Condition Stories

A condition story is a pre-computed Markdown narrative for a single clinical condition. It assembles all relevant information from the knowledge graph (onset, medications, labs, complications, monitoring status) into a single readable document. The relationships between conditions, medications, and labs are inferred by the clinical knowledge base.

Why condition stories?

A condition like Type 2 Diabetes Mellitus doesn't exist in isolation. Understanding it requires knowing:

• When it was first diagnosed and across which sources
• Which medications are currently treating it
• Which labs monitor its control
• Whether there are complications (CKD, neuropathy, retinopathy)
• Whether those complications are being adequately monitored

This information is distributed across multiple nodes in the graph. A condition story assembles it once so an agent can get a complete and coherent portrait of the condition in a single place.

Example

# Type 2 Diabetes Mellitus

**Status:** Active since 2016-03
**Codes:** SNOMED 44054006 · ICD-10 E11.9
**Sources:** 3 sources (FHIR Bundle, Summary_20230907.xml, Labs_2023.xml)

## Current Treatment

| Medication    | Dose       | Since   |
| ------------- | ---------- | ------- |
| Metformin     | 1000mg BID | 2016-04 |
| Empagliflozin | 10mg QD    | 2020-08 |

## Monitoring Labs

| Lab             | Latest    | Date       | In Range |
| --------------- | --------- | ---------- | -------- |
| HbA1c           | 7.1%      | 2023-09-07 | ✓        |
| eGFR            | 52 mL/min | 2023-08-15 | ⚠ Low    |
| Fasting glucose | 118 mg/dL | 2023-09-07 | ✓        |

## Complications

- **Chronic Kidney Disease (Stage 3b)**: active since 2021
  - Complication of: Type 2 Diabetes Mellitus
  - Monitored by: eGFR, urine albumin/creatinine ratio
  - Last eGFR: 52 mL/min (below target of ≥60)

Patient-reported context

When a symptom memory has been matched against the condition, the story includes a ## Patient-Reported section built from those relationships:

## Patient-Reported

### Conflicts with record

- "My blood sugar readings have actually been running closer to 180 some mornings"

### Corroborations

- "My fasting sugars have been running around 120 most mornings"

Entries are grouped by relationship type (conflicts, corroborations, additional context) and rendered from the memory's content. This section only appears when at least one matching memory exists; it never contains an empty header. adherence memories, which relate to medications rather than conditions, surface the same way on the medication's own file instead. See Memory for how memories are generated and matched to facts.

Token budget

Condition stories respect the token_budget parameter. The narrative engine prioritizes:

1. Current status and onset
2. Active medications
3. Recent labs
4. Complications and their monitoring status
5. Historical details and source provenance

Lower budgets produce shorter summaries focused on the most clinically relevant information. A compact format gives a one-paragraph overview; narrative gives the full story with tables.

Accessing condition stories

Via MCP (see MCP Tools):

read_patient("/patient/{id}/conditions/active/{condition}/_story.md")
read_patient("/patient/{id}/conditions/active/{condition}/_story.md", format="compact")

Via REST (see REST API reference):

GET /v1/patients/{patientId}/read?path=/conditions/active/type_2_diabetes_mellitus/_story.md&format=narrative

See Read a Condition Story for a practical walkthrough.