Blood Transfusion Timing Analysis - MIMIC-IV v3.1 Validation

Overview

This repository contains validation analyses for blood transfusion timing strategies using MIMIC-IV v3.1 database. The study examines outcomes for early versus late transfusion strategies in ICU patients, with special focus on cardiovascular cohorts.

Research Question

Does the timing of blood transfusion (based on hemoglobin thresholds) affect patient outcomes in critically ill patients?

Key Analyses

1. Transfusion Group Classification

• Early transfusion: Hemoglobin (Hgb) ≥ 8 g/dL
• Late transfusion: Hemoglobin (Hgb) < 8 g/dL

2. General Population Analysis

• Base cohort characteristics (demographics, hemoglobin levels)
• Summary statistics by transfusion group
• Gender, age, race, and ethnicity distributions
• Hemoglobin band distributions (stratified by severity)
• Mortality rates and ICU length of stay

3. Cardiovascular Cohort Analysis

Focused analysis on patients with primary cardiovascular diagnoses, including:

• Pre- and post-transfusion hemoglobin changes (Δ Hgb)
• Timing to first transfusion after ICU admission
• Hospital mortality rates
• Admission type (elective vs. emergency)
• Demographics (gender, age groups, race, insurance, admission type)

4. Top Primary Diagnoses

Identification of the top 5 primary diagnoses among transfusion patients in each group.

Data Source

MIMIC-IV v3.1 (Medical Information Mart for Intensive Care)

• physionet-data.mimiciv_3_1_icu.* - ICU events and stays
• physionet-data.mimiciv_3_1_hosp.* - Hospital admissions, lab events, patients, diagnoses

Key Variables

• RBC Transfusion Item IDs: 225168, 226368, 227070, 221013
• Hemoglobin Lab Item IDs: 50811, 51222, 51640
• Pre-transfusion window: 24 hours before transfusion
• Post-transfusion window: 6-48 hours after transfusion

Methodology

Cohort Selection

1. Identified all ICU patients receiving RBC transfusions
2. Retrieved pre-transfusion hemoglobin values (within 24h)
3. Classified patients into Early/Late transfusion groups
4. Linked demographics, outcomes, and diagnoses

Statistical Summaries

• Patient counts and percentages
• Mean, median, min, max, standard deviation for continuous variables
• Mortality rates
• Length of stay in ICU

Cardiovascular Subgroup

Filtered to primary cardiovascular diagnoses using ICD codes containing:

• Cardiovascular, coronary, heart, valve, myocardial conditions

Key Findings

The analysis generates comprehensive demographic and outcome comparisons between early and late transfusion groups, enabling:

• Assessment of transfusion timing strategies
• Identification of outcome differences
• Understanding of patient characteristics in each group

Outputs

All results are exported as CSV files:

• summary_results.csv
• gender_distribution.csv
• age_distribution.csv
• race_distribution.csv
• hgb_band_distribution.csv
• demographics_full.csv (cardiovascular cohort)
• Subgroup CSVs by age, gender, race, insurance, admission type

Visualizations

• Age group distribution by transfusion timing
• Hospital mortality comparison (cardiovascular patients)

Technical Notes

• Analysis performed using Google BigQuery via Python client
• Google Colab environment
• Data tables loaded with google.colab.data_table extension
• Matplotlib and Seaborn for visualizations

Dependencies

google-cloud-bigquery
pandas
matplotlib
seaborn

Usage

This notebook validates transfusion strategy outcomes using MIMIC-IV v3.1, providing robust evidence for clinical decision-making regarding optimal transfusion thresholds in ICU settings.

---

Dataset: MIMIC-IV v3.1
Platform: Google Cloud BigQuery
Environment: Google Colab