AgentSkillsCN

chronic-disease-registry-query

根据 CDC 推荐的成人与儿科免疫接种计划，核对患者免疫接种状态，识别超期未接种与即将接种的疫苗，依据患者病情标记禁忌症，并生成 ImmunizationRecommendation FHIR 资源。当用户询问“检查免疫接种”、“疫苗状态”、“哪些疫苗该接种了”、“免疫接种审查”、“疫苗接种史”、“超期疫苗”、提到“免疫接种计划”、“补种疫苗”时，或需要疫苗接种建议时，可使用此技能。切勿用于跨科室的免疫接种人群统计（应使用患者科室概览），或仅针对过敏的查询。

SKILL.md
--- frontmatter
name: chronic-disease-registry-query
description: |
  Builds and queries disease registries from FHIR data for diabetes, hypertension, COPD, asthma, CHF, and CKD. Generates patient counts, severity distributions, treatment patterns, and outcome trends. Use when user asks to "query registry", "disease registry", "registry report", "how many patients with [condition]", "severity breakdown", "treatment patterns", "disease outcomes", or mentions "registry management", "chronic disease tracking". Do NOT use for individual patient review, quality measure calculation (use quality-measure-dashboard), or general panel overview without disease-specific depth (use patient-panel-overview).
metadata:
  author: LangCare
  version: 1.0.0
  mcp-server: langcare-mcp-fhir
  category: population-health

Chronic Disease Registry Query

Overview

Construct virtual disease registries by querying FHIR Condition, Observation, MedicationRequest, and Encounter resources. Support six registries: diabetes, hypertension, COPD, asthma, CHF, and CKD. For each registry, determine patient enrollment, severity/stage distribution, current treatment patterns, outcome metric trends, and intervention triggers. See references/registry-criteria.md for inclusion codes and stratification logic.

FHIR Resources Used

ResourcePurposeKey Fields
ConditionRegistry inclusion, comorbidities, complicationscode, clinicalStatus, severity, onsetDateTime, subject
ObservationSeverity metrics, outcome trackingcode, valueQuantity, effectiveDateTime, component
MedicationRequestTreatment pattern analysismedicationCodeableConcept, status, dosageInstruction, authoredOn
EncounterVisit frequency, hospitalization trackingclass, status, period, reasonCode, subject
PatientDemographics, age/sex stratificationbirthDate, gender, deceasedBoolean
ProcedureSurgical interventions, dialysiscode, performedDateTime, status
CarePlanCare plan adherencestatus, category, activity

Instructions

Step 1: Select Registry

Determine which disease registry to query. Supported registries and their primary codes (see references/registry-criteria.md for full code sets):

RegistrySNOMED CodeICD-10 PrefixKey Metric LOINC
Diabetes44054006, 46635009E10, E114548-4 (A1c)
Hypertension38341003I10, I11-I1385354-9 (BP panel)
COPD13645005J4419926-5 (FEV1/FVC)
Asthma195967001J4519926-5 (FEV1/FVC)
CHF42343007I5042637-9 (BNP), 33762-6 (NT-proBNP)
CKD709044004N1833914-3 (eGFR), 14959-1 (Microalbumin/Cr)

Step 2: Build Registry Population

Query patients with the target condition:

code
Tool: fhir_search
resourceType: "Condition"
queryParams: "code=[snomed-code]&clinical-status=active&_count=200&_include=Condition:subject"

If zero results, retry with ICD-10:

code
Tool: fhir_search
resourceType: "Condition"
queryParams: "code=http://hl7.org/fhir/sid/icd-10-cm|[icd10-prefix]&clinical-status=active&_count=200"

Collect all unique patient IDs. Fetch Patient demographics:

code
Tool: fhir_search
resourceType: "Patient"
queryParams: "_id=[id1],[id2],...&_count=50"

Exclude deceased patients (deceasedBoolean = true or deceasedDateTime present).

