Michael Mestek’s scientific contributions

In Response We have adapted Bland and Altman’s analysis only in that we recognize that one of our measurements is from a technology widely used in clinical practice.1,2 In our experience, readers are more interested in reviewing differences with respect to a well-established device and that averaging observations from what are essentially a reference and a device under test adds nothing to their understanding. We take this opportunity to clarify that Bland and Altman never use the phrase “confidence interval” in the article referenced by Dr Drummond. They do, however, provide a figure remarkably similar to our Figure 2. We are disappointed that, again, Dr Drummond failed to mention that we qualified and expanded on this analysis by presenting data on subject demographics (Table 1), medical conditions (Table 2), bootstrapped delta with respect to the reference (including 95% limits of agreement; Table 3), bootstrapped concordance and correlation coefficients (Table 4) and statistics on the nonstationary nature of the collected data (Table 4). In all of these cases, the reported statistics were presented separately for volunteers and patients, and then in combination. We can only assume this detail was inconvenient to the message he wished to convey. For at least the past 10 years, it has been well reported that limits of agreement may vary with respect to the sample used in their calculation.3 Like Bland and Altman, earlier authors note that “the limits by themselves provide only a reference interval and should never be used as the determining factor to conclude agreement between two devices.”3 Leaving aside the fact that our analysis was part of a wider statistical treatment, we use the limits precisely as the reference interval for which it is intended. No conclusion of agreement was drawn from this interval. We note with interest that the authors of the articles Dr Drummond cites for describing the additional statistical analysis required4 and for highlighting the underreporting of this analysis5 contain a common member, himself. While we admire this personal campaign, we must point out that providing the confidence interval of the limits of agreement of the statistic in question is beneficial primarily if we want to quantify confidence in conclusions that are being drawn from that data. If we are presenting the limits of agreement as a simple reference interval, then there is a point of diminishing return in providing this statistic of a statistic of some statistics. We disagree that the limits of agreement should answer the question “are these measurement systems equivalent, can I use them interchangeably?” We do not believe that any single metric can answer that question (with or without confidence intervals). We do not, however, see any conflict between his assertion that there is insufficient information in our article to declare whether the systems can be used interchangeably and our conclusions that the technology “may be a useful adjunct to continuous pulse oximetry monitoring” and that that its use “warrants assessment.”

Background: Intermittent measurement of respiratory rate via observation is routine in many patient care settings. This approach has several inherent limitations that diminish the clinical utility of these measurements because it is intermittent, susceptible to human error, and requires clinical resources. As an alternative, a software application that derives continuous respiratory rate measurement from a standard pulse oximeter has been developed. We sought to determine the performance characteristics of this new technology by comparison with clinician-reviewed capnography waveforms in both healthy subjects and hospitalized patients in a low-acuity care setting. Methods: Two independent observational studies were conducted to validate the performance of the Medtronic Nellcor Respiration Rate Software application. One study enrolled 26 healthy volunteer subjects in a clinical laboratory, and a second multicenter study enrolled 53 hospitalized patients. During a 30-minute study period taking place while participants were breathing spontaneously, pulse oximeter and nasal/oral capnography waveforms were collected. Pulse oximeter waveforms were processed to determine respiratory rate via the Medtronic Nellcor Respiration Rate Software. Capnography waveforms reviewed by a clinician were used to determine the reference respiratory rate. Results: A total of 23,243 paired observations between the pulse oximeter-derived respiratory rate and the capnography reference method were collected and examined. The mean reference-based respiratory rate was 15.3 ± 4.3 breaths per minute with a range of 4 to 34 breaths per minute. The Pearson correlation coefficient between the Medtronic Nellcor Respiration Rate Software values and the capnography reference respiratory rate is reported as a linear correlation, R, as 0.92 ± 0.02 (P < .001), whereas Lin's concordance correlation coefficient indicates an overall agreement of 0.85 ± 0.04 (95% confidence interval [CI] +0.76; +0.93) (healthy volunteers: 0.94 ± 0.02 [95% CI +0.91; +0.97]; hospitalized patients: 0.80 ± 0.06 [95% CI +0.68; +0.92]). The mean bias of the Medtronic Nellcor Respiration Rate Software was 0.18 breaths per minute with a precision (SD) of 1.65 breaths per minute (healthy volunteers: 0.37 ± 0.78 [95% limits of agreement: -1.16; +1.90] breaths per minute; hospitalized patients: 0.07 ± 1.99 [95% limits of agreement: -3.84; +3.97] breaths per minute). The root mean square deviation was 1.35 breaths per minute (healthy volunteers: 0.81; hospitalized patients: 1.60). Conclusions: These data demonstrate the performance of the Medtronic Nellcor Respiration Rate Software in healthy subjects and patients hospitalized in a low-acuity care setting when compared with clinician-reviewed capnography. The observed performance of this technology suggests that it may be a useful adjunct to continuous pulse oximetry monitoring by providing continuous respiratory rate measurements. The potential patient safety benefit of using combined continuous pulse oximetry and respiratory rate monitoring warrants assessment.This is an open-access article distributed under the terms of the Creative Commons Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download and share the work provided it is properly cited. The work cannot be changed in any way or used commercially.

