Electronic health record data is unable to effectively characterize measurement error from pulse oximetry: a...

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Clinical Measurement

Background:

• Electronic health records (EHR) data have been utilized in research to assess race-based inaccuracies in pulse oximetry (SpO2) measurements.
• Existing studies may overlook the critical impact of SpO2 value variability over time (deviation time) on measurement accuracy.
• This variability poses a significant challenge to the reliability of EHR data for quantifying SpO2 error.

Purpose of the Study:

• To demonstrate that temporal variability in SpO2 measurements renders EHR data unsuitable for quantifying pulse oximetry error.
• To analyze the impact of 'deviation time' on the accuracy of SpO2 measurements derived from EHR data.
• To provide a critical evaluation of methodologies using EHR data for pulse oximetry performance assessment.

Main Methods:

• Utilized the MIMIC-IV Waveform dataset, sampling SpO2 values from 198 intensive care unit patients as reference.
• Simulated EHR-derived SpO2 errors using various deviation times and simulated laboratory oxygen saturation measurements.
• Evaluated simulated pulse oximeter performance (average root mean squared error of 2%) and reproduced regulatory submission analyses, including Bland-Altman plots.

Main Results:

• All quantified error metrics (mean error, standard deviation, A_RMS error) increased linearly with the logarithm of time deviation.
• At a 10-minute deviation time, the average root mean squared (A_RMS) error escalated from a baseline of 2% to over 4%.
• Increasing deviation time significantly amplified measurement errors, demonstrating a clear time-dependent degradation of data reliability.

Conclusions:

• Electronic health records (EHR) data are not reliable for accurately quantifying pulse oximetry (SpO2) error due to inherent temporal variability.
• Prior research relying on EHR data to assess SpO2 imprecision requires cautious interpretation.
• The findings underscore the need for methodologies that account for temporal dynamics when evaluating SpO2 measurement accuracy.