Step 3: Determine Severity/Stage Distribution

Query condition-specific severity metrics.

Diabetes -- A1c-based stratification:

code
Tool: fhir_search
resourceType: "Observation"
queryParams: "code=http://loinc.org|4548-4&date=ge[12-months-ago]&_sort=-date&_count=200"

Stratify: Well controlled (A1c < 7%), Moderate (7-9%), Poor (> 9%), Unknown (no recent A1c).

Hypertension -- BP-based staging:

code
Tool: fhir_search
resourceType: "Observation"
queryParams: "code=http://loinc.org|85354-9&date=ge[6-months-ago]&_sort=-date&_count=200"

Stratify per AHA: Normal (< 120/80), Elevated (120-129/<80), Stage 1 (130-139/80-89), Stage 2 (>= 140/90), Crisis (> 180/120).

CHF -- BNP/NYHA-based:

code
Tool: fhir_search
resourceType: "Observation"
queryParams: "code=http://loinc.org|42637-9&date=ge[12-months-ago]&_sort=-date&_count=200"

LOINC 42637-9 = BNP. Also check for NYHA classification in Condition.severity or Observation with LOINC 88020-3.

CKD -- eGFR staging:

code
Tool: fhir_search
resourceType: "Observation"
queryParams: "code=http://loinc.org|33914-3&date=ge[12-months-ago]&_sort=-date&_count=200"

Stratify: Stage 1 (eGFR >= 90 with kidney damage), Stage 2 (60-89), Stage 3a (45-59), Stage 3b (30-44), Stage 4 (15-29), Stage 5 (< 15).

COPD/Asthma -- Spirometry:

code
Tool: fhir_search
resourceType: "Observation"
queryParams: "code=http://loinc.org|19926-5&date=ge[12-months-ago]&_sort=-date&_count=200"

LOINC 19926-5 = FEV1/FVC. GOLD staging for COPD: Mild (>= 80%), Moderate (50-79%), Severe (30-49%), Very Severe (< 30%).

Step 4: Analyze Treatment Patterns

Query active medications for registry patients by therapeutic class.

Diabetes medications:

code
Tool: fhir_search
resourceType: "MedicationRequest"
queryParams: "patient=[patient-id]&status=active&_count=50"

Classify medications into: Metformin, Sulfonylureas, DPP-4 inhibitors, GLP-1 agonists, SGLT2 inhibitors, Insulin (basal, bolus, premixed), Thiazolidinediones, Other. See references/registry-interventions.md for drug class mappings.

Calculate: % on monotherapy, % on dual therapy, % on triple therapy, % on insulin, % on cardioprotective agents (GLP-1 or SGLT2).

Hypertension medications: Classify: ACE inhibitors, ARBs, Calcium channel blockers, Thiazide diuretics, Beta blockers, Mineralocorticoid receptor antagonists, Other.

CHF medications: Track guideline-directed medical therapy (GDMT): % on ACEi/ARB/ARNI, % on beta blocker, % on MRA, % on SGLT2i, % on all four pillars.

Step 5: Track Outcome Trends

Query historical observations for trend analysis (last 24 months, quarterly intervals):

code
Tool: fhir_search
resourceType: "Observation"
queryParams: "code=[loinc-code]&date=ge[24-months-ago]&_sort=date&_count=500"

For each quarter, compute: mean value, median, % at target, % worsening. Present trend as a table.

Step 6: Identify Comorbidity Burden

For each registry patient, query additional active conditions:

code
Tool: fhir_search
resourceType: "Condition"
queryParams: "patient=[patient-id]&clinical-status=active"

Calculate: mean number of comorbidities, most common comorbidity pairs (e.g., diabetes + hypertension, CHF + CKD), % with >= 3 chronic conditions.

Step 7: Check for Complications

Query condition-specific complications. See references/registry-criteria.md for complication codes.

