dismech: Skill for Claude Code

Disease Trajectories Mining

Use this skill when you need to mine DT (Disease Trajectories / DisTraj) artifacts and convert them into dismech comorbidity entries.

Quick start

1. Locate a DT JSON file (often includes a phase_dict or edge list).
2. Extract normalized edges with the script below.
3. Pick candidate pairs and map to comorbidity YAML signals.

Example:

python .claude/skills/disease-trajectories/scripts/dt_extract_edges.py path/to/dt.json --format tsv > /tmp/dt_edges.tsv

Workflow

1) Locate DT artifacts

• Search for candidate files:
  • rg --files -g "*.json" and look for names like phase_dict, trajectories, edges.
• If the DT data is external, download and keep the raw file in a scratch location (do not edit in place).

2) Inspect schema quickly

Use a quick introspection to identify top-level keys:

python - <<'PY'
import json
from pathlib import Path
p = Path("path/to/dt.json")
obj = json.loads(p.read_text())
print(type(obj))
if isinstance(obj, dict):
    print(list(obj.keys())[:20])
PY

If there is a phase_dict mapping, it usually encodes pair keys like ICD_A-ICD_B and may include sex stratification. If there is an edges/pairs list, inspect the field names for A/B, sex, and directionality.

3) Extract normalized edges

Use the bundled script:

python .claude/skills/disease-trajectories/scripts/dt_extract_edges.py path/to/dt.json --format tsv > /tmp/dt_edges.tsv

What the script does:

• Handles phase_dict mappings with pair keys like E12-L28.
• Handles edge lists under edges, links, pairs, data, or trajectories.
• Normalizes fields to a consistent row format with disease_a_id, disease_b_id, directionality metrics, sex, p-value, FDR, and source path.

4) Filter candidate pairs

Use standard tools on the TSV output (examples):

• Filter for a specific ICD pair:
  • rg "^E12\tL28\t" /tmp/dt_edges.tsv
• Filter by directionality:
  • awk -F '\t' 'NR==1 || $11=="A_BEFORE_B"' /tmp/dt_edges.tsv
• Filter by sex:
  • awk -F '\t' 'NR==1 || $3=="male"' /tmp/dt_edges.tsv