DOI: 10.1161/circ.148.suppl_1.104 ISSN: 0009-7322

Abstract 104: Photoplethysmogram Signal Characteristics as a Non-Invasive Surrogate of Diastolic Blood Pressure During Cardiopulmonary Resuscitation

Dieter Bender, Grace Sutoris, Marc Hershey, Kathryn Graham, Vinay M Nadkarni, Robert A Berg, Ryan W Morgan, Robert M Sutton, C. Nataraj
  • Physiology (medical)
  • Cardiology and Cardiovascular Medicine

Background: During cardiopulmonary resuscitation (CPR), diastolic blood pressure (DBP) is a key driver of coronary perfusion pressure and thereby, blood flow during CPR. Although resuscitation guidelines advise measuring DBP as an indicator of CPR quality, the need for an invasive arterial catheter to measure DBP limits scope of patients for whom a BP-directed CPR strategy can be implemented.

Aims and Hypothesis: We aimed to study the photoplethysmogram (PPG) signal acquired from pulse oximetry monitoring as a non-invasive indicator of CPR quality. We hypothesized that a machine learning algorithm using dynamic PPG features would accurately classify CPR segments according to whether they met established pediatric DBP thresholds associated with improved outcomes (≥25mmHg in infants and ≥30 mmHg in older children).

Methods/Approach: This was an NHLBI-funded ancillary study of the ICU-RESUScitation clinical trial (NCT02837497). This initial pilot study included 12 children who received CPR for an IHCA and had both arterial BP and pulse oximetry monitoring in place. Physiologic data were summarized into five-second epochs and DBP for each epoch was dichotomized by whether it met age-based DBP thresholds. Time-domain features ( k =25) were systematically engineered from the PPG signal, and a Support Vector Machine (SVM) model with a Gaussian kernel was trained to classify the CPR quality samples.

Results: The analysis yielded n= 268 five-second CPR epochs, of which 147 (55.9%) met DBP thresholds. In the testing set of 134 previously unseen samples (50% of n , n low =59, n high =75), the SVM model achieved a classification accuracy of 91% with 100% precision and specificity, 79.7% sensitivity, and a negative predictive value of 86.2% only influenced by false negatives. Throughout development, six features describing the dicrotic notch characteristics reflected in the PPG waveform were empirically observed as significant CPR quality indicators.

Conclusion: In this multicenter pilot investigation, the PPG waveform derived from pulse oximetry readings showed promise as a non-invasive surrogate of invasively measured DBP during pediatric CPR. This has the potential to greatly broaden the population for whom physiologic monitoring during CPR is possible.

More from our Archive