Respiratory rate is recognized as a clinically important parameter for monitoring respiratory status on the general care floor (GCF). Currently, intermittent manual assessment of respiratory rate is the standard of care on the GCF. This technique has several clinically-relevant shortcomings, including the following: (1) it is not a continuous measurement, (2) it is prone to observer error, and (3) it is inefficient for the clinical staff. We report here on an algorithm designed to meet clinical needs by providing respiratory rate through a standard pulse oximeter. Finger photoplethysmograms were collected from a cohort of 63 GCF patients monitored during free breathing over a 25-min period. These were processed using a novel in-house algorithm based on continuous wavelet-transform technology within an infrastructure incorporating confidence-based averaging and logical decision-making processes. The computed oximeter respiratory rates (RRoxi) were compared to an end-tidal CO2 reference rate (RRETCO2). RRETCO2 ranged from a lowest recorded value of 4.7 breaths per minute (brpm) to a highest value of 32.0 brpm. The mean respiratory rate was 16.3 brpm with standard deviation of 4.7 brpm. Excellent agreement was found between RRoxi and RRETCO2, with a mean difference of -0.48 brpm and standard deviation of 1.77 brpm. These data demonstrate that our novel respiratory rate algorithm is a potentially viable method of monitoring respiratory rate in GCF patients. This technology provides the means to facilitate continuous monitoring of respiratory rate, coupled with arterial oxygen saturation and pulse rate, using a single non-invasive sensor in low acuity settings.

SESSION TYPE: COPD Posters IIPRESENTED ON: Wednesday, October 24, 2012 at 01:30 PM - 02:30 PMPURPOSE: A new method of analyzing a pulse oximeter waveform (RRoxi algorithm) allows continuous non-invasive respiratory rate monitoring. The RRoxi algorithm uses respiratory-induced modulations in the photoplethysmogram secondary to changes in hemodynamic and autonomic activity throughout the respiratory cycle. Patients with chronic obstructive pulmonary disease (COPD) may exhibit abnormal cardiorespiratory interactions that could influence the modulations used by RRoxi to extract respiratory rate. We therefore sought to determine the accuracy of RRoxi in patients with COPD.METHODS: With IRB approval and patient consent, we enrolled 22 non-intubated patients diagnosed with COPD (age: 61.8 ± 11.8 yr; BMI: 28.6 ± 7.7 kg/m2, 4 female/ 18 male). During an observational study period of ~20 min on the general care floor or in the ICU, photoplethysmogram data were acquired from an index finger using a pulse oximetry sensor. Nasal/oral capnography waveforms were acquired simultaneously to determine a reference respiratory rate. Respiratory rate values were computed offline by RRoxi. Accuracy was calculated as root mean square deviation (RMSD) between RRoxi and the reference respiratory rate, and Pearson correlation coefficients were computed to assess their relationship.RESULTS: RRoxi determined a respiratory rate value 93% of the monitoring period, yielding a total of 4,917 paired observations (RRoxi values and the capnography-based reference). Mean and range of the reference respiratory rate were 17.6 ± 4.4 and 8.1 to 32 breaths per minute, respectively. The mean difference between the reference and RRoxi respiratory rate measurements was 0.7 ± 1.6 breaths per minute. The accuracy of RRoxi was 1.7 breaths per minute, as measured by RMSD and the agreement between RRoxi and the reference was R2=0.87.CONCLUSIONS: These results demonstrate that in patients with COPD, the RRoxi algorithm was accurate to within 1.7 breaths per minute (RMSD) and able to determine a respiratory rate value during 93% of the monitoring period.CLINICAL IMPLICATIONS: These accuracy and availability results suggest that RRoxi should be clinically acceptable to provide continuous non-invasive respiratory rate monitoring in similar hospitalized patients.DISCLOSURE: Michael Mestek: Employee: Covidien Respiratory & Monitoring Solutions ScientistPaul Addison: Employee: Covidien Respiratory & Monitoring Solutions ScientistAnna-Maria Neitenbach: Employee: Covidien Respiratory & Monitoring Solutions Research Associate.Sergio Bergese: Other: Dr. Bergese received financial funding for data collection on hospitalized patients for this study.Scott Kelley: Employee: Dr. Kelley is the Chief Medical Officer at Covidien Respiratory & Monitoring SolutionsNo Product/Research Disclosure InformationCovidien, Boulder, CO.