Diabetes complications:

code
Tool: fhir_search
resourceType: "Condition"
queryParams: "code=http://snomed.info/sct|422034002,http://snomed.info/sct|127013003,http://snomed.info/sct|421986006&clinical-status=active&_count=100"

SNOMED: 422034002 = Diabetic retinopathy, 127013003 = Diabetic renal disease, 421986006 = Diabetic peripheral neuropathy.

Step 8: Flag Intervention Triggers

Based on references/registry-interventions.md, flag patients meeting intervention criteria:

  • •Therapy escalation: At-goal metric worsening over 2+ consecutive measurements
  • •Specialist referral: Severity exceeding primary care management threshold
  • •Hospitalization risk: High-risk combination of metrics (e.g., CHF with BNP > 500 + recent ED visit)
  • •Care management enrollment: Multiple intervention triggers active simultaneously

Step 9: Present Registry Report

code
CHRONIC DISEASE REGISTRY: [Disease Name]
==========================================
Registry Size: [N] patients
Report Date: [today]

DEMOGRAPHICS
------------
Mean Age: [age] | Male: [%] | Female: [%]
Age Distribution: 18-40: [N] | 41-65: [N] | 66+: [N]

SEVERITY DISTRIBUTION
---------------------
[Severity Level 1]: [N] ([%])
[Severity Level 2]: [N] ([%])
[Severity Level 3]: [N] ([%])
Unknown/No Data:    [N] ([%])

TREATMENT PATTERNS
------------------
[Drug Class 1]:     [N] ([%])
[Drug Class 2]:     [N] ([%])
Monotherapy:        [N] ([%])
Dual therapy:       [N] ([%])
Triple+ therapy:    [N] ([%])

OUTCOME TRENDS (Quarterly, Last 12 Months)
-------------------------------------------
Quarter  | Mean [Metric] | % at Target | Trend
Q1 2024  | [value]       | [%]         | --
Q2 2024  | [value]       | [%]         | [up/down/stable]
Q3 2024  | [value]       | [%]         | [up/down/stable]
Q4 2024  | [value]       | [%]         | [up/down/stable]

COMPLICATIONS
-------------
[Complication 1]: [N] ([%])
[Complication 2]: [N] ([%])
[Complication 3]: [N] ([%])

INTERVENTION TRIGGERS
---------------------
Patients needing therapy escalation:  [N]
Patients needing specialist referral: [N]
High hospitalization risk:            [N]
Care management candidates:           [N]

Examples

Example 1: Diabetes Registry Query

User says: "Pull up the diabetes registry. Show severity breakdown and treatment patterns."

Actions:

  1. •Search Condition for SNOMED 44054006 and 46635009 (T1DM + T2DM). Returns 94 patients.
  2. •Fetch demographics. Exclude 2 deceased. Registry = 92 patients.
  3. •Query A1c (LOINC 4548-4) in past 12 months. Stratify by control level.
  4. •Query MedicationRequest for each patient. Classify by drug class.
  5. •Query diabetes complications (retinopathy, nephropathy, neuropathy).

Result:

code
CHRONIC DISEASE REGISTRY: Diabetes Mellitus
=============================================
Registry Size: 92 patients (89 T2DM, 3 T1DM)
Report Date: 2024-11-15

DEMOGRAPHICS
------------
Mean Age: 58.4 | Male: 48 (52.2%) | Female: 44 (47.8%)

SEVERITY (A1c-based)
---------------------
Well Controlled (< 7%):    52 (56.5%)
Moderate (7.0-8.9%):       24 (26.1%)
Poor Control (>= 9%):       9 (9.8%)
No Recent A1c:               7 (7.6%)

TREATMENT PATTERNS
------------------
Metformin:           78 (84.8%)
SGLT2 inhibitor:     31 (33.7%)
GLP-1 agonist:       22 (23.9%)
DPP-4 inhibitor:     12 (13.0%)
Sulfonylurea:        18 (19.6%)
Insulin (any):       24 (26.1%)
Monotherapy:         34 (37.0%)
Dual therapy:        38 (41.3%)
Triple+ therapy:     20 (21.7%)
Cardioprotective (GLP-1/SGLT2): 41 (44.6%)

