Skip to main content

Clinical Decision

Category: AI & ML
Node id: clinical-decision

Standards: HIPAA

Clinical scoring algorithms (qSOFA, NEWS2, CHA2DS2-VASc, Wells PE) and custom LLM-powered decision rules

What it does​

  • qSOFA sepsis screening score
  • NEWS2 national early warning score
  • CHA2DS2-VASc stroke risk scoring
  • Wells PE probability assessment
  • Custom LLM-powered clinical decision rules
  • Structured risk level classification

Use cases​

  • Sepsis early detection in emergency departments
  • Automated patient deterioration scoring
  • Stroke risk assessment for atrial fibrillation patients
  • Pulmonary embolism pre-test probability
  • Custom clinical protocol evaluation via AI

Runtime​

  • Performance tier: variable

Input ports​

NameTypeRequiredDescription
patient_datajsonYesPatient data with vitals and clinical parameters

Output ports​

NameTypeRequiredDescription
resultjsonNoClinical decision result (score, risk level, details, interpretation, recommendations)
metadatajsonNoExecution metadata (algorithm used)
successbooleanNoOperation success flag

Settings​

Basic​

SettingNameTypeRequiredDefaultAllowed valuesDescription
AlgorithmalgorithmenumYesqsofa, news2, cha2ds2vasc, wells_pe, customClinical scoring algorithm to apply
ModelmodelstringNogpt-4oModel name for custom algorithm
Decision Rulesdecision_rulestextNo""Custom decision rules (for custom algorithm)

Auth​

SettingNameTypeRequiredDefaultAllowed valuesDescription
LLM Endpointllm_endpointstringNo""LLM API endpoint (required for custom algorithm)
API Keyapi_keysecretNoBearer token for LLM endpoint (custom algorithm)

Errors​

CodeDescription
ALGORITHM_REQUIREDNo algorithm specified in configuration
PATIENT_DATA_REQUIREDMissing patient_data input
UNSUPPORTED_ALGORITHMAlgorithm name not recognized
LLM_ENDPOINT_REQUIREDCustom algorithm requires llm_endpoint
LLM_REQUEST_FAILEDHTTP request to LLM endpoint failed

Runtime notes​

  • Each algorithm expects different patient_data fields: qSOFA needs systolic_bp/respiratory_rate/gcs; NEWS2 needs additional spo2/pulse/temperature/consciousness; CHA2DS2-VASc needs age/sex/comorbidity booleans
  • result shape varies by algorithm — qSOFA/NEWS2 have risk_level, CHA2DS2-VASc has recommendation, Wells PE has risk_level with float score
  • The custom algorithm delegates to an LLM and returns free-form JSON — validate the decision shape in downstream nodes
  • Built-in algorithms are deterministic and fast; custom algorithm has variable latency due to LLM call

See Build any workflow for how to put nodes together and Ports, types and data flow for how port types connect.

Something unclear? Send documentation feedback by email.