dingxingdi Skill: clinical-experience-driven reasoning
1. Capability Definition & Real Case
- Professional Definition: The ability to anchor clinical decision-making (diagnosis, medication, and instructions) in Case-Based Reasoning (CBR) by retrieving analogous historical trajectories from Electronic Health Records (EHR). It utilizes a coarse-to-fine framework: first identifying a cohort of 'similar patients' using structured clinical markers (e.g., diagnosis and medication codes) and then performing a secondary text retrieval over those patients' discharge summaries to extract task-relevant clinical experiences.
- Dimension Hierarchy: Specialized Knowledge Navigation->Corpus-Specific Retrieval & Reasoning->clinical-experience-driven reasoning
Real Case
[Case 1]
- Initial Environment: A large-scale repository of Electronic Health Records (EHR) containing both structured tabular data (treatment codes) and unstructured discharge summaries for thousands of previous hospital visits.
- Real Question: My father (CKD Stage 4) had mitral valve replacement 2 weeks ago. Warfarin 5 mg/day, but INR remains low. Should the discharge dose be adjusted?
- Real Trajectory: 1. Analyze the current patient's profile: CKD Stage 4, post-mitral valve replacement, low INR on 5mg Warfarin. 2. Query the EHR specifically for 'similar patients' who share the CKD Stage 4 and valve replacement markers. 3. Retrieve Case A (Male, 65): Warfarin reduced from 5mg to 3mg due to renal clearance issues. 4. Retrieve Case B (Female, 72): Warfarin increased to 4mg/day after INR dropped. 5. Synthesize these diverse 'experiences' to conclude that early adjustment is common due to INR instability and requires gradual, context-specific changes.
- Real Answer: Yes, Warfarin dosing in CKD Stage 4 post-surgery often requires early adjustment due to INR instability. Past cases suggest close monitoring and gradual dose changes based on renal clearance and specific recovery patterns.
- Why this demonstrates the capability: This case demonstrates that 'General Medical Knowledge' (standard dosage) is insufficient for complex co-existing conditions. The agent must retrieve 'Past Case Experiences' to provide a context-grounded reasoning path that accounts for specific variables like renal clearance and post-surgical instability.
[Case 2]
- Initial Environment: A RAG system with access to historical discharge summaries. The target patient is a male with bilateral ulnar distribution paresthesias and periscapular pain.
- Real Question: Which diagnoses should be documented into the patient's discharge summary based on his symptoms and MRI?
- Real Trajectory: 1. Identify the patient's symptoms (paresthesias, periscapular pain) and MRI findings (disc-osteophyte complex, spinal canal narrowing). 2. Use structural similarity ranking to find 'similar patients' in the EHR with identical symptom profiles. 3. Extract the 'Discharge Diagnosis' sections from those similar patients. 4. Identify that similar patients with C6-C7 herniations were consistently diagnosed with 'disc osteophyte' and 'spinal stenosis'. 5. Map these analogous outcomes back to the target patient's specific presentation to confirm the final labels.
- Real Answer: Disc osteophyte, Spinal stenosis
- Why this demonstrates the capability: This demonstrates the use of a 'similar-patient cohort' as a knowledge base. Success depends on the agent's ability to first identify analogous patients and then correctly extract the 'decision-making result' (the final diagnosis) from their records to apply it to the current target case.
Pipeline Execution Instructions
To synthesize data for this capability, you must strictly follow a 3-phase pipeline. Do not hallucinate steps. Read the corresponding reference file for each phase sequentially:
Phase 1: Environment Exploration Read the exploration guidelines to discover raw knowledge seeds:
references/EXPLORATION.mdPhase 2: Trajectory Selection Once Phase 1 is complete, read the selection criteria to evaluate the trajectory:
references/SELECTION.mdPhase 3: Data Synthesis Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data:
references/SYNTHESIS.md