COMPLICATIONS
-------------
Retinopathy:    11 (12.0%)
Nephropathy:    14 (15.2%)
Neuropathy:     19 (20.7%)
Any complication: 28 (30.4%)

INTERVENTION TRIGGERS
---------------------
Need therapy escalation (A1c rising): 6 patients
Need ophthalmology referral (no eye exam): 23 patients
Need nephrology referral (eGFR < 30): 3 patients

Example 2: CHF Registry with GDMT Analysis

User says: "How many heart failure patients are on full GDMT? Show me the gaps."

Actions:

  1. •Search Condition SNOMED 42343007 (CHF). Returns 34 patients.
  2. •Query MedicationRequest for each patient. Check for 4 GDMT pillars: ACEi/ARB/ARNI, beta blocker, MRA, SGLT2i.
  3. •Query BNP (LOINC 42637-9) for severity.

Result:

code
CHF REGISTRY - GDMT ANALYSIS
==============================
Registry Size: 34 patients

GDMT PILLAR COVERAGE
---------------------
ACEi/ARB/ARNI:    28 (82.4%)
Beta Blocker:      30 (88.2%)
MRA:               18 (52.9%)
SGLT2 Inhibitor:   14 (41.2%)

COMBINATION ANALYSIS
--------------------
All 4 pillars:     11 (32.4%)
3 of 4 pillars:    12 (35.3%)
2 of 4 pillars:     8 (23.5%)
1 or fewer:          3 (8.8%)

LARGEST GAP: SGLT2 inhibitors (20 patients not on one)
- 8 of these have no contraindication documented
- ACTION: Review for SGLT2i initiation

PATIENTS MISSING MULTIPLE PILLARS:
1. Robert Ellis (MRN: 44102) - On beta blocker only - BNP: 890 - HIGH PRIORITY
2. Carol Fisher (MRN: 55210) - On ACEi + BB - BNP: 620 - Add MRA + SGLT2i
3. James Ward (MRN: 33098) - On BB only - BNP: 445 - Needs ACEi/ARB + MRA + SGLT2i

Troubleshooting

Registry population is much smaller than expected

  • •Check for alternate coding. Many FHIR servers use ICD-10 exclusively. Add ICD-10 codes to the search: code=http://hl7.org/fhir/sid/icd-10-cm|E11,http://snomed.info/sct|44054006.
  • •Conditions may be stored with verificationStatus = provisional or unconfirmed. Include these if the registry should capture suspected cases: verification-status=confirmed,provisional.
  • •Some systems record historical conditions with clinical-status = inactive but never update to active after confirmation. Consider including inactive with recent onsetDateTime.

Medication classification fails due to non-standard codes

  • •Fall back to medicationCodeableConcept.text for string matching against drug class keywords (e.g., "metformin", "lisinopril", "carvedilol").
  • •If the server uses Medication references instead of CodeableConcept, resolve the reference with fhir_read on the Medication resource to get the code.
  • •Request medications with _include=MedicationRequest:medication to pull contained Medication resources inline.

Observation queries return data for patients outside the registry

  • •Always filter returned Observations by matching subject.reference against the registry patient ID list.
  • •If the server supports it, use patient-specific queries: patient=[id]&code=[loinc]&_sort=-date&_count=1 to get the most recent value per patient. This is slower but more precise.

Related Skills

  • •patient-panel-overview -- lighter-weight panel aggregate without registry depth
  • •quality-measure-dashboard -- formal HEDIS/CMS measure calculation
  • •care-gap-identifier -- individual patient care gap identification
  • •diabetes-panel-review -- individual patient diabetes review
  • •renal-function-dashboard -- individual CKD patient analysis