clinical-knowledge

par nvidia

Apprend aux agents les plages de référence cliniques, les codes de conditions, les définitions de mesures de qualité, les classifications de médicaments et le contexte réglementaire afin qu'ils puissent signaler les anomalies…

npx skills add https://github.com/nvidia/dgx-spark-playbooks --skill clinical-knowledge

Clinical Reference Knowledge

Regulatory Context

The 21st Century Cures Act (2016) and ONC's Interoperability Final Rule (2020) require US healthcare organizations to expose patient data through standardized FHIR APIs. As of 2024, ~70% of US hospitals support FHIR R4 (source: ONC). CMS ties quality measure reporting to reimbursement through programs like MIPS (Merit-based Incentive Payment System) and the Hospital Value-Based Purchasing Program. Failure to report -- or poor performance -- results in payment adjustments of up to 9%.

HIPAA (Health Insurance Portability and Accountability Act) prohibits transmitting Protected Health Information (PHI) to external services without a BAA (Business Associate Agreement). This is the primary reason clinical AI must run locally: cloud LLM APIs are not BAA-covered by default, and even those that offer BAAs (e.g., Azure OpenAI) face institutional resistance from hospital compliance teams.

Common Lab Reference Ranges

These are general adult reference ranges. Values vary by lab, assay, and patient characteristics. Always defer to the performing laboratory's reference range when available via FHIR referenceRange.

LabNormal RangeConcerningUnitNotes
HbA1c< 5.7% (non-diabetic), < 7.0% (diabetic target)> 9.0% = poor control%ADA 2024 guidelines; target may be relaxed to < 8.0% for elderly/frail
Fasting Glucose70-100100-125 = prediabetes, >= 126 = diabeticmg/dLMust be fasting; random glucose >= 200 also diagnostic
Creatinine0.7-1.3 (male), 0.6-1.1 (female)> 1.5 = impaired renalmg/dLAffected by muscle mass; less reliable in elderly
eGFR> 90 (normal), 60-89 (mild decrease)30-59 = moderate CKD, 15-29 = severe, < 15 = kidney failuremL/min/1.73m2CKD-EPI 2021 equation (race-neutral)
BUN7-20> 20 with rising creatinine = renal concernmg/dLElevated by dehydration, high protein diet
Total Cholesterol< 200200-239 = borderline, >= 240 = highmg/dL
LDL< 100 (general), < 70 (high-risk ASCVD)> 160 = highmg/dLACC/AHA 2018
HDL> 40 (male), > 50 (female)< 40 = cardiovascular risk factormg/dL
Triglycerides< 150150-499 = elevated, >= 500 = severe (pancreatitis risk)mg/dL
Systolic BP< 120 (normal), 120-129 (elevated)130-139 = Stage 1 HTN, >= 140 = Stage 2 HTNmmHgACC/AHA 2017; CMS165 uses 140/90 threshold
Diastolic BP< 8080-89 = Stage 1 HTN, >= 90 = Stage 2 HTNmmHg
BNP< 100100-400 = possible HF, > 400 = likely HFpg/mLAge-adjusted: higher cutoffs in elderly; obesity lowers BNP
NT-proBNP< 300 (rule-out)Age-stratified: >450 (<50y), >900 (50-75y), >1800 (>75y)pg/mLMore stable than BNP; renal clearance affects levels
Potassium3.5-5.0< 3.5 = hypokalemia, > 5.5 = hyperkalemia (cardiac risk)mEq/LCritical for patients on ACEi/ARB/spironolactone
Sodium136-145< 130 = moderate hyponatremiamEq/LCommon in HF patients
ALT7-56> 3x ULN = significant hepatotoxicityU/LMonitor with statin therapy
Hemoglobin13.5-17.5 (male), 12.0-16.0 (female)< 12 (male) or < 11 (female) = anemiag/dLCommon in CKD (erythropoietin deficiency)

When reporting lab values:

  • Always flag values outside the normal range with the severity (mild / moderate / severe)
  • Note the date of the observation -- a result from 2 years ago has different clinical significance than one from yesterday
  • If the FHIR Observation includes a referenceRange, use that instead of the table above

Condition Codes

SNOMED CT Codes (Primary)

CodeConditionICD-10 EquivalentPrevalence (US adults)
44054006Type 2 Diabetes MellitusE11.x~11% (37M)
46635009Type 1 Diabetes MellitusE10.x~0.5% (1.6M)
38341003Essential HypertensionI10~47% (116M)
84114007Heart FailureI50.x~2.4% (6.7M)
40055000Chronic Kidney DiseaseN18.x~15% (37M)
53741008Coronary Artery DiseaseI25.x~7% (20M)
13645005COPDJ44.x~6% (16M)
195967001AsthmaJ45.x~8% (25M)
49436004Atrial FibrillationI48.x~2% (6M)
73211009Diabetes (unspecified)E11.9Used in older records

ICD-10-CM to SNOMED Crosswalk

When FHIR Condition resources use ICD-10 coding (system http://hl7.org/fhir/sid/icd-10-cm), map as follows:

  • E11.* → Type 2 Diabetes (SNOMED 44054006)
  • E10.* → Type 1 Diabetes (SNOMED 46635009)
  • I10 → Essential Hypertension (SNOMED 38341003)
  • I50.* → Heart Failure (SNOMED 84114007)
  • N18.* → Chronic Kidney Disease (SNOMED 40055000)

Note: A Condition resource may have both SNOMED and ICD-10 codes in the coding array. Always check all entries, not just coding[0].

