Retrospective Analysis of Quality Control Data Using Pooled Blood to Compare Agreement within Two Models of Blood Gas Analyzers

Abstract

Background

Two previous reports from the same group concluded that the analytical reliability of Instrumentation Laboratory GEM4000 analyzers (GEM4K) deteriorated during a 24-hour period, based on results between samples from the same patient but collected at different times. Our routine blood gas Between-Laboratory Survey is done every 2 weeks using a freshly pooled heparinized blood sample taken to each analyzer location to verify comparability among our GEM4K and Radiometer ABL90 (Rad90) blood gas analyzers. Because another report found a few very large differences in glucose results between the GEM4K and central laboratory analyzers, we reviewed the glucose results on plasma from our Chemistry Between-Laboratory Surveys that includes comparisons between our central laboratory analyzers (Beckman DxC800; DxC800) and our GEM4K and Rad90 blood gas analyzers.

Method

Using data from our Blood Gas and Chemistry Surveys over a 55-week period, we calculated the mean, standard deviation (SD), and concentration intervals of the 27 sets of results by the GEM4K or Rad90 analyzers. Agreement in plasma glucose between DxC800 and blood gas analyzers was evaluated by the limits of agreement and intraclass correlation coefficient analysis.

Results

For each analyte, the Rad90 had lower average SD than the GEM4K for the 55-week period, although both brands of analyzers showed acceptable performance. For plasma glucose results on our Chemistry Survey, the GEM4K results agreed more closely with the DxC800 results than the Rad90 results.

Conclusions

Based on both our Blood Gas and Chemistry Surveys, we conclude that both brands of analyzers performed within analytically and clinically acceptable limits throughout the year, with no evidence for the type of errors reported previously.

Impact Statement

Our report evaluates quality control data on fresh pooled heparinized blood or plasma collected over a 55-week period. Our findings contradict reports from other authors regarding the imprecision and variability of a blood gas analyzer.

Introduction

Cembrowski et al. (1) compared a patient’s results that were obtained within 2 hours of each collection and analyzed by the same blood gas/electrolyte analyzers, using either the GEM4K or the Radiometer ABL800 analyzers. Among the conclusions was that the paired results from the GEM analyzers showed better agreement during the period of 2:00 AM to 2:00 PM than during the period of 2:00 PM to 2:00 AM. While their comparisons were on samples from the same patient, the samples were collected at approximately 2-h time intervals that varied throughout the day. A follow-up study published as an abstract (2), compared GEM4K results by point-of-care (POC) analyzers to GEM4K results by laboratory analyzers collected on the same patient within 30 min of each other. They reported reduced accuracy of the GEM4K in nonmorning hours, which was presumably 6–8 hours postcalibration of internal Process Control Solution C. However, the conclusions of the original study (1) were disputed in a letter to the editor that included one author employed by Instrumentation Laboratory (IL) (3). Because their comparisons between results on the same patient differed by up to 2 h and involved critical care analytes that are subject to rapid and significant physiologic changes, these comparisons could have some component of physiologic changes that would contribute to the presumed imprecision. Furthermore, the GEM4K uses other internal control solutions to monitor analyzer performance with every sample tested throughout the day, which should ensure analyzer stability over a longer period of time.

At our hospital, to comply with College of American Pathology (CAP) regulation COM.04250, we have a program that has been in routine use for over 25 years to compare blood gas and electrolyte results among our numerous GEM4K and Rad90 analyzers located both in near-patient laboratories and at POC locations. Every 2 weeks, a technologist prepares a pool of leftover human heparinized blood and takes multiple syringes of this pooled blood to 5 GEM4K analyzers located in a stat laboratory, to 3 Rad90 analyzers located in a separate clinical pediatric laboratory, and to several similar analyzers at POC locations.

To study the between-analyzer agreement for each type of analyzer, we evaluated results from our quality assessment program over a 1-year period. To make the comparisons equivalent, we analyzed data from 3 GEM4K laboratory analyzers, 3 Rad90 laboratory analyzers, 1 POC GEM4K analyzer in a pediatric cardiac catheterization laboratory, and 1 POC Rad90 analyzer in a pediatric cardiac intensive-care-unit (PCICU). We felt this study had the advantage of utilizing virtually identical pooled fresh human blood specimens having a nearly ideal sample matrix that were tested independently by all analyzers over a 90-min period.

