key: cord-0015553-ycsmo86x
authors: Flick, Moritz; Hoppe, Phillip; Matin Mehr, Jasmin; Briesenick, Luisa; Kouz, Karim; Greiwe, Gillis; Fortin, Jürgen; Saugel, Bernd
title: Non-invasive measurement of pulse pressure variation using a finger-cuff method (CNAP system): a validation study in patients having neurosurgery
date: 2021-02-25
journal: J Clin Monit Comput
DOI: 10.1007/s10877-021-00669-1
sha: 898788dea0c579762a95814672c14481d272356f
doc_id: 15553
cord_uid: ycsmo86x

The finger-cuff system CNAP (CNSystems Medizintechnik, Graz, Austria) allows non-invasive automated measurement of pulse pressure variation (PPV(CNAP)). We sought to validate the PPV(CNAP)-algorithm and investigate the agreement between PPV(CNAP) and arterial catheter-derived manually calculated pulse pressure variation (PPV(INV)). This was a prospective method comparison study in patients having neurosurgery. PPV(INV) was the reference method. We applied the PPV(CNAP)-algorithm to arterial catheter-derived blood pressure waveforms (PPV(INV−CNAP)) and to CNAP finger-cuff-derived blood pressure waveforms (PPV(CNAP)). To validate the PPV(CNAP)-algorithm, we compared PPV(INV−CNAP) to PPV(INV). To investigate the clinical performance of PPV(CNAP), we compared PPV(CNAP) to PPV(INV). We used Bland–Altman analysis (absolute agreement), Deming regression, concordance, and Cohen's kappa (predictive agreement for three pulse pressure variation categories). We analyzed 360 measurements from 36 patients. The mean of the differences between PPV(INV−CNAP) and PPV(INV) was −0.1% (95% limits of agreement (95%-LoA) −2.5 to 2.3%). Deming regression showed a slope of 0.99 (95% confidence interval (95%-CI) 0.91 to 1.06) and intercept of −0.02 (95%-CI −0.52 to 0.47). The predictive agreement between PPV(INV−CNAP) and PPV(INV) was 92% and Cohen’s kappa was 0.79. The mean of the differences between PPV(CNAP) and PPV(INV) was −1.0% (95%-LoA−6.3 to 4.3%). Deming regression showed a slope of 0.85 (95%-CI 0.78 to 0.91) and intercept of 0.10 (95%-CI −0.34 to 0.55). The predictive agreement between PPV(CNAP) and PPV(INV) was 82% and Cohen’s kappa was 0.48. The PPV(CNAP)-algorithm reliably calculates pulse pressure variation compared to manual offline pulse pressure variation calculation when applied on the same arterial blood pressure waveform. The absolute and predictive agreement between PPV(CNAP) and PPV(INV) are moderate.

Pulse pressure variation (PPV) caused by mechanical ventilation can predict fluid responsiveness [1, 2] . PPV is determined by heart-lung interactions; mechanical ventilation with positive airway pressure causes cyclic changes in venous return and cardiac preload resulting in variable changes in the arterial blood pressure waveform that can be quantified by PPV [3, 4] . Automated measurement of PPV requires continuous recording and analysis of the arterial blood pressure waveform, usually invasively using an arterial catheter.

In recent years, innovative finger-cuff technologies became available that allow continuous recording of the arterial blood pressure waveform and PPV calculation in a non-invasive manner [5] [6] [7] [8] [9] [10] . The CNAP system (CNAP Monitor 500; CNSystems Medizintechnik, Graz, Austria) is one commercially available finger-cuff system and has been validated for arterial blood pressure and cardiac output measurements [11] . Using a proprietary algorithm, the CNAP system also automatically calculates PPV (PPV CNAP ).

We here sought to (a) validate the PPV CNAP -algorithm and (b) investigate the absolute and predictive agreement between PPV CNAP and arterial catheter-derived manually offline calculated PPV (PPV INV ).