CMS Quality Measures

CMS122v12 -- Diabetes: Hemoglobin A1c (HbA1c) Poor Control (> 9%)

ComponentDefinition
DenominatorPatients 18-75 with diabetes (Type 1 or Type 2) and at least 2 encounters during the measurement period
NumeratorPatients with most recent HbA1c > 9.0%, OR no HbA1c recorded during the measurement period
ExclusionsHospice care, palliative care, advanced illness with frailty (2+ encounters for advanced illness AND frailty diagnosis), dementia medications (donepezil, rivastigmine, memantine, galantamine)
Performance rateLower is better (inverse measure)
Payment impactPart of MIPS quality reporting; affects Medicare reimbursement

CMS165v12 -- Controlling High Blood Pressure

ComponentDefinition
DenominatorPatients 18-85 with essential hypertension diagnosed before or during the measurement period
NumeratorPatients with most recent BP < 140/90 mmHg
ExclusionsHospice, palliative care, ESRD, kidney transplant, advanced illness with frailty, pregnancy
Performance rateHigher is better
NoteBP must be measured during an outpatient encounter; home BP readings are not counted in the standard measure

CMS135v12 -- Heart Failure: ACEi/ARB/ARNI Therapy for LVEF < 40%

ComponentDefinition
DenominatorPatients 18+ with heart failure AND documented LVEF < 40% (HFrEF)
NumeratorPatients prescribed ACE inhibitor, ARB, or ARNI (sacubitril/valsartan)
ExclusionsHospice, allergy/intolerance to all three classes, bilateral renal artery stenosis, pregnancy, hyperkalemia > 5.5
NoteLVEF data often in DiagnosticReport or CarePlan, not always queryable via Condition alone

CMS134v12 -- Diabetes: Medical Attention for Nephropathy

ComponentDefinition
DenominatorPatients 18-75 with diabetes
NumeratorPatients with nephropathy screening (urine albumin test) OR evidence of nephropathy treatment (ACEi/ARB) OR nephropathy diagnosis
ExclusionsHospice, palliative care, advanced illness with frailty

Drug Classifications

When checking medication coverage, recognize these drug class groupings. Matching should be case-insensitive partial string matching on the medication name from FHIR medicationCodeableConcept.text or .coding[].display.

Diabetes Medications

ClassDrugsNotes
BiguanidemetforminFirst-line therapy
Sulfonylureasglipizide, glyburide, glimepirideHypoglycemia risk
Insulininsulin lispro, insulin glargine, insulin aspart, insulin detemir, insulin degludec, NPH insulinMatch any string containing "insulin"
GLP-1 Receptor Agonistsliraglutide (Victoza), semaglutide (Ozempic/Wegovy/Rybelsus), dulaglutide (Trulicity), exenatide (Byetta/Bydureon), tirzepatide (Mounjaro)Weight loss benefit; cardiovascular benefit
SGLT2 Inhibitorsempagliflozin (Jardiance), dapagliflozin (Farxiga), canagliflozin (Invokana), ertugliflozin (Steglatro)Cardiovascular + renal benefit; monitor for DKA
DPP-4 Inhibitorssitagliptin (Januvia), saxagliptin, linagliptin, alogliptinWeight-neutral
Thiazolidinedionespioglitazone, rosiglitazoneHF risk; edema

Antihypertensives

ClassDrugsNotes
ACE Inhibitorslisinopril, enalapril, ramipril, benazepril, fosinopril, quinaprilCough side effect; monitor K+ and creatinine
ARBslosartan, valsartan, irbesartan, olmesartan, telmisartan, candesartan, azilsartanAlternative if ACEi cough
ARNIssacubitril/valsartan (Entresto)HFrEF guideline-directed; do NOT combine with ACEi
CCBsamlodipine, nifedipine, diltiazem, verapamilDiltiazem/verapamil contraindicated in HFrEF
Beta-Blockersmetoprolol (tartrate or succinate), atenolol, carvedilol, bisoprolol, propranolol, nebivololOnly carvedilol, metoprolol succinate, bisoprolol for HF
Thiazide Diureticshydrochlorothiazide (HCTZ), chlorthalidone, indapamideFirst-line for HTN
Loop Diureticsfurosemide, bumetanide, torsemideVolume management in HF, not primary HTN therapy
Aldosterone Antagonistsspironolactone, eplerenoneHF benefit; monitor K+

Statins (HMG-CoA Reductase Inhibitors)

IntensityDrugs
Highatorvastatin 40-80mg, rosuvastatin 20-40mg
Moderateatorvastatin 10-20mg, rosuvastatin 5-10mg, simvastatin 20-40mg, pravastatin 40-80mg
Lowsimvastatin 10mg, pravastatin 10-20mg, lovastatin 20mg