In an evaluation of the accuracy of blood glucose results by the IL GEM 3000 and 4000 analyzers, Liang et al. (4) utilized data from different specimens collected within 5 min of each other, with one sample analyzed as plasma by a central laboratory analyzer and the other analyzed as heparinized blood by blood gas (BG) analyzers. They concluded that these BG analyzers did not meet the requirements of the 2016 FDA guidance. We note that, of their 2671 pairs of results, approximately 11 differed by over 100 and up to 270 mg/dL, which are greater analytical differences than expected for identical samples analyzed by chemistry and BG analyzers. This suggests at least some of these specimens had significantly different glucose concentrations, perhaps caused by food intake, or administration of glucose or insulin.

Another report (5) compared glucose results between different specimens collected relatively close together. This study concluded that the agreement between the GEM4K and the central laboratory results was acceptable between 2:00 AM and 8:00 AM, but deteriorated in the 6 h after 8:00 AM, noting many differences of >40 mg/dL glucose.

To evaluate the differences between glucose results on our BG analyzers compared to our chemistry analyzers, we reviewed the between-laboratory glucose results on pooled plasma samples analyzed for our routine Chemistry Surveys done every 2 weeks that includes both our central laboratory analyzers (DxC800) and our BG analyzers (GEM 4K and Rad 90).

Materials and Methods

Study Materials

As part of our routine between-laboratory (BL) quality control survey to verify comparability of instruments doing the same testing, per CAP regulation COM.04250, every 2 weeks we pool heparinized BG samples destined for discard that remain after testing and reporting. Leftover blood samples are randomly selected based on volume remaining after testing, with no effort made to select any concentrations of analyte. Before testing, all pooled blood is checked for hemolysis by centrifugation. Twenty-five mL of blood is pooled into a plastic medicine cup, mixed, then allowed to stand for about 30 min. This pooled blood is remixed, then aliquoted into 7 3-mL syringes, which are sealed then laid on ice as they are taken on a rotating basis to each laboratory for analysis. The results are compiled and a report is filled out and sent to each laboratory for review by the laboratory manager and director.

Instruments

The GEM4K analyzer (Instrumentation Laboratory) is a whole blood analyzer that measures pH, pCO₂, pO₂, Na⁺, K⁺, Ca²⁺, glucose, lactate, and total hemoglobin (Hb) and several forms of Hb including oxy-Hb, carboxy-Hb, and met-Hb. As part of its quality control, it utilizes 4 process control solutions (PCS) to monitor stability of the analytical responses. PCS A and B are run at least every 4 h and after every sample. PCS C is run once per 24 h, and PCS D is run every 12 h. The ages of these analyzers were 1 year for the laboratory analyzers and 4 years for the POC analyzer.

The Rad90 analyzer (Radiometer America) analyzes whole blood for pH, pCO₂, pO₂, Na⁺, K⁺, Ca²⁺, Cl⁻, glucose, lactate, Hb, and Hb forms by oximetry. This analyzer utilizes a combination of solid-state amperometric and potentiometric ion-selective sensors, optical pO₂, and spectrophotometric oximetric measurements. To compensate for drift, a calibrant solution is run to bracket every test measurement, and every 8 h, 1 level of each of 3 levels of liquid internal QC solutions is automatically run. The Rad90 laboratory analyzers have been in routine use for 3 years and the POC analyzer for 2 years.

Institutional Review of Protocol

Our study protocol was reviewed by the Duke University Health System Institutional Review Board and was declared exempt from further review as it did not meet the definition of human research.

Study Protocol

Typically, between the hours of 7:00 PM and 10:00 PM on our evening shift, we prepare and carry these samples in syringes to the various laboratory and POC locations for analysis. All samples are laid on ice to minimize metabolic changes as they are taken to the various locations, with all samples typically analyzed within 90 min on 4 GEM4K and 4 Rad90 analyzers.

Using data collected over a 55-week period, we calculated the daily mean and standard deviations (SD) of each set of results by either the GEM4K or Rad90 analyzers. To evaluate the variation within each type of analyzer, we then calculated the mean and SD of each set of 27 daily SDs obtained on each type of analyzer over the 55-week period.