SESSION TYPE: Heart Failure PostersPRESENTED ON: Wednesday, October 24, 2012 at 01:30 PM - 02:30 PMPURPOSE: A recently developed algorithm (RRoxi) provides continuous non-invasive respiratory rate from a pulse oximeter waveform. This approach is based on modulations in the photoplehtysmogram secondary to hemodynamic and autonomic changes during the respiratory cycle. Medical conditions characterized by abnormal cardiorespiratory interactions, such as congestive heart failure (CHF), may influence the fundamental modulations used by RRoxi to extract respiratory rate. We sought to determine the accuracy of RRoxi in patients with known CHF.METHODS: With IRB approval and patient consent, we studied 12 hospitalized patients with diagnosed CHF (left ventricular ejection fraction < 30%; age: 60 ± 14.7 yr; BMI: 30 ± 5.9 kg/m2, 1 female/11 male). During an observational monitoring period of ~20 min, photoplethysmogram data were acquired from an index finger using a pulse oximetry sensor. Nasal/oral capnography waveforms were acquired simultaneously to determine a reference respiratory rate. Respiratory rate values were computed offline by RRoxi. Accuracy was calculated as root mean square deviation (RMSD) between RRoxi and the reference respiratory rate and Pearson correlation coefficients were computed to assess their relationship.RESULTS: During the study period, RRoxi computed a respiratory rate value 93% of the time patients were monitored, yielding a total of 2,738 paired observations (RRoxi values and the capnography-base reference respiratory rate). Mean and range of the reference respiratory rate were 16.7 ± 3.0 and 10.7 to 26.1 breaths per minute, respectively. The mean difference between the measurements was 0.4 ± 1.5 breaths per minute. The accuracy of RRoxi was 1.6 breaths per minute, as measured by RMSD. The agreement between RRoxi and the reference was R2=0.76.CONCLUSIONS: These results demonstrate that the RRoxi algorithm is accurate to within 1.6 breaths per minutes (RMSD) in patients with CHF and determined a respiratory rate during 93% of the monitoring period.CLINICAL IMPLICATIONS: These accuracy results for RRoxi suggest that this algorithm should be clinically acceptable to provide continuous non-invasive respiratory rate monitoring in similar patients.DISCLOSURE: Michael Mestek: Employee: Covidien Respiratory & Monitoring Soultions ScientistPaul Addison: Employee: Covidien Respiratory & Monitoring Solutions ScientistAnna-Maria Neitenbach: Employee: Covidien Respiratory & Monitoring Solutions Research AssociateSergio Bergese: Other: Dr. Bergese received financial funding to complete data collection for the study.Scott Kelley: Employee: Dr. Kelley is the Chief Medical Officer of Covidien Respiratory & Monitoring SoultionsNo Product/Research Disclosure InformationCovidien, Boulder, CO.

