|
Pulse oximetry is a common tool to perform a non-invasive optical estimate (SpO2) of arterial blood oxygen saturation level (SaO2). Although the principle of pulse oximetry has been established for a long time Recent clinical studies reported oximeter over-estimation bias in black patients. Measurement accuracy is an important factor, as over-estimation could impact clinical decision-making. Prior Monte-Carlo (MC) simulation-based studies showed increased melanin could reduce the oximeter signal intensity. These studies didn’t show the impact of pigmentation on calibration equation development in a population cohort. Extending MC simulations to study the influence of bias in calibration model enrollment, along with the corresponding optical estimation errors would offer insight into the basis of important clinical observations. Here, an MC simulation platform was developed to assess how pigmentation distribution in the racial demographics could impact calibration model development. MC simulations of oximeter measurements from <1200 simulated patient finger models were generated using a stochastic sampling-based technique, where patient optical properties (including pigmentation) were statistically assigned to generate a variation of measurements across different population cohorts. MC simulations of oximeter calibration studies representative of prior FDA 510(k) guidelines e.g.- minimum 20% darkly pigmented population) in comparison with alternative enrollment distributions. Performance of oximeter calibration equations was evaluated with unique population distributions of test subjects. Results showed that even if the calibration equations were developed from a representative population cohort, the predicted SpO2 show overestimation in high pigmentation cohorts. This over-estimation minimizes when the calibration is generated from distributions with an increased pigmented subject enrollment. The sensitivity to detect hypoxia in the highly pigmented cohort (sensitivity=0.95) is lower than the low pigmented cohort(sensitivity=0.98) when the representative population distribution was used to develop the calibration equation.
|