For routine evaluation of our quality control results, we calculate the mean and SD of results for each analyte by all analyzers, then compare these to acceptable ranges, which are based on, but more rigorous than, CLSI guidelines (6). However, for this study we were concerned with the daily SDs for either the 4 results by the GEM4K analyzers or the 4 results by the Rad90 analyzers. Because this study was not designed to assign a “true” value to either set of results, we gave little attention to the agreement between each brand of analyzer. However, as noted later, all results but 1 were within acceptable limits for our routine quality evaluations.

Our Chemistry Survey is done approximately every 2 weeks to verify BL agreement of our chemistry tests. At each survey, 2 preparations of pooled heparinized plasma that remain after testing is complete are used. These samples are prepared, then taken to the analyzers in various laboratories at our medical center. We reviewed the plasma glucose results from our Chemistry Surveys that included results from our central laboratory analyzers (DxC800) and our BG analyzers (GEM4K and Rad90).

Statistical Methods

Concentration of analytes as measured by GEM4K and Rad90 are summarized using the mean, SD, and interval (the smallest to the largest number). To assess the difference between GEM4K and Rad90 measurements, a linear mixed-effects model was used to account for repeated measurements made on the same sample. Instrument precision was evaluated by assessing the coefficient of variation (CV) and the daily variability with results presented using the mean (SD) of the daily standard deviations and the mean (SD) of the daily range (the difference between the largest and the smallest number). To examine the agreement in plasma glucose measurements between the Beckman DxC and BG analyzers, the intraclass correlation coefficient (ICC) and its 95% confidence intervals (CI) were calculated (7). The limits of agreement (LOA) were then used to evaluate the mean bias and variability (mean ± 2SD). The threshold for assessing statistical significance was set to α = 0.05. No adjustment for multiple testing was performed. All analyses were done using SAS v.9.4 (SAS Institute, Inc.).

Results

The mean, SD, and intervals for each analyte are shown in Table 1. No statistical difference was determined in measurements between GEM4K and Rad90 for any analyte except Na. We observed that the daily means of Na results were always higher by 2–4 mmol/L (mean difference = 3.2 mmol/L) for the Rad90 compared to the GEM4K. This is apparently due to the manufacturers using different set points for their sodium calibrators. Because the samples were not preselected for any concentration of analyte, the measurement intervals mostly cover a range of concentrations or values expected for the population that has such tests ordered. As examples, no pH values were in the alkaline range; no K results were in the hypokalemic range; no ionized Ca results were in the hypercalcemic range; all lactate results were elevated; and no glucose or total Hb values were either extremely low or high. The CV for each analyte was also calculated and the overall precision of the 2 analyzers is comparable.

Table 1.

Concentration of analytes measured by GEM4K and Rad90.

Analyte	GEM4K (n = 107)			Rad90 (n = 108)			LSMdc	P valuec
	Mean (SD)a	CV	Intervalb	Mean (SD)a	CV	Intervalb
pH	7.34 (0.05)	0.7%	7.25–7.45	7.34 (0.05)	0.7%	7.25–7.43	0.00	0.787
pCO2 (mmHg)	37.1 (5.60)	15.0%	22.0–49.0	38.9 (6.80)	17.5%	24.0–64.0	−1.77	0.304
pO2 (mmHg)	76.1 (22.7)	29.9%	36.0–135	80.1 (23.6)	29.4%	41.0–143	−4.08	0.523
Na (mmol/L)	143.3 (2.0)	1.6%	137–150	146.5 (2.0)	1.6%	141–152	−3.21	<0.001
K (mmol/L)	5.09 (0.91)	17.9%	4.20–7.80	5.05 (0.90)	17.8%	4.20–7.80	0.03	0.893
ion Ca (mmol/L)	1.13 (0.04)	3.5%	1.01–1.22	1.13 (0.03)	3.0%	1.05–1.19	0.00	0.739
Glucose (mg/dL)	99.0 (18.3)	18.5%	62.0–152	99.4 (17.1)	17.2%	69.0–149	−0.15	0.976
Lactate (mmol/L)	6.71 (1.54)	22.9%	2.30–9.50	6.99 (1.29)	18.5%	4.60–9.30	−0.27	0.487
Hb (g/dL)	11.1 (0.80)	7.5%	9.70–13.1	11.5 (0.90)	7.6%	9.80–13.2	−0.34	0.152
%O2Hb	91.6 (4.50)	4.9%	76.1–96.4	91.7 (4.80)	5.3%	75.4–97.5	−0.16	0.901

Data based on 27 sets of results (3 or 4 results per set) from each instrument on each day of QC testing from March 7, 2018 to March 21, 2019.

mmHg × 0.1333 = kPa.

