An exploratory data quality analysis of time series physiologic signals using a large-scale intensive care unit...

Area of Science:

• Clinical Informatics
• Biomedical Data Science
• Critical Care Medicine

Background:

• Physiological data (heart rate, blood pressure) are crucial for clinical decisions in intensive care units (ICUs).
• Electronic health records contain vital signs data valuable for diagnosing and predicting clinical outcomes.
• Limited research exists on the data quality of high-frequency time-series vital signs in ICUs, essential for predictive modeling.

Purpose of the Study:

• To assess data quality issues (completeness, accuracy, timeliness) of minute-by-minute time-series vital signs in the MIMIC-III dataset.
• To evaluate the suitability of this data for developing predictive models of ICU outcomes.

Main Methods:

• Analyzed minute-by-minute time-series data for heart rate (HR), respiratory rate (RR), blood oxygen saturation (SpO2), and arterial blood pressure (ABP) from the MIMIC-III dataset.
• Included 16009 patient-ICU stays from 9410 unique adult patients.
• Measured data completeness, accuracy, and timeliness across the duration of ICU stays.

Main Results:

• Approximately 30% of patient-ICU stays lacked at least 1 minute of data for HR, RR, and SpO2.
• Around 56% of patient-ICU stays had no ABP data for at least 1 minute.
• Only a small fraction of ICU stays (e.g., 12.5% for HR, 4.4% for ABP) met ≥99% data availability across all quality metrics.

Conclusions:

• Significant data quality challenges exist in minute-by-minute vital signs data within the MIMIC-III dataset.
• These data completeness, accuracy, and timeliness issues have critical implications for data scientists and researchers developing ICU predictive models.
• Further efforts are needed to improve the quality and usability of high-frequency physiological data for clinical research and decision support.