On better estimating and normalizing the relationship between clinical parameters: comparing respiratory modulations in the photoplethysmogram and blood pressure signal (DPOP versus PPV)

Abstract

DPOP (ΔPOP or Delta-POP) is a noninvasive parameter which measures the strength of respiratory modulations present in the pulse oximeter waveform. It has been proposed as a noninvasive alternative to pulse pressure variation (PPV) used in the prediction of the response to volume expansion in hypovolemic patients. We considered a number of simple techniques for better determining the underlying relationship between the two parameters. It was shown numerically that baseline-induced signal errors were asymmetric in nature, which corresponded to observation, and we proposed a method which combines a least-median-of-squares estimator with the requirement that the relationship passes through the origin (the LMSO method). We further developed a method of normalization of the parameters through rescaling DPOP using the inverse gradient of the linear fitted relationship. We propose that this normalization method (LMSO-N) is applicable to the matching of a wide range of clinical parameters. It is also generally applicable to the self-normalizing of parameters whose behaviour may change slightly due to algorithmic improvements.

Figure 1

Deriving DPOP from the pleth. The amplitude of the cardiac component of the pleth waveform modulates due to respiratory-related changes in intrathoracic pressure which alters cardiac function. Principally, this stems from decreased left ventricular stroke volume during inspiration, leading to decreased pulse amplitude during this phase of respiration.

Figure 2

DPOP versus PPV data plot. Here the “best fit” line shown is calculated using the standard linear least-squares regression method.

Figure 3

Example of a pulse oximeter waveform. The waveform from one of the patients in the study exhibits distinct long-term increases and decreases of the baseline.

Figure 4

Distribution of gradient-associated errors in DPOP. A five-point smoothing distribution is overlaid on the histograms to aid interpretation. (a) Theoretical distribution using simple model with amplitude modulations ranging from 0 to A _min⁡ and a gradient error from −A _min⁡ to A _min⁡. These are arbitrarily chosen ranges and other ranges lead to a similar positively skewed distribution of the error. (b) Actual distribution or errors from the LMSO method. Note that these will also include errors from other sources including random signal noise, although a positive skew is obvious in the data.

Figure 5

Alternative relationship line fits. (a) Least-squares forced through the origin: this method accounts for the relationship passing through the origin as expected from physiological considerations. (b) LMS method: this method is less susceptible to outliers, especially those which may skew the distribution. (c) LMSO method (LMS forced through origin): this combines the robust nature of the LMS method with the requirement for the relationship to pass through the origin.

Figure 6

Relationship line normalization. By normalizing the relationship in this way, DPOP may be substituted directly for PPV as a fluid responsiveness parameter. In particular, the same threshold for indication of hypovolemia may be used for both parameters.