CV = coefficient of variation.

GEM4K = Instrumentation Laboratory GEM 4000 Blood Gas Analyzer.

Rad90 = Radiometer ABL90 Blood Gas Analyzer.

^aMean (SD): the overall mean and standard deviation of 107 measurements from 4 GEM4K analyzers and 108 measurements from 4 Rad90 analyzers between March 7, 2018, and March 21, 2019.

^bInterval: the smallest to the largest of the 107 (or 108) results.

^cDifferences of least squares means of analyte measurements (LSMd) between GEM4K and RAD90 were estimated using linear mixed-effects models to account for repeated measurements within the same sample. Unadjusted P value of statistical test of the difference is reported.

For each day of the study, we calculated the mean and SD of each set of 4 results by either the GEM4K or Rad90 analyzers. This gave 27 sets of SDs for each analyte by each brand of analyzer. To evaluate the overall variation within each type of analyzer, we then calculated the mean and SD of each set of these 27 daily SDs obtained on each type of analyzer for each analyte over the 12-month period. The results of this analysis are shown in Table 2. For each analyte, the Rad90 had a lower average SD for the 12-month period, which shows superior analytical performance by the Rad90, although both types of analyzers had acceptable performance throughout the year.

Table 2.

Instrument precision within GEM 4000 or Radiometer 90 analyzers (n = 27 sets^*).

Analyte	GEM4K (n = 27)		Rad90 (n = 27)
	Mean and (SD) of daily SDsa	Interval of differencesb	Mean and (SD) of daily SDsa	Interval of differencesb
pH	0.009 (0.004)	0–0.04	0.005 (0.004)	0–0.03
pCO2 (mmHg)	0.78 (0.551)	0–5.0	0.483 (0.374)	0–3.0
pO2 (mmHg)	2.39 (0.96)	2.0–11.0	1.74 (1.73)	1.0–18.0
Na (mmol/L)	0.822 (0.305)	1.0–4.0	0.453 (0.451)	0–4.0
K (mmol/L)	0.066 (0.035)	0–0.30	0.038 (0.043)	0–0.40
Ionized Ca (mmol/L)	0.015 (0.007)	0.01–0.07	0.008 (0.004)	0–0.04
Glucose (mg/dL)	2.88 (1.15)	3.0–11.0	1.26 (0.91)	0–8.0
Lactate (mmol/L)	0.164 (0.053)	0.10–0.60	0.131 (0.053)	0.10–0.50
Hb (g/dL)	0.169 (0.088)	0–0.90	0.134 (0.081)	0.10–0.80
%O2Hb	0.765 (0.393)	0.60–5.10	0.447 (0.382)	0.20–3.0

^*n = 27 sets of results on each day of QC testing from March 7, 2018, to March 21, 2019, were used for analysis. mmHg × 0.1333 = kPa.

GEM4K = Instrumentation Laboratory GEM 4000 Blood Gas Analyzer.

Rad90 = Radiometer ABL90 Blood Gas Analyzer.

^aMean (SD) of daily SDs: this column is the calculated means of the 27 sets of daily SDs of the 4 (or 3) measurements from the 4 instruments. The variations of these 27 daily SDs are shown in parentheses as (SD).

^bInterval of differences: this shows the interval of the daily ranges. The daily ranges are the differences between the largest and the smallest result for each analyte on each day of testing.

In Table 3, we show the ranges of acceptable differences for each analyte, which are more rigorous than CLSI guidelines (6), and the number of control results that were out of our acceptable ranges. This shows that nearly all differences by both analyzers were within acceptable differences from their means. The GEM4K had one pCO₂ result that was flagged, with results of 40 mmHg, 40 mmHg, 40 mmHg, and 45 mmHg on 1 day of testing, with the result of 45 being out of our acceptable range by 1 mmHg. Curiously, the Rad90 had 1 “bad” day for pO₂ results: 143 mmHg, 125 mmHg, 125 mmHg, and 141 mmHg. However, the 143 mmHg result was barely within our 7.5% criteria for acceptability, so it was not flagged.

Table 3.

Acceptable differences from the mean for each analyte.