This was a prospective method comparison study comparing non-invasive finger-cuff-derived with invasive arterial catheter-derived arterial blood pressure as well as PPV in patients having neurosurgery. Here, we report PPV results. The arterial blood pressure results will be reported separately. The study was approved by the ethics committee (Ethikkomission der Ärztekammer Hamburg, Hamburg, Germany; registration number PV6048) and conducted in operating rooms of the University Medical Center Hamburg-Eppendorf between April and October 2019. All patients provided written informed consent.

We included adult patients (≥ 18 years) who were scheduled for neurosurgery and required invasive arterial blood pressure monitoring using an arterial catheter as part of routine care. We excluded patients with vascular implants at the upper extremities, finger oedema, impairment in peripheral perfusion (e.g., Raynaud syndrome, peripheral artery disease, or arterial-venous shunts), cardiac arrhythmia, valvular heart disease grade 2 or above, excessive movement and/ or seizures, or cardiac assist devices. For this analysis of PPV only patients with appropriate ventilator settings (tidal volume ≥ 8 mL kg −1 predicted body weight, respiratory rate ≥ 10 min −1 , and positive end-expiratory pressure ≤ 5 cm H 2 O) were included.

The PPV CNAP -algorithm is a computer algorithm to detect and analyze ventilation-induced swings in the arterial blood pressure waveform and automatically calculate PPV as illustrated in Fig. 1 . In short, the PPV CNAP -algorithm applies an adapted beat detection algorithm [12] to the arterial blood pressure waveform to obtain systolic and diastolic arterial blood pressure, pulse pressure (PP), and pulse interval (PI). PP and PI are compared to the average PP and PI of previous heart beats. If the difference between a new PP or PI value and their average values of previous heart beats exceeds a certain threshold, the beat could be a premature beat and is therefore excluded from further calculation. The tolerance level is adaptively adjusted by the variance of PP and PI. Next, a PP minimum-maximum detector is applied to the time series of PP values. The detector is made somewhat "fuzzy" to ignore small variations in the PP series. The time appearance of minimum PP (PP min ) and maximum PP (PP max ) undergo a plausibility check by using the average of previous verified PP min and PP max . After verification, PPV is calculated as: PPV = 200 × (PP max -PP min )/(PP max + PP min ) (%).

Note that one PP max /PP min pair corresponds to a half of a respiratory cycle. PPV CNAP is calculated by averaging six PP max /PP min pairs corresponding to the last three respiratory cycles. Additionally, outlier detection is used; PPV values higher than 40% are completely rejected. Further, if the difference between a new PPV value and the previous one is higher than a certain threshold, this new PPV value is used for calculation only with a 50% weight. If the PPV value is confirmed by the next measurement, the PPV value is then considered with a full 100% weight.

If plausibility checks of new beats or new PP min or PP max fail too often, all average and variance variables are reset. The PPV CNAP -algorithm is newly initialized and re-starts the calculation of average and variance variables from scratch.

After induction of general anesthesia, all patients were ventilated with a tidal volume of 8 mL kg −1 predicted body weight, a respiratory rate of ≥ 10 min −1 adjusted to endexpiratory carbon dioxide, and a positive end-expiratory pressure of ≤ 5 cm H 2 O. After insertion of the radial arterial Fig. 1 Schematic illustration of the PPV CNAP -algorithm. SAP systolic arterial blood pressure, DAP diastolic arterial blood pressure, PP pulse pressure, PI pulse interval, PPV pulse pressure variation catheter, the CNAP system's upper-arm-cuff was attached on the ipsilateral arm. The CNAP finger-cuff was placed on the index and middle finger of the contralateral arm. CNAP finger-cuff arterial blood pressure measurements were calibrated to oscillometric arterial blood pressure measurements every 30 min in the first 27 patients. We changed this to the maximal calibration interval of 60 min during the study and calibrated every 60 min in the last 9 patients. Arterial blood pressure recording was started after positioning of the patient in the operating room and continued until the end of surgery. The continuous arterial blood pressure waveforms measured non-invasively with the CNAP system and invasively with the arterial catheter were simultaneously displayed and recorded on the patient monitor (Infinity Delta Monitor; Dräger, Lübeck, Germany). Both waveforms were extracted to a personal computer (eData Data Grabber; Dräger) and beat-to-beat measurements were used for further offline analysis.

