Pulse Wave Analysis to Estimate Cardiac Output: Comment

We read with great interest the recent review in Anesthesiology, “Pulse Wave Analysis to Estimate Cardiac Output.”1 As military anesthesiologists, we are very excited about future potential allow advanced waveform analysis help better guide treatment when noninvasive monitors not available or impractical for clinical situation. Our group has investigated use of pulse wave applications, particularly traumatic hemorrhage. There several nuances think important discuss that have thus far limited these devices.Technologies imputing hemodynamic parameters from primarily multivariate linear and polynomial2 regression identify associations between features such as blood pressure, stroke volume, cardiac output (CO), systemic vascular resistance (SVR), preload. For example, while there been updates Vigileo software (Edwards Lifesciences Corp., USA), original algorithm was a simple based on formula CO = κ (α × PWTT β) HR, where κ, α, β constants unique each patient determined by transit time (PWTT), CO, heart rate (HR).3 These systems enhanced care variety settings physiologic states, high degree accuracy.However, models often rely assumptions relate pertinent variables. An example is used Windkessel model VolumeView ClearSight Corp.). authors point out, ranges invasiveness calibration techniques employed approximate parameters, but inaccuracy technologies extreme states implies this approach critical limitations. Although polynomial allows modeling nonlinearities data (unlike regression), it can be sensitive outliers produce highly inaccurate predictions at extremes range. In fact, discussed Edwards patent application technology VolumeView: “The accuracy method may low some situations basic empirical relationships valid.” This caution perioperative relevant physiology clinicians need reliable information make decisions.Advances computer electronic medical records promoted computer-based learning algorithms analyze data.4 involves fall under category machine learning, including neural networks, more commonly referred artificial intelligence. methods enable computers trends processes too nuanced traditional identify.A study compared volume prediction during liver transplant 34 patients using EV1000 (a system uses analysis) deep convolutional network.5 The outperformed all phases except anhepatic phase (where comparative performance equivocal), most notably had drastically improved reperfusion, stress which arguably patient. previous studies demonstrated limitations rapid cardiovascular changes, particular, SVR. Since extracted its own, without clinician input, found hidden changes were complex capture.In vivo, many indices collinear. hypovolemia, compensation occurs increases rate, contractility, SVR, impact models. Furthermore, they autocorrelated since series data. Consider walking into an operating room trying predict next set vitals monitor. provider would likely much accurate than you because lack memory states. individual related, nonlinear ways specific individual’s their apparent dynamics fluid through vessel become unpredictable abnormal physiology, hemodynamics type monitoring becomes increasingly important. It follows standard statistical approaches significant limitations.The observed arrival other plethysmography consequence dynamic interplay requires intelligence fully capture. Thus, major improvements diagnostic ability achieved analysis. true preload, unreliable. approach, uniquely equipped capture variables, significantly enhance our understanding human responses monitor noninvasively.The declare no competing interests.