Analyte	Acceptable difference from mean, with unitsa	GEM4K (n = 27)	Rad90 (n = 27)
		Out of rangeb	Out of rangeb
pH	+0.03 units	0	0
pCO2	+3 mmHg or 7.5%	1	0
pO2	+5 mmHg or 7.5%	0	0
Na	+3.0 mmol/L	0	0
K	+0.3 mmol/L	0	0
Ionized Ca	+0.04 mmol/L	0	0
Glucose	+6 mg/dL or 10%	0	0
Lactate	+0.4 mmol/L or 10%	0	0
Hb	+0.5 g/dL or 5.0%	0	0
%O2Hb	+3%	0	0

Data based on 27 sets of the results each day of QC testing from March 7, 2018, to March 21, 2019.

mmHg × 0.1333 = kPa.

GEM4K = Instrumentation Laboratory GEM 4000 Blood Gas Analyzer.

Rad90 = Radiometer ABL90 Blood Gas Analyzer.

^aThe acceptable ranges of differences from the mean of all data from both the GEM4K and Rad90 used routinely to evaluate our between-instrument differences (6).

^bOut of range: this column shows the number of results for each analyte that were out of our acceptable differences, calculated as the difference between each measurement and the mean of the daily measurements from all analyzers.

To determine the differences in glucose results between the GEM4K and central laboratory analyzers, we reviewed the glucose results from our routine Chemistry Between-Laboratory Surveys that includes results from our central laboratory analyzers (DxC800) and our BG analyzers (GEM4K and Rad90). This survey uses pooled heparinized plasma for all analyzers. Table 4 shows a summary of the comparisons between the mean of the glucose results by the DxC800 compared to individual results by the GEM4K and Rad90 analyzers. Although the interval of glucose concentrations is relatively narrow (77–166 mg/dL), our comparisons showed no differences of significant clinical concern: from −6 to +10 mg/dL for GEM4K and from −8 to +10 mg/dL for the Rad90. These comparisons also showed the GEM4K results agreed more closely with the DxC800 results than did the Rad90 results. As shown by the LOA in Table 4, the GEM4K has less bias (mean difference); however, the variabilities are similar. Moreover, the ICC for GEM 4K with DxC800 are above 0.98 with a narrow 95% CI, whereas the ICC for Rad90 with DxC800 are around 0.95 with a much broader 95% CI (Table 4).

Table 4.

Comparison of plasma glucose results (mg/dL) between the central laboratory analyzer and the blood gas analyzers.

Instrument	Number of measures	Interval of resultsa	Interval of differencesb	Mean + SD of differencesc	ICCd (95% CI)	LOAe (mean ± 2SD)
Beckman DxC	48	77–166	——	——	——	——
GEM4K #1	48	77–171	−6 to 8	0.35 ± 3.39	0.98 [0.97, 0.99]	0.35 ± 6.78
GEM4K #2	48	74–176	−9 to 10	0.94 ± 3.73	0.98 [0.96, 0.99]	0.94 ± 7.46
Rad90 #1	46	81–169	−8 to 10	4.22 ± 3.38	0.95 [0.49, 0.99]	4.22 ± 6.76
Rad90 #2	46	83–172	−8 to 8	3.96 ± 3.21	0.96 [0.55, 0.99]	3.96 ± 6.42
Rad90 #3	42	83–170	−6 to 9	4.67 ± 2.62	0.96 [0.22, 0.99]	4.67 ± 5.24

Results are from routine quality control data to evaluate between-analyzer comparisons on 2 pooled plasma samples done approximately at 2-week intervals over a 1-year period. The Beckman DxC is the central laboratory analyzer.

^aInterval of results: the smallest number and the largest number of all the measurements.

^bInterval of differences: differences in measurements were calculated (the result from the blood gas analyzer—the result by the DxC analyzer). The smallest number and the largest number of the difference are presented.

^cMean ± SD of differences: mean and SD of the differences in measurements between blood gas analyzer and DxC analyzer are calculated.

^dICC: intraclass correlation coefficient of measurements between DxC analyzer and blood gas analyzer with 95% confidence interval (CI).

^eLOA: limits of agreement, which is the mean ± 2 times the standard deviation (SD).

