mirror of
https://github.com/NVIDIA/dgx-spark-playbooks.git
synced 2026-06-24 07:09:31 +00:00
285 lines
11 KiB
Python
285 lines
11 KiB
Python
|
"""FHIR query helpers for Clinical Intelligence.
|
||
|
|
|
||
|
|
Import this in analysis scripts:
|
||
|
|
import sys
|
||
|
|
sys.path.insert(0, '/sandbox/clinical-intelligence/skills/analysis-methods/scripts')
|
||
|
|
from fhir_helpers import *
|
||
|
|
|
||
|
|
All HTTP calls use subprocess/curl (requests library does NOT work in sandbox).
|
||
|
|
"""
|
||
|
|
|
||
|
|
import subprocess
|
||
|
|
import json
|
||
|
|
import os
|
||
|
|
import pandas as pd
|
||
|
|
from urllib.parse import urlencode
|
||
|
|
|
||
|
|
BASE_URL = "https://r4.smarthealthit.org"
|
||
|
|
CANVAS_DIR = os.path.expanduser("~/.openclaw/canvas")
|
||
|
|
CANVAS_URL_BASE = "http://localhost:18789/__openclaw__/canvas"
|
||
|
|
|
||
|
|
|
||
|
|
def save_chart_to_canvas(fig, filename, dpi=150, facecolor='#1a1a1a'):
|
||
|
|
"""Save a matplotlib figure to the canvas directory and print the access URL."""
|
||
|
|
os.makedirs(CANVAS_DIR, exist_ok=True)
|
||
|
|
path = os.path.join(CANVAS_DIR, filename)
|
||
|
|
fig.savefig(path, dpi=dpi, facecolor=facecolor, bbox_inches='tight')
|
||
|
|
print(f"\nCanvas URL: {CANVAS_URL_BASE}/{filename}")
|
||
|
|
return path
|
||
|
|
|
||
|
|
|
||
|
|
# ── Core FHIR functions ────────────────────────────────────────
|
||
|
|
|
||
|
|
def fhir_get(path, params=None):
|
||
|
|
"""GET a FHIR endpoint via curl. Returns parsed JSON."""
|
||
|
|
url = f"{BASE_URL}/{path}" if not path.startswith("http") else path
|
||
|
|
if params:
|
||
|
|
url = f"{url}?{urlencode(params)}"
|
||
|
|
r = subprocess.run(["curl", "-sf", "--max-time", "30", url],
|
||
|
|
capture_output=True, text=True, timeout=35)
|
||
|
|
if r.returncode != 0 or not r.stdout.strip():
|
||
|
|
import sys
|
||
|
|
print(f"WARNING: curl failed for {url} (exit code {r.returncode})", file=sys.stderr)
|
||
|
|
if r.stderr:
|
||
|
|
print(f" stderr: {r.stderr[:200]}", file=sys.stderr)
|
||
|
|
return {"entry": []}
|
||
|
|
return json.loads(r.stdout)
|
||
|
|
|
||
|
|
|
||
|
|
def get_all_pages(path, params=None):
|
||
|
|
"""Fetch all pages of a FHIR Bundle."""
|
||
|
|
all_entries = []
|
||
|
|
data = fhir_get(path, params)
|
||
|
|
all_entries.extend(data.get('entry', []))
|
||
|
|
while True:
|
||
|
|
next_url = None
|
||
|
|
for link in data.get('link', []):
|
||
|
|
if link.get('relation') == 'next':
|
||
|
|
next_url = link['url']
|
||
|
|
break
|
||
|
|
if not next_url:
|
||
|
|
break
|
||
|
|
data = fhir_get(next_url)
|
||
|
|
all_entries.extend(data.get('entry', []))
|
||
|
|
return all_entries
|
||
|
|
|
||
|
|
|
||
|
|
# ── Patient cohort ─────────────────────────────────────────────
|
||
|
|
|
||
|
|
def get_patients_with_condition(snomed_code):
|
||
|
|
"""Get unique patient IDs for a SNOMED condition code.
|
||
|
|
Uses bare codes (no system URI) for the SMART test server."""
