← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / indication-dossier / references / 05-synthesis.md

Phase 5 — Synthesis

When: waypoints/regulatory_trials.json exists.

Objective

Read all waypoint files and write the final Markdown report. Also write research_output.json for downstream consumers.

Output structure

Final Output: waypoints/indication_dossier_report.md

# Indication Dossier: [Name]

**Definition**: [one-sentence clinical definition]
**ICD-10**: [codes, or "Not a standard diagnostic entity"]
**Parent Indication**: [parent name, or "Root"]

## 1. Population Definition & Epidemiology

### 1.1 Diagnostic Criteria
[How this population is identified. Consensus definitions, diagnostic criteria.
Note if criteria are evolving or non-standard.]

### 1.2 Prevalence & Incidence
[Global and US prevalence. Incidence trends. Data quality notes.]

### 1.3 Demographics & Risk Factors
[Age, sex, racial/ethnic distribution. Key risk factors and comorbidities.]

### 1.4 Natural History
[Disease course, progression, staging. Mortality data. Key inflection points.]

## 2. Disease Biology

### 2.1 Pathophysiology
[Key biological pathways. Validated vs. emerging therapeutic targets.]

### 2.2 Biomarkers
[Diagnostic, prognostic, pharmacodynamic. FDA-qualified vs. exploratory.
Connect to clinical trial utility.]

## 3. Standard of Care

### 3.1 Approved Therapies
[Approved drugs with mechanism, approval year, limitations.
Or: "No therapies are currently approved for this indication."]

### 3.2 Treatment Guidelines
[Current SOC algorithm. Specialty society recommendations.]

### 3.3 Unmet Need
[What current therapy fails to address. Disease modification gap.
Underserved populations.]

## 4. Clinical Endpoints & Regulatory Path

### 4.1 Accepted Endpoints
[FDA/EMA accepted primary endpoints. Surrogate vs. clinical.
Or: "No FDA guidance exists for this indication."]

### 4.2 Regulatory Precedents
[Key approvals, guidance documents, advisory committee discussions.]

### 4.3 Trial Design Parameters
[Typical trial sizes, durations, comparators, per-patient costs.
Based on Phase 3 trial patterns from ClinicalTrials.gov.]

## 5. Key Trials

### 5.1 Landmark Trials
[3-5 most important trials. For each: NCT ID, drug, results, impact.]

### 5.2 Notable Failures
[Significant failures and mechanism-level lessons learned.]

## Appendix: Sources

[Numbered list of every source consulted, grouped under bold subheadings by
source type. Numbering is continuous across groups. Title is the hyperlink;
accessed date closes the entry.]

Actions

Step 1: Load all waypoint data - Read waypoints/meta.json - Read waypoints/epidemiology.json - Read waypoints/biology_soc.json - Read waypoints/regulatory_trials.json - Read waypoints/sources_evaluated.json - If writing reveals a missing value for a field already in a waypoint, one targeted fetch to fill it is fine; record the new source in sources_evaluated.json. Do not open new research threads.

Step 2: Write waypoints/indication_dossier_report.md

CRITICAL: Write to <workdir>/waypoints/indication_dossier_report.md (inside waypoints/).

Follow the output_structure above. Key guidance:

Sections 1-3: Narrative prose with inline citations. Be specific with numbers (prevalence rates, trial sizes, approval years). Coverage status belongs in research_output.json, not the prose. In the report, only name what is partial or missing; never label a section as covered.

Section 4: Mix of prose and structured data. Include tables for endpoint comparisons and trial parameters where helpful.

Section 5: Structured entries for each trial. Connect to lessons for future trial design.

Throughout: Frame from the patient population perspective. Use the framing guidance from SKILL.md.

Figures: Only generate a chart when it shows a relationship or distribution that prose cannot. Give each a one-line caption that names the data source. After rendering, Read the image and answer: does this tell the reader something the surrounding prose does not? Are title, axes with units, and legend all present and legible? Do axis ticks and chart type match the data (no fractional ticks on years, counts, or categories; no misleading scales)? If any answer is no, delete the figure rather than regenerating.

Step 3: Write waypoints/research_output.json

{
  "indication_name": "...",
  "parent_indication": "...",
  "meta": {...},
  "epidemiology": {...},
  "biology_soc": {...},
  "regulatory_trials": {...},
  "sources_evaluated": [...],
  "coverage_summary": {
    "epidemiology": {"covered": [...], "partial": [...], "missing": [...]},
    "biology_soc": {"covered": [...], "partial": [...], "missing": [...]},
    "regulatory_trials": {"covered": [...], "partial": [...], "missing": [...]}
  }
}

Step 4: Mark complete

CRITICAL: Write waypoints/indication_dossier_report.md FIRST, then this step.

Write waypoints/progress.json:

{
  "complete": true,
  "output_file": "indication_dossier_report.md",
  "current_phase": "synthesis"
}

Guidance