Diagnostic Support Summaries

Overview

Generate comprehensive, evidence-based diagnostic summaries that synthesize a patient's clinical presentation, examination findings, and test results into structured differential diagnoses with supporting evidence. This skill applies validated diagnostic criteria, Bayesian clinical reasoning, and evidence-based medicine principles to support — not replace — physician diagnostic decision-making.

When to Use

• Synthesizing complex clinical presentations into structured differentials
• Preparing diagnostic case summaries for conferences or consultations
• Summarizing diagnostic workup progress with next-step recommendations
• Supporting clinical reasoning with evidence-based diagnostic criteria
• Generating teaching case summaries for medical education
• Documenting diagnostic reasoning for medical-legal purposes

Required Inputs

| Input | Description | Format | |-------|-------------|--------| | Clinical presentation | Symptoms, duration, progression, associated features | Structured HPI | | Physical exam findings | Pertinent positives and negatives by system | Structured exam | | Diagnostic test results | Labs, imaging, pathology, special tests | Structured results with reference ranges | | Patient context | Age, sex, comorbidities, medications, risk factors | Structured demographics | | Clinical question | Specific diagnostic question being addressed | Free text |

Methodology

Step 1: Clinical Data Organization

Structure the clinical information using a systematic framework:

Presenting Syndrome Identification:

• Identify the primary presenting syndrome (e.g., acute chest pain, chronic cough, unexplained weight loss)
• Classify by organ system, acuity (acute/subacute/chronic), and severity
• Note temporal pattern (sudden, gradual, intermittent, progressive)

Pertinent Feature Extraction:

• Pertinent positives: findings that support specific diagnoses
• Pertinent negatives: findings that argue against specific diagnoses
• Red flags: findings suggesting dangerous or emergent conditions
• Pattern recognition: classic presentation patterns (e.g., pleuritic chest pain + dyspnea + recent immobilization = PE concern)

Step 2: Differential Diagnosis Generation

Build a prioritized differential using anatomic, pathophysiologic, and probabilistic reasoning:

Framework: "VINDICATE + P"

• Vascular: thrombotic, embolic, hemorrhagic, vasculitic
• Infectious: bacterial, viral, fungal, parasitic
• Neoplastic: primary, metastatic, paraneoplastic
• Degenerative: wear-and-tear, aging-related
• Iatrogenic/Intoxication: drug-related, procedure-related
• Congenital: genetic, developmental
• Autoimmune/Allergic: systemic autoimmune, organ-specific
• Traumatic: acute injury, repetitive stress
• Endocrine/Metabolic: hormonal, electrolyte, metabolic
• Psychogenic: functional, somatoform, psychiatric

Step 3: Evidence Mapping

For each differential diagnosis, map the supporting and refuting evidence:

Diagnostic Criteria Application:

• Apply validated diagnostic criteria where available (e.g., Duke criteria for endocarditis, SLICC criteria for SLE, Light criteria for pleural effusion)
• Calculate pre-test probability using clinical prediction rules where applicable (Wells score for PE, CHA2DS2-VASc for stroke risk)
• Note sensitivity and specificity of key findings for each diagnosis

Evidence Strength for Each Diagnosis:

• Strong support: pathognomonic finding or multiple concordant features
• Moderate support: several consistent features, some expected features absent
• Weak support: possible but fewer consistent features
• Against: key expected features absent or contradictory findings present

Step 4: Workup Assessment

Evaluate the current diagnostic workup status:

• Completed tests: results and their diagnostic implications
• Pending tests: expected timeline and what they will clarify
• Recommended next tests: prioritized by diagnostic yield and clinical urgency
• Test characteristics: sensitivity, specificity, likelihood ratios for recommended tests

Step 5: Summary Generation

Produce the diagnostic support summary:

Summary Components:

1. One-sentence clinical synopsis
2. Ranked differential with probability estimates and evidence basis
3. Key supporting/refuting evidence for top 3 diagnoses
4. Current workup status with pending and recommended tests
5. Clinical reasoning narrative connecting evidence to diagnoses
6. Urgency assessment for time-sensitive diagnoses