|
||
|
|
entries = get_all_pages("Condition", {"code": snomed_code, "_count": "200"})
|
||
|
|
patient_ids = list(set(
|
||
|
|
e['resource']['subject']['reference'].split('/')[-1]
|
||
|
|
for e in entries
|
||
|
|
if 'subject' in e.get('resource', {})
|
||
|
|
))
|
||
|
|
print(f"Found {len(patient_ids)} patients with condition {snomed_code}")
|
||
|
|
return patient_ids
|
||
|
|
|
||
|
|
|
||
|
|
# ── Lab retrieval (BATCHED) ────────────────────────────────────
|
||
|
|
|
||
|
|
def get_latest_labs_batch(loinc_code, patient_ids=None):
|
||
|
|
"""Fetch the most recent observation for a LOINC code for ALL patients in one call.
|
||
|
|
Returns dict: patient_id -> (value, unit, date).
|
||
|
|
|
||
|
|
This makes 1-2 FHIR calls regardless of patient count.
|
||
|
|
Do NOT use get_latest_lab() in a loop — use this instead."""
|
||
|
|
entries = get_all_pages("Observation", {
|
||
|
|
"code": loinc_code, "_count": "500", "_sort": "-date"
|
||
|
|
})
|
||
|
|
result = {}
|
||
|
|
pid_set = set(patient_ids) if patient_ids else None
|
||
|
|
for e in entries:
|
||
|
|
obs = e['resource']
|
||
|
|
ref = obs.get('subject', {}).get('reference', '')
|
||
|
|
pid = ref.split('/')[-1] if '/' in ref else ref
|
||
|
|
if pid in result:
|
||
|
|
continue # already have most recent (sorted by -date)
|
||
|
|
if pid_set is not None and pid not in pid_set:
|
||
|
|
continue
|
||
|
|
date = obs.get('effectiveDateTime', 'Unknown')
|
||
|
|
if 'valueQuantity' in obs:
|
||
|
|
result[pid] = (obs['valueQuantity']['value'],
|
||
|
|
obs['valueQuantity'].get('unit', ''), date)
|
||
|
|
elif 'valueString' in obs:
|
||
|
|
result[pid] = (obs['valueString'], '', date)
|
||
|
|
else:
|
||
|
|
result[pid] = (None, None, date)
|
||
|
|
return result
|
||
|
|
|
||
|
|
|
||
|
|
def get_latest_lab(patient_id, loinc_code):
|
||
|
|
"""Get the most recent lab for ONE patient. Returns (value, unit, date).
|
||
|
|
Only use for single-patient lookups. For cohorts, use get_latest_labs_batch()."""
|
||
|
|
r = fhir_get("Observation", {"patient": patient_id, "code": loinc_code,
|
||
|
|
"_sort": "-date", "_count": "1"})
|
||
|
|
if not r.get('entry'):
|
||
|
|
return None, None, None
|
||
|
|
obs = r['entry'][0]['resource']
|
||
|
|
date = obs.get('effectiveDateTime', 'Unknown')
|
||
|
|
if 'valueQuantity' in obs:
|
||
|
|
return obs['valueQuantity']['value'], obs['valueQuantity'].get('unit', ''), date
|
||
|
|
elif 'valueString' in obs:
|
||
|
|
return obs['valueString'], '', date
|
||
|
|
return None, None, None
|
||
|
|
|
||
|
|
|
||
|
|
# ── Blood pressure (component observation) ─────────────────────
|
||
|
|
|
||
|
|
def get_latest_bp(patient_id):
|
||
|
|
"""Get most recent BP. Handles both panel (85354-9) and standalone formats.
|
||
|
|
Returns (systolic, diastolic, date)."""