We randomly selected 10 60-s episodes of each patient. Within these episodes, we identified a period with at least three visible swings in PP in the non-invasive and invasive arterial waveform, which were used for further analysis.

We calculated PPV INV−CNAP by applying the PPV CNAP -algorithm to the arterial blood pressure waveform recorded invasively using the arterial catheter. PPV CNAP was automatically calculated using the PPV CNAP -algorithm based on the arterial blood pressure waveform recorded non-invasively with the CNAP system. PPV INV was calculated manually from the arterial blood pressure waveform recorded invasively using the arterial catheter (reference method).

Descriptive data are reported as mean ± standard deviation (SD) for continuous data and as absolute frequency and percentage for categorical data.

Using Bland-Altman analysis accounting for repeated measurements within individuals [13, 14] , we compared (a) PPV INV−CNAP and PPV INV to validate the PPV CNAP -algorithm per se and (b) PPV CNAP and PPV INV to investigate the absolute agreement between PPV CNAP and PPV INV . For each comparison, we calculated the mean of the differences between the two methods (test method minus reference method), the SD of the mean of the differences, and the 95% limits of agreement (95%-LoA; i.e., mean of the differences ± 1.96 SD of the mean of the differences) and the 95% confidence intervals (95%-CI) around the 95%-LoA to quantify the trueness and precision of agreement [15, 16] . We additionally describe the correlation between PPV INV−CNAP and PPV INV and between PPV CNAP and PPV INV by Deming regression for scattered plots with 95%-CI [17, 18] . We assessed the predictive agreement for fluid responsiveness across three predefined categories (PPV < 9%, PPV 9 to 13%, PPV > 13%) between PPV CNAP and PPV INV . These PPV categories reflect PPV thresholds used for clinical decision making regarding fluid therapy in clinical practice [19, 20] . The predictive agreement across these three categories was calculated as the number of concordant paired measurements divided by the total number of paired measurements. In addition, we calculated Cohen's kappa [21] . 

A total of 44 patients were available for this analysis, but eight were excluded. We excluded four patients due to cardiac arrhythmia, two patients because of technical failure of the CNAP system, and two patients because of study protocol violations (Fig. 2) . We thus included 36 patients with a total of 360 measurements in the final analysis. Patient characteristics are presented in Table 1 .

The mean of the differences ± SD between PPV INV−CNAP and PPV INV was −0.1 ± 1.2% (95%-LoA −2.5 to 2.3%) (Fig. 3 , Table 2 ). For the comparison between PPV INV−CNAP and (Fig. 3) . The predictive agreement for fluid responsiveness between PPV INV−CNAP and PPV INV was 92% with a Cohen's kappa of 0.79 (Table 3 ).

The mean of the differences ± SD between PPV CNAP and PPV INV was −1.0 ± 2.7% (95%-LoA −6.3 to 4.3%) (Fig. 4 , Table 2 ). The Deming regression for the correlation between PPV CNAP and PPV INV showed a slope of 0.85 (95%-CI 0.78 to 0.91) and an intercept of 0.10 (95%-CI −0.34 to 0.55) (Fig. 4) . The predictive agreement for fluid responsiveness between PPV CNAP and PPV INV was 82% with a Cohen's kappa of 0.48 (Table 4 ).

In this prospective method comparison study, we aimed to validate the PPV CNAP -algorithm and investigate the absolute and predictive agreement between PPV CNAP and PPV INV in patients having neurosurgery.