The presence of respiratory information within the pulse oximeter signal (PPG) is a well-documented phenomenon. However, extracting this information for the purpose of continuously monitoring respiratory rate requires: (1) the recognition of the multi-faceted manifestations of respiratory modulation components within the PPG and the complex interactions among them; (2) the implementation of appropriate advanced signal processing techniques to take full advantage of this information; and (3) the post-processing infrastructure to deliver a clinically useful reported respiratory rate to the end user. A holistic algorithmic approach to the problem is therefore required. We have developed the RR(OXI) algorithm based on this principle and its performance on healthy subject trial data is described herein. Finger PPGs were collected from a cohort of 139 healthy adult volunteers monitored during free breathing over an 8-min period. These were subsequently processed using a novel in-house algorithm based on continuous wavelet transform technology within an infrastructure incorporating weighted averaging and logical decision making processes. The computed oximeter respiratory rates (RR(oxi)) were then compared to an end-tidal CO2 reference rate RR(ETCO2). RR(ETCO2) ranged from a lowest recorded value of 2.97 breaths per min (br/min) to a highest value of 28.02 br/min. The mean rate was 14.49 br/min with standard deviation of 4.36 br/min. Excellent agreement was found between RR(oxi) and RR(ETCO2), with a mean difference of -0.23 br/min and standard deviation of 1.14 br/min. The two measures are tightly spread around the line of agreement with a strong correlation observable between them (R2 = 0.93). These data indicate that RR(oxi) represents a viable technology for the measurement of respiratory rate of healthy individuals.

In deriving limits of agreement by the methods discussed in Bland and Altman, 1 we followed recommendations from the US Food and Drug Administration guidance for pulse oxim-etry equivalence testing 2 and applied the described correction for multiple observations per individual. 1 In employing a modified Bland-Altman diagram, we recognized that 1 of the 2 signals collected (the Etco 2-based respiration rate) is a well-known method of measurement commonly used in clinical practice. We therefore use this signal as a reference. We thank the independent reviewers of our article and the Statistical Editor of Anesthesia & Analgesia for suggesting that we present bootstrapped data for both hospital and laboratory subjects together and independently. For a technology at this stage, our approach is, we believe, far more rigorous than the majority of publications in this area. Regrettably, Dr Drummond chose not to comment on the characteristics of the observed data in his analysis. While technically not independent, the reported parameters are not a simple averaging of data and are highly nonstationary through time. They can change substantially over a 1-minute period (as can the effective sample rate—the cardiac pulse rate—and the pulse morphology). The purpose of this study was to indicate how these changing values compare to an accepted reference. Dr Drummond also failed to comment on the bootstrapping analysis of a large number of additional statistics, something he makes great effort to recommend elsewhere. 3 The substantial effect Dr Drummond describes will manifest only if the signal differences are completely random. An expected " RMSD [root mean square deviation]… increase " of 3.16 is a theoretical increase in random noise resulting from a 10-fold downsampling of a signal—not something that we would expect in this case and, as RMSD includes a systemic error component, not something that is technically correct either. If the RMSD has a significant contribution from the mean bias, or a variation in local biases across subjects, then we would expect little or no change to the RMSD metric. Indeed, if Dr Drummond's concern is that both signals are significantly autocorrelated through time, then he should surely also expect the delta between them to be autocorrelated. If this is the case, then modeling this delta as a wholly random error would not seem appropriate. We do not use the term " gold standard " anywhere in our article. It is inappropriate for Dr Drummond to use this term in quotation marks or indeed suggest its use in our article. We described the clinician reviewed Etco 2 as a " reference. " It is used as a reference because it is a well understood and accepted measurement of respiration rate validated by qualified clinicians. Indicating performance relative to an established reference underpins the regulatory process used to clear many of the medical devices Dr Drummond has himself used over his long career. To present the potential differences between these 2 measurements as an issue is rather to miss the whole point of the study. REFERENCES 1. Bland JM, Altman DG. Agreement between methods of measurement with multiple observations per individual. J Biopharm Stat. 2007;