|
||
|
|
r = fhir_get("Observation", {"patient": patient_id, "code": "85354-9",
|
||
|
|
"_sort": "-date", "_count": "1"})
|
||
|
|
if r.get('entry'):
|
||
|
|
obs = r['entry'][0]['resource']
|
||
|
|
systolic = diastolic = None
|
||
|
|
for comp in obs.get('component', []):
|
||
|
|
code = comp.get('code', {}).get('coding', [{}])[0].get('code', '')
|
||
|
|
if code == '8480-6':
|
||
|
|
systolic = comp.get('valueQuantity', {}).get('value')
|
||
|
|
elif code == '8462-4':
|
||
|
|
diastolic = comp.get('valueQuantity', {}).get('value')
|
||
|
|
if systolic is not None:
|
||
|
|
return systolic, diastolic, obs.get('effectiveDateTime', 'Unknown')
|
||
|
|
|
||
|
|
r = fhir_get("Observation", {"patient": patient_id, "code": "8480-6",
|
||
|
|
"_sort": "-date", "_count": "1"})
|
||
|
|
if r.get('entry'):
|
||
|
|
obs = r['entry'][0]['resource']
|
||
|
|
return obs.get('valueQuantity', {}).get('value'), None, obs.get('effectiveDateTime', 'Unknown')
|
||
|
|
|
||
|
|
return None, None, None
|
||
|
|
|
||
|
|
|
||
|
|
def get_latest_bp_batch(patient_ids):
|
||
|
|
"""Fetch most recent BP for all patients in a cohort.
|
||
|
|
Returns dict: patient_id -> (systolic, diastolic, date).
|
||
|
|
|
||
|
|
Makes 1-2 FHIR calls regardless of patient count.
|
||
|
|
Do NOT use get_latest_bp() in a loop — use this instead."""
|
||
|
|
pid_set = set(patient_ids)
|
||
|
|
result = {}
|
||
|
|
|
||
|
|
# First try BP panel (85354-9)
|
||
|
|
entries = get_all_pages("Observation", {"code": "85354-9", "_count": "500", "_sort": "-date"})
|
||
|
|
for e in entries:
|
||
|
|
obs = e['resource']
|
||
|
|
ref = obs.get('subject', {}).get('reference', '')
|
||
|
|
pid = ref.split('/')[-1] if '/' in ref else ref
|
||
|
|
if pid in result or pid not in pid_set:
|
||
|
|
continue
|
||
|
|
systolic = diastolic = None
|
||
|
|
for comp in obs.get('component', []):
|
||
|
|
code = comp.get('code', {}).get('coding', [{}])[0].get('code', '')
|
||
|
|
if code == '8480-6':
|
||
|
|
systolic = comp.get('valueQuantity', {}).get('value')
|
||
|
|
elif code == '8462-4':
|
||
|
|
diastolic = comp.get('valueQuantity', {}).get('value')
|
||
|
|
if systolic is not None:
|
||
|
|
result[pid] = (systolic, diastolic, obs.get('effectiveDateTime', 'Unknown'))
|
||
|
|
|
||
|
|
# Fallback: standalone systolic (8480-6) for patients without a panel
|
||
|
|
missing = pid_set - set(result.keys())
|
||
|
|
if missing:
|
||
|
|
entries = get_all_pages("Observation", {"code": "8480-6", "_count": "500", "_sort": "-date"})
|
||
|
|
for e in entries:
|
||
|
|
obs = e['resource']
|
||
|
|
ref = obs.get('subject', {}).get('reference', '')
|
||
|
|
pid = ref.split('/')[-1] if '/' in ref else ref
|
||
|
|
if pid in result or pid not in missing:
|
||
|
|
continue
|
||
|
|
val = obs.get('valueQuantity', {}).get('value')
|
||
|
|
if val is not None:
|
||
|
|
result[pid] = (val, None, obs.get('effectiveDateTime', 'Unknown'))
|
||
|
|
|
||
|
|
return result
|
||
|
|
|
||
|
|
|
||
|
|
# ── Medications (BATCHED) ──────────────────────────────────────
|
||
|
|
|
||
|
|
def get_all_medications_batch(patient_ids):
|
||
|
|
"""Fetch active medications for all patients in a cohort.
|
||
|
|
Returns dict: patient_id -> list of medication name strings.
|
||
|
|
|
||
|
|
Makes 1-2 FHIR calls regardless of patient count."""