To validate the PPV CNAP -algorithm per se (independent from waveform recording), we applied the PPV CNAP -algorithm to the arterial blood pressure waveform recorded invasively using an arterial catheter. The absolute agreement-i.e., the trueness and precision of agreement [15, 16] -between PPV INV−CNAP and the manually calculated PPV INV was high. The Deming regression analysis showed a significant correlation between PPV INV−CNAP and PPV INV and the predictive agreement was substantial according to Cohen's kappa. Our results suggest that the PPV CNAP -algorithm reliably calculates PPV and that its measurements are interchangeable with the reference method-the manual offline calculation of PPV-when applied on the same arterial blood pressure waveform. As a next step, we compared PPV CNAP to the reference PPV INV . The absolute agreement between PPV CNAP and PPV INV was lower than between PPV INV−CNAP and PPV INV and the Deming regression indicated a minor proportional difference between the methods. Nonetheless, the predictive agreement between PPV CNAP and PPV INV was moderate according to Cohen's kappa.

In this study, we used arterial catheter-derived manually calculated PPV (PPV INV ) as the reference method. There are no consensus guidelines on how to perform PPV validation studies and interpret their results. Specifically, it remains undefined what constitutes clinically acceptable PPV measurement performance. The absolute agreement between PPV CNAP and PPV INV was similar compared with previous studies evaluating the measurement performance of PPV CNAP in critically ill patients [22, 23] and patients having major open abdominal [24] or vascular surgery [25] . A pilot study in only 10 critically ill patients revealed a mean of the differences between PPV CNAP and arterial catheter-derived manually calculated PPV of −2.1% with 95%-LoA of −8.3 to 4.1% [22] . However, the study also included patients who were ventilated with tidal volumes less than 8 mL kg -1 , which were excluded in our study. In a cohort of 47 critically ill patients with acute circulatory failure, the mean of the differences between PPV CNAP and PPV calculated manually from a femoral arterial blood pressure waveform was −0.6% with 95%-LoA of −6.3 to 5.2% [23] . The authors excluded 17% of patients because the CNAP system was unable to properly record the noninvasive arterial blood pressure waveform [23] . In contrast, we were able to record an arterial blood pressure waveform with the CNAP system in all patients. Our results are in line with a previous study in 35 patients having vascular surgery which showed similar moderate absolute agreement between PPV CNAP and arterial catheter-derived manually calculated PPV before and after volume expansion [25] . Even though our results are in line with previous findings, it is challenging to interpret the absolute agreement of PPV CNAP with PPV INV as no clearly defined thresholds for clinically acceptable PPV differences exist.

When investigating non-invasively measured dynamic cardiac preload variables, their predictive capabilities regarding the prediction of fluid responsiveness may even be more important than absolute agreement with invasive reference measurements. In the before-mentioned study in vascular surgery patients, volume expansion was performed to investigate the ability of PPV CNAP to predict fluid responsiveness. PPV CNAP predicted fluid responsiveness very well according to receiver operating characteristics curve analysis [25] . This was also shown in other studies directly testing the capabilities of PPV CNAP to predict fluid responsiveness, i.e., an increase in cardiac output after a fluid challenge. PPV CNAP and PPV calculated from an invasive arterial blood pressure waveform seem to have similar predictive value [23, 24] . We did not perform a fluid challenge or passive legraising test to directly test how well PPV CNAP predicts fluid responsiveness. Instead, we categorized PPV measurements considering clinical decision making based on predefined PPV thresholds for the prediction of fluid responsiveness [19] . PPV CNAP measurements falling in the same category as the respective PPV INV values would subsequently lead to the same decision regarding fluid therapy. The predictive agreement between PPV CNAP and PPV INV across the three categories was over 90% and Cohen's kappa indicated a substantial predictive agreement. In line with the results of Bland-Altman analysis, the predictive agreement between PPV CNAP and PPV INV was slightly lower, but still over 80% and Cohen's kappa indicated moderate agreement.

We did not perform preload-changing interventions such as a fluid challenge or passive leg-raising test to assess fluid responsiveness. Nevertheless, we analyzed the agreement between the test and the reference method stratified by different PPV categories according to clinically established [19] . Data pairs were selected randomly, but data selection bias cannot be ruled out definitely. We did not perform an a priori sample size calculation. Narrow 95%-CI around the 95%-LoA of the means of the differences between PPV INV−CNAP and PPV INV as well as PPV CNAP and PPV INV suggest that the sample size was sufficient though. Additionally, the change of the calibration interval for the CNAP system during the study may have affected the results. We only included patients having neurosurgery and the results can thus not be generalized to other-especially critically ill-patients.