Output Specification

The output includes:

clinical_synopsis: one-sentence summary of the case

presenting_syndrome: syndrome name, organ_system, acuity, severity

differential_diagnoses (ranked list): diagnosis name, icd10, probability_estimate (high/moderate/low/unlikely), supporting_evidence list, refuting_evidence list, diagnostic_criteria_met (criteria name, elements met, elements total), key_discriminating_tests

workup_status: completed_tests (test, result, interpretation, diagnostic_implication), pending_tests (test, expected_turnaround, diagnostic_question), recommended_tests (test, rationale, sensitivity, specificity, urgency)

clinical_reasoning_narrative: prose explaining the diagnostic logic

urgency_flags: time-sensitive diagnoses requiring immediate action

evidence_references: guideline or literature citations supporting the analysis

Analysis Framework

Diagnostic Probability Stratification

| Probability Tier | Estimated Likelihood | Action | |-----------------|---------------------|--------| | Must not miss | Any probability, high severity | Rule out immediately regardless of probability | | High probability | Greater than 50% | Primary working diagnosis, confirm | | Moderate probability | 15-50% | Active differential, targeted testing | | Low probability | 5-15% | Consider if initial workup negative | | Unlikely | Less than 5% | Do not pursue unless red flags emerge |

Pre-Test to Post-Test Probability

Apply likelihood ratios to update diagnostic probabilities:

• LR+ greater than 10: Strong rule-in (large increase in probability)
• LR+ 5-10: Moderate rule-in
• LR+ 2-5: Small increase in probability
• LR- 0.2-0.5: Small decrease in probability
• LR- 0.1-0.2: Moderate rule-out
• LR- less than 0.1: Strong rule-out (large decrease in probability)

Examples

Input: 45-year-old female presenting with 3 weeks of progressive fatigue, joint pain (MCPs and wrists bilateral), morning stiffness lasting over 60 minutes, and new malar rash. ANA positive 1:640, dsDNA positive, C3/C4 low, CBC shows mild leukopenia.

Diagnostic Summary (abbreviated):

• Synopsis: 45F with polyarthritis, malar rash, and serologic findings concerning for systemic lupus erythematosus
• Top Differential:
  1. SLE (HIGH) — meets 4+ SLICC criteria (arthritis, malar rash, ANA+, dsDNA+, low complement, leukopenia). Criteria: 4/11 ACR or 4/17 SLICC met
  2. Mixed connective tissue disease (LOW) — overlapping features but anti-U1 RNP not tested
  3. Rheumatoid arthritis (LOW) — symmetric small joint arthritis fits, but rash and serology favor SLE
• Recommended: anti-Smith antibody, anti-U1 RNP, urinalysis with microscopy (lupus nephritis screening), anti-CCP (RA differentiation), complement levels trending
• Urgency: Renal involvement screening is time-sensitive

Guidelines

1. This is decision support, not diagnosis — all output requires physician review and clinical judgment
2. Must-not-miss diagnoses always appear regardless of probability (e.g., PE in chest pain, SAH in headache)
3. Apply Occam's razor judiciously — prefer one unifying diagnosis but acknowledge when multiple diagnoses are more likely
4. Cite diagnostic criteria explicitly — specify which criteria are met and which are not
5. Quantify uncertainty — use probability tiers rather than definitive statements

Validation Checklist

• [ ] All pertinent positives and negatives from the input are addressed
• [ ] Differential diagnoses are prioritized by probability and severity
• [ ] Must-not-miss diagnoses are included regardless of probability
• [ ] Diagnostic criteria are applied correctly with elements enumerated
• [ ] Recommended tests are prioritized by diagnostic yield and urgency
• [ ] Clinical reasoning narrative is logically coherent
• [ ] No diagnostic conclusions are stated as definitive — all framed as supportive

HIPAA Compliance Notes

• Diagnostic summaries contain sensitive PHI including diagnoses and test results
• When used for teaching or case conferences, fully de-identify per Safe Harbor (remove all 18 identifiers)
• Store diagnostic support outputs in the medical record system under appropriate access controls
• Ensure AI-generated diagnostic summaries are clearly labeled as decision support, not final diagnoses
• Maintain audit trails for all diagnostic summary generation and access