Heart Failure Medications (Guideline-Directed Medical Therapy)

The four pillars of HFrEF therapy (ACC/AHA 2022):

  1. ACEi/ARB/ARNI -- sacubitril/valsartan preferred over ACEi/ARB
  2. Beta-Blocker -- carvedilol, metoprolol succinate, or bisoprolol only
  3. Aldosterone Antagonist -- spironolactone or eplerenone (if eGFR > 30, K+ < 5.0)
  4. SGLT2 Inhibitor -- dapagliflozin or empagliflozin (regardless of diabetes status)

Matching Strategy

medication_name = fhir_med_text.lower()

is_on_insulin = "insulin" in medication_name
is_on_glp1 = any(drug in medication_name for drug in
    ["liraglutide", "semaglutide", "dulaglutide", "exenatide", "tirzepatide",
     "victoza", "ozempic", "trulicity", "byetta", "mounjaro", "rybelsus"])
is_on_sglt2 = any(drug in medication_name for drug in
    ["empagliflozin", "dapagliflozin", "canagliflozin", "ertugliflozin",
     "jardiance", "farxiga", "invokana", "steglatro"])
is_on_acei = any(drug in medication_name for drug in
    ["lisinopril", "enalapril", "ramipril", "benazepril", "fosinopril", "quinapril"])
is_on_arb = any(drug in medication_name for drug in
    ["losartan", "valsartan", "irbesartan", "olmesartan", "telmisartan",
     "candesartan", "azilsartan"])
is_on_betablocker = any(drug in medication_name for drug in
    ["metoprolol", "atenolol", "carvedilol", "bisoprolol", "propranolol", "nebivolol"])
is_on_statin = any(drug in medication_name for drug in
    ["atorvastatin", "rosuvastatin", "simvastatin", "pravastatin", "lovastatin"])

Clinical Comorbidity Patterns

Common co-occurring conditions to watch for during analysis:

  • Cardiorenal-metabolic overlap: Diabetes + Hypertension + CKD occur together in ~30% of diabetic patients
  • Heart failure + CKD: eGFR < 30 limits medication options (spironolactone, SGLT2i dose adjustment)
  • Diabetes + CAD: Statin therapy should be high-intensity; GLP-1/SGLT2i have cardiovascular benefit
  • Atrial fibrillation + Heart failure: Common pairing; rate control with beta-blocker preferred

Guardrails

  • These reference ranges are general guidelines based on published clinical guidelines (ADA, ACC/AHA, KDIGO), not diagnostic criteria
  • Never state that a patient "has" a condition based on a lab value alone
  • Always include the disclaimer: "This is for informational and research purposes, not clinical decision-making"
  • When flagging care gaps, use language like "may warrant review" not "requires treatment"
  • Acknowledge that CMS measure logic here is simplified -- production implementations use the full eCQM (electronic Clinical Quality Measure) specifications from CMS
  • Lab reference ranges should defer to the performing laboratory's range when available

Plus de skills de nvidia

compileiq-debug
nvidia
Utilisez quand quelque chose ne va pas : Search() bloque, toutes les évaluations retournent INVALID_SCORE, les scores ne s'améliorent pas, chaque configuration retourne le même nombre, erreurs ptxas…
create-github-pr
nvidia
Créer des pull requests GitHub en utilisant l'interface en ligne de commande gh. Utiliser lorsque l'utilisateur souhaite créer une nouvelle PR, soumettre du code pour révision, ou ouvrir une pull request. Mots-clés de déclenchement -…
nemoclaw-maintainer-cross-issue-sweep
nvidia
Analyse les autres problèmes ouverts pour trouver ceux qu’une PR donnée pourrait également corriger ou casser accidentellement. Génère des opportunités de correctifs adjacents et des risques de contradiction avec fichier:ligne…
fhir-basics
nvidia
Apprend aux agents comment fonctionnent les API FHIR R4, quelles ressources sont disponibles, comment les interroger avec des paramètres de recherche, et comment analyser correctement tous les formats de réponse…
compileiq-validate-result
nvidia
Utiliser APRÈS qu'une recherche soit terminée et AVANT de réclamer un accélérateur ou d'expédier un ACF. Charge le CSV dump_results, extrait les K meilleurs candidats (mono-objectif)…
changelog-audit
nvidia
Auditer le CHANGELOG.md de Warp avant une publication : récupérer les entrées perdues, trier par impact utilisateur, affiner le langage des entrées, ajuster les retours à la ligne et (en mode branche de publication) mettre à jour la comparaison…
maintain-dynamic-plugins
nvidia
Maintenir les chargeurs de plugins dynamiques NeMo Relay, les manifestes, les SDK natifs Rust, le protocole worker gRPC, le SDK worker Python, la documentation, les tests et la couverture du workflow de publication
dgx-diagnose
nvidia
Diagnostiquer les problèmes courants du DGX Station GB300 — plantages CUDA, ciblage incorrect du GPU, bugs de conteneur vLLM/SGLang, problèmes d'état MIG, erreurs NVLink/Fabric Manager,…