🛡️ MIMIC-IV Lite

Overview

De-identified clinical data from MIMIC-IV Demo v2.2 (PhysioNet): 100 patients, 7 hospital tables, 275 admission records.

Data Structure

Table	What It Captures
Admissions	Hospital admission records: admit/discharge times, admission type
Patients	Demographics: age, gender
Diagnoses (ICD)	Diagnosis codes: primary and secondary conditions
Lab Events	Laboratory results: blood cultures, lactate, WBC, creatinine
Prescriptions	Medication orders: antibiotics, vasopressors, timing
ICU Stays	Intensive care records: unit type, LOS

Causal Reasoning Model

Three nested causal chains trace anomalous length-of-stay to root causes:

1. Sepsis Pathway: infection → organ dysfunction → delayed treatment → prolonged stay 2. Timing Delays: weekend/night admission → delayed labs/antibiotics → worse outcomes 3. Comorbidity Burden: elderly + multi-organ involvement → compounding complications

Sample Questions

• "Why did this patient have an anomalously long hospital stay?"
• "How did admission timing affect treatment delays?"
• "What comorbidities compounded the sepsis pathway?"

Prescriptive Recommendations

Ask what to do: get clinical intervention recommendations. The engine recommends early antibiotic initiation for sepsis patients, expedited lab processing for weekend admissions, and the full SSC sepsis bundle for patients with organ dysfunction.

• "What actions should we take for sepsis patients with delayed antibiotics?"
• "How can we reduce treatment delays for weekend and night admissions?"

Want More?

MIMIC-IV Extended adds vitals, cultures, procedures, transfers, fluid balance, and pharmacy data (17 tables, 8 causal pathways) plus additional prescriptive recommendations. Sign in to access it.