query for the #1 priority item from your query plan.

Requirements:
- Professional opening referencing the clinical review
- Clinical summary with only the documented findings (no interpretation)
- Clear statement of the documentation opportunity
- 4-6 multiple choice response options
- Must include: "Clinically undetermined" and "Other: ___" as final options
- Non-leading language throughout — no preferred answer implied

Part 6: Step 5 — Education Notes

STEP 5 - PHYSICIAN EDUCATION NOTE

Based on this case analysis, what 1-2 sentence documentation guidance would be most useful to share with the attending physician after this case closes? This is for educational follow-up, not for the query itself.

Also: Is there a pattern in this case's documentation gaps that suggests a broader physician education need for this service line?

Part 7: Complete the Remaining Queries

After completing steps 1–5, continue drafting the remaining priority queries:

Now draft AHIMA-compliant queries for priority items #2 and #3 from the query plan, using the same format as the Step 4 query draft.

Real Example: Complex CHF with Complications Case

Setup: De-identified case — patient admitted with CHF exacerbation, also has CKD Stage 3 documented, albumin 2.2, weight loss documented in nursing notes but not by physician, started on diuretics, required Foley for urine output monitoring.

Chain results across 5 steps:

Step 1 output: Identifies 4 documentation gaps: (1) CHF type unspecified (systolic vs. diastolic), (2) CKD with possible AKI given diuretic use (3) Possible malnutrition — not physician-documented, (4) Urinary catheter documented but no indication documented

Step 2 output: Ranks: #1 CHF type (high DRG impact, strong echo evidence), #2 Malnutrition (medium impact, nutrition consult supporting documentation available), #3 AKI (medium impact, creatinine trending data needed), #4 Catheter indication (low impact)

Step 3 output: Query plan with clinical validation flag on malnutrition (need nutrition consult to be specific about severity before querying)

Step 4 output: Compliant CHF type query with 5 response options, professionally formatted

Step 5 output: Education note suggesting physician document CHF ejection fraction in admission note rather than relying on prior echo reference

Total time: 12 minutes for complete 5-step workup on a complex case vs. 45–60 minutes manually.

What to Do When It Breaks

• Claude loses context between steps — If the conversation gets very long, paste a brief "Context: We are in the middle of a CDI case analysis for [brief description]" reminder at the start of each new step
• Step 2 rankings seem off — Provide more clinical specificity: "The echo from [relative time] shows EF of 35% — update your DRG impact analysis"
• Queries don't match your facility format — Add your facility's query template at the top of Step 4: "Draft using this exact format: [paste template]"
• Clinical validation flags seem overly cautious — Ask: "Clarify the clinical evidence threshold for [diagnosis] query. Is the evidence I described sufficient?"

Variations

• Simpler version: Run only Steps 1–3 for documentation gap identification and prioritization; draft queries manually from the ranked list — still saves 20–30 minutes on complex cases
• Extended version: Add a Step 6 asking for a one-paragraph case summary suitable for your CDI program's quality reporting log

What to Do Next

• This week: Run this chain on your 2–3 most complex pending cases and compare quality and time vs. your normal process
• This month: Build this into your workflow for any case with 3 or more potential query opportunities
• Advanced: Combine this with your CDI Project (Level 4 guide) so Claude has your AHIMA guidelines loaded when it does the query drafting step

Advanced guide for Clinical Documentation Improvement Specialist professionals. Requires Claude {{tool:Claude.plan}} subscription ({{tool:Claude.price}}).