|
||
|
|
entries = get_all_pages("MedicationRequest", {"status": "active", "_count": "500"})
|
||
|
|
result = {}
|
||
|
|
pid_set = set(patient_ids)
|
||
|
|
for e in entries:
|
||
|
|
med = e['resource']
|
||
|
|
ref = med.get('subject', {}).get('reference', '')
|
||
|
|
pid = ref.split('/')[-1] if '/' in ref else ref
|
||
|
|
if pid not in pid_set:
|
||
|
|
continue
|
||
|
|
name = (
|
||
|
|
med.get('medicationCodeableConcept', {}).get('text')
|
||
|
|
or med.get('medicationCodeableConcept', {}).get('coding', [{}])[0].get('display')
|
||
|
|
or 'Unknown'
|
||
|
|
)
|
||
|
|
result.setdefault(pid, []).append(name)
|
||
|
|
return result
|
||
|
|
|
||
|
|
|
||
|
|
def get_medications(patient_id):
|
||
|
|
"""Get active medications for ONE patient. Returns list of names.
|
||
|
|
Only use for single-patient lookups. For cohorts, use get_all_medications_batch()."""
|
||
|
|
r = fhir_get("MedicationRequest", {"patient": patient_id, "status": "active",
|
||
|
|
"_count": "100"})
|
||
|
|
meds = []
|
||
|
|
for e in r.get('entry', []):
|
||
|
|
med = e['resource']
|
||
|
|
name = (
|
||
|
|
med.get('medicationCodeableConcept', {}).get('text')
|
||
|
|
or med.get('medicationCodeableConcept', {}).get('coding', [{}])[0].get('display')
|
||
|
|
or 'Unknown'
|
||
|
|
)
|
||
|
|
meds.append(name)
|
||
|
|
return meds
|
||
|
|
|
||
|
|
|
||
|
|
def check_drug_class(med_list, drug_names):
|
||
|
|
"""Check if any medication in med_list matches any drug in drug_names (case-insensitive)."""
|
||
|
|
med_lower = [m.lower() for m in med_list]
|
||
|
|
return any(drug in med_text for drug in drug_names for med_text in med_lower)
|
||
|
|
|
||
|
|
|
||
|
|
# ── Cohort DataFrame builder ──────────────────────────────────
|
||
|
|
|
||
|
|
def build_cohort_df(patient_ids, loinc_code, lab_name, drug_check_fn=None):
|
||
|
|
"""Build a DataFrame with lab values and medications for a patient cohort.
|
||
|
|
Uses batched FHIR queries — only 2-3 HTTP calls total.
|
||
|
|
|
||
|
|
Args:
|
||
|
|
patient_ids: list of patient ID strings
|
||
|
|
loinc_code: LOINC code for the lab (e.g. '4548-4' for HbA1c)
|
||
|
|
lab_name: column name for the lab value (e.g. 'HbA1c')
|
||
|
|
drug_check_fn: optional function(med_list) -> bool for medication check
|
||
|
|
"""
|
||
|
|
print(f"Fetching {lab_name} for {len(patient_ids)} patients (batched)...")
|
||
|
|
labs = get_latest_labs_batch(loinc_code, patient_ids)
|
||
|
|
print(f"Fetching medications (batched)...")
|
||
|
|
meds_map = get_all_medications_batch(patient_ids)
|
||
|
|
|
||
|
|
rows = []
|
||
|
|
for pid in patient_ids:
|
||
|
|
lab_val, unit, date = labs.get(pid, (None, None, None))
|
||
|
|
meds = meds_map.get(pid, [])
|
||
|
|
row = {
|
||
|
|
"patient_id": pid,
|
||
|
|
lab_name: lab_val,
|
||
|
|
"lab_date": date,
|
||
|
|
"medications": ", ".join(meds) if meds else "None",
|
||
|
|
"med_count": len(meds)
|
||
|
|
}
|
||
|
|
if drug_check_fn is not None:
|
||
|
|
row["on_target_med"] = drug_check_fn(meds)
|
||
|
|
rows.append(row)
|
||
|
|
|
||
|
|
df = pd.DataFrame(rows)
|
||
|
|
print(f"Built DataFrame: {len(df)} patients, {df[lab_name].notna().sum()} with {lab_name}")
|
||
|
|
return df
|