Discussion

We believe our study using pools of fresh heparinized blood gave an appropriate evaluation of the within-model variability of 2 types of BG analyzers. Identical samples of a nearly ideal matrix were analyzed within a relatively short time and the within-model variability of 4 analyzers for each of 10 analytes was evaluated. We believe these results on virtually identical samples gave a true indication of analytical variability among the 4 analyzers of each model. The previous study compared differences between samples collected on the same patient but separated by up to 2 h (1). Because these critical care analytes can change markedly within an hour, even within a few minutes, some component of variability due to diet, physiologic changes, or therapeutic interventions could have been present.

Our results on plasma comparing glucose results between our central laboratory analyzers (DxC800) and BG analyzers also found very acceptable differences for all comparisons over a 1-year period, as noted in Table 4. The results on plasma showed the GEM4K results agreed more closely with the DxC800 (mean difference <1 mg/dL) than did the Rad90 to the DxC800 (mean differences ∼4 mg/dL). Both the earlier study (1) and the studies comparing glucose results (4, 5) used dissimilar specimens collected within 60 min of each other. However, the very large differences of 100–270 mg/dL on some comparisons (4) are not compatible with analytical variability, suggesting these outliers were likely due to preanalytical causes, such as glucose or insulin administration between the collection times. Other less-obvious nonanalytical differences between results on the same patient in these studies may have occurred (1, 2, 4, 5).

For our routine monitoring of between-analyzer agreement over this 55-week period, we note that only 1 analyte by the GEM4K was flagged as out of the acceptable range, that being a pCO₂ result out by 1 mmHg. The Rad90 had no outliers flagged, although one set of pO₂ results were barely within our acceptability criteria. While the Rad90 analyzers performed exceptionally well with excellent agreement among the 4 analyzers, the GEM4K analyzers also demonstrated very acceptable agreement among themselves.

The strengths of our study were that: (a) all samples were virtually identical on the day of testing by all 8 analyzers; (b) the matrices of these samples were nearly ideal, using fresh pooled nonhemolyzed heparinized blood from patients; (c) all testing was done within a relatively brief period of time, usually within 90 min; (d) all testing was done under routine calibration and quality control conditions, with no efforts made to calibrate prior to testing; and (e) the data were collected every 2 weeks over a 55-week period, which should detect any significant deterioration of method performance.

The weaknesses of our study were that (a) because the blood selected for pooling was for quality control purposes to evaluate between-analyzer agreement, the concentrations of each analyte were not preselected to achieve any particular range of values. Thus, the concentration intervals studied were relatively small; (b) this testing was done during a relatively consistent time of day, during the hours of 7:00 PM to 10:00 PM, although this would have been during the time that greater variability was noted in other reports (1, 2, 5).

The conclusions of our study are that (a) both brands of analyzers performed within acceptable limits throughout the year; (b) the Rad90 analyzers had excellent within-analyzer variation that was lower than the GEM4K for virtually every analyte; and (c) the GEM4K glucose results on plasma agreed more closely with the DxC results than did the Rad90. To summarize, our results on virtually identical samples do not support the findings of other studies (1, 2, 4, 5).

 

Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Author Contributions: All authors confirmed they have contributed to the intellectual content of this paper and have met the following 4 requirements: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; (c) final approval of the published article; and (d) agreement to be accountable for all aspects of the article thus ensuring that questions related to the accuracy or integrity of any part of the article are appropriately investigated and resolved.

J. Toffaletti planned the study, directed the research, evaluated the data, and prepared/edited the manuscript. K. Buckner collected the data, calculated preliminary statistics, and reviewed the manuscript. B. Liu and C.L. Green analyzed the data, provided statistical methods section, and interpretation of results and reviewed the manuscript.

Authors’ Disclosures or Potential Conflicts of Interest: Upon manuscript submission, all authors completed the author disclosure form. Disclosures and/or potential conflicts of interest: Employment or Leadership: None declared. Consultant or Advisory Role: J. Toffaletti, Instrumentation Laboratory. Stock Ownership: None declared. Honoraria: J. Toffaletti, Instrumentation Laboratory. Research Funding: J. Toffaletti: research support from Instrumentation Laboratory. B. Liu and C.L. Green: research reported in this publication was supported by the National Center for Advancing Translational Sciences of the National Institutes of Health under Award Number UL1TR002553. Expert Testimony: None declared. Patents: None declared. Role of Sponsor: The funding organizations played no role in the design of study, choice of enrolled patients, review and interpretation of data, preparation of manuscript, or final approval of manuscript.

Acknowledgments: We appreciate the help of Elizabeth Handel who tabulated some of the results.