In conclusion, the PPV CNAP -algorithm reliably calculates PPV compared to manual offline PPV calculation when applied on the same arterial blood pressure waveform. The absolute and predictive agreement between PPV CNAP and PPV INV are moderate.

Funding The study was supported by an institutional restricted research grant from CNSystems Medizintechnik (Graz, Austria). Open Access funding enabled and organized by Projekt DEAL.

Conflict of interest MF has received honoraria for consulting from CNSystems Medizintechnik (Graz, Austria). PH, JMM, LB, KK, and GG have no conflict of interest to declare. JF is co-founder and CEO of CNSystems Medizintechnik (Graz, Austria). BS has received institutional restricted research grants, honoraria for giving lectures, and refunds of travel expenses from CNSystems Medizintechnik (Graz, Austria). BS has received honoraria for consulting, honoraria for giving lectures, and refunds of travel expenses from Edwards Lifesciences (Irvine, CA, USA). BS has received honoraria for consulting, institutional restricted research grants, honoraria for giving lectures, and refunds of travel expenses from Pulsion Medical Systems (Feldkirchen, Germany). BS has received institutional restricted research grants from Retia Medical (Valhalla, NY, USA). BS has received honoraria for giving lectures from Philips Medizin Systeme Böblingen (Böblingen, Germany). BS has received honoraria for consulting, institutional restricted research grants, and refunds of travel expenses from Tensys Medical (San Diego, CA, USA).

The study was approved by the ethics committee (Ethikkomission der Ärztekammer Hamburg, Hamburg, Germany; registration number PV6048).

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.

Prediction of fluid responsiveness: an update

Arterial pulse pressure variation with mechanical ventilation

Clinical use of respiratory changes in arterial pulse pressure to monitor the hemodynamic effects of PEEP

Relation between respiratory changes in arterial pulse pressure and fluid responsiveness in septic patients with acute circulatory failure

Measurement of blood pressure

Techniques for non-invasive monitoring of arterial blood pressure

Non-invasive arterial pressure monitoring revisited

Less invasive hemodynamic monitoring in critically ill patients

Non-invasive measurement of pulse pressure variation using a finger-cuff method in obese patients having laparoscopic bariatric surgery

Non-invasive continuous arterial pressure and pulse pressure variation measured with Nexfin® in patients following major upper abdominal surgery: a comparative study

Continuous noninvasive pulse wave analysis using finger cuff technologies for arterial blood pressure and cardiac output monitoring in perioperative and intensive care medicine: a systematic review and meta-analysis

An open-source algorithm to detect onset of arterial blood pressure pulses. Computers in Cardiology

Statistics notes: variables and parameters

Agreement between methods of measurement with multiple observations per individual

Metrology part 1: definition of quality criteria

Metrology part 2: procedures for the validation of major measurement quality criteria and measuring instrument properties

Evaluation of regression procedures for methods comparison studies

General deming regression for estimating systematic bias and its confidence interval in method-comparison studies

Assessing the diagnostic accuracy of pulse pressure variations for the prediction of fluid responsiveness: a "gray zone" approach

Monitoring of pulse pressure variation using a new smartphone application (Capstesia) versus stroke volume variation using an uncalibrated pulse wave analysis monitor: a clinical decision making study during major abdominal surgery

The measurement of observer agreement for categorical data

Evaluation of a novel automated non-invasive pulse pressure variation algorithm

Prediction of fluid responsiveness by a continuous non-invasive assessment of arterial pressure in critically ill patients: comparison with four other dynamic indices

Non-invasive assessment of fluid responsiveness using CNAP technology is interchangeable with invasive arterial measurements during major open abdominal surgery

Automated, continuous and non-invasive assessment of pulse pressure variations using CNAP((R)) system