Three-year monitoring and comparison of results from two identical blood gas analyzers.

Objectives: Measurement comparability between blood gas analyzers within a laboratory is of utmost importance. This study analyzed the data obtained from a three-year period. Design and methods: For quality monitoring one blood sample was tested on two identical blood gas analyzers at each of three shifts/day for three years. Deming regression analysis was used to determine result correlation and statistical identity between the two analyzers for pH, pCO2, pO2, sodium, potassium, chloride, ionized calcium, glucose, and lactate. Failures in the two-analyzer comparison were determined by the performance limits from the Institute of Quality Management in Healthcare (IQMH) and from the manufacturer respectively.
Results: Correlation coefficients were greater than 0.96 (0.9622-0.9975) for all tested analytes. The measurements of every analyte on both analyzers were statistically identical. In the two-analyzer comparison failure numbers/1000 tests for pO2 and glucose varied with the performance limits (IQMH: 0.6 and 49.2; the manufacturer: 19.3 and 4.4, respectively). In addition, persistent glucose failures (>5/week) between the two analyzers occurred occasionally. Conclusions: Results of all tested analytes between the two blood gas analyzers were interchangeable. Recurring glucose discrepancies might be a result of different lots of cartridges used on each analyzer, which were not identified during the initial installation.

Introduction

Blood gas analyzers are important laboratory instruments in critical care settings. They provide multiple tests in panels to determine acid/base status, the balance of electrolytes and metabolites, and oxygen delivery capacity in both Core Laboratory and Point of Care environments [[1], [2], [3]]. Inaccurate results may change medical decisions and impact patient safety. Therefore, manufacturers have been developing new technologies to improve the measurement and quality monitoring of blood gas analyzers.

The cartridge-based GEM Premier blood gas analyzers use an active quality process following the principles of risk management. The onboard intelligent quality management (iQM) system includes several process control solutions, calibration valuation products, pattern recognition software, and system hardware checks. It provides continuous monitoring of analytical processes with real-time automatic error detection, correction, and documentation [3,4]. The iQM has been verified at multiple laboratories for clinical use and demonstrated better sigma performance, faster error detection time, and higher error detection rates [5,6].

Usually, more than one blood gas analyzer is implemented in critical care units or Core Laboratories to provide urgent testing and backup testing. Similar to other chemistry analyzers, results from different blood gas analyzers are used interchangeably. Comparability of results is especially important for blood gas analyzers since multiple parameters are measured at the same time [7]. The results from multiple analyzers should be comparable in a single laboratory system or across all sites of an organization, this is required by many regulatory agencies for accreditation [8,9]. Laboratories have the responsibility to verify the interchangeability of analyzers at the time of installation and to monitor the interchangeability over the analyzers’ service life.

Limited studies had evaluated the result interchangeability or comparability of multiple blood gas analyzers by between-analyzer comparison during installation [7,10]. A report showed that all analytical parameters on the blood gas analyzers proved conformant, but slight differences between a couple of analyzers in the comparison study were observed [10]. Another report found that the total imprecision across different analyzers was low and indicated that multiple analyzers at different clinical units can be regarded as a single system. However, the authors also indicated that analyzer deviation should be taken into account since a certain degree of variance is inevitable when considering a large number of analyzers [7]. Few published reports are available regarding result interchangeability on the same model of blood gas analyzers in long period of service.

Two GEM Premier 4000 blood gas analyzers were installed in our Core Laboratory in 2018. After initial validation, we routinely tested one random whole blood sample on both analyzers at three shifts per day as additional quality monitoring, and as a check of result interchangeability. The procedure produced a large amount of data over a three-year period. This allowed us to confirm if the results on the two blood gas analyzers were interchangeable, to identify any failures in analyzer comparison that were not found in the initial validation, and to provide information for monitoring the analyzers appropriately.

Methods and materials

Analyzers

GEM Premier 4000 blood gas analyzer (Werfen, Ontario, Canada)

The operation and quality monitoring of the two GEM 4000 blood gas analyzers in the Core Laboratory were established to meet regulatory and manufacturer requirements. pH, pCO2, sodium, potassium, chloride, and ionized calcium are measured using potentiometric sensors, pO2 is measured by an amperometric sensor, and glucose and lactate are measured using enzyme-based amperometric sensors [3]. Samples measured on these analyzers were mostly collected from Emergency Departments and inpatient units. Samples from Intensive Care Units and Operating Rooms were analyzed by point-of-care testing and were not included in this study.

The onboard iQM process control solutions, which are traceable to National Institute of Standards and Technology primary standards, are processed through the same fluidic pathway as patient samples. Therefore, any obstructions or malfunctions from the sampler along the entire analytical pathway can be detected. New cartridges are verified by the iQM system as scheduled by the manufacturer. Any changes beyond the manufacturer's control limits are analyzed by pattern recognition software to diagnose, correct, and document. Only after the correction is successfully completed will the system be satisfactory for testing patient samples [3]. Cartridges containing 600 tests were used. Usually, different lots of cartridges were installed on the two analyzers respectively.

Architect c16000 chemistry analyzer (Abbott Diagnostics, Ontario, Canada)

Glucose was measured by hexokinase method on two automated chemistry analyzers in the Core Laboratory. The methods were regularly monitored by standard procedures and the performance met our laboratory quality standards.

Analyzer comparison

Comparison between two blood gas analyzers

After patient specimens had been tested, one arterial or venous whole blood sample was randomly selected and tested anonymously on both blood gas analyzers within a 5-min interval. This quality monitoring protocol was performed during each of the three shifts every day (midnight shift: 23:30 to 07:30, day shift: 07:30 to 15:30, evening shift: 15:30 to 23:30). The results of pH, pCO2, pO2, sodium, potassium, chloride, ionized calcium, glucose, and lactate were analyzed. This study presented the data obtained from July 2018 to June 2021.

Comparison between blood gas analyzer and chemistry analyzer

Once a week, one random whole blood sample was tested on one of the blood gas analyzers. The sample was then centrifuged immediately and the plasma was tested on one of the chemistry analyzers. The results of the glucose comparisons between July 2018 and June 2021 were included in this study.

Total allowable performance limits

Total allowable performance limits from the Institute of Quality Management in Healthcare (IQMH, Toronto, Ontario, Canada) [11] and from the manufacturer [3] (Table 1) were used for determining the acceptance of measurements in the two-analyzer comparison.

Table 1

Total allowable performance limits from IQMH and the manufacturer for tests on blood gas analyzer.

Test	IQMH	Manufacturer
pH	0.03	0.04
pCO2	<40 mmHg, +/− 4 mmHg	5mmHg or 8% whichever is greater
	≥40 mmHg, ±9%
pO2	<80 mmHg, ±15 mmHg	9mmHg or 10% whichever is greater
	≥80 mmHg, ±9%
Sodium	+/− 4 mmol/L	4 mmol/L within clinical range (120–160 mmol/L), 5 mmol/L outside the clinical range
Chloride	+/− 4 mmol/L	4 mmol/L or 5% whichever is greater
Potassium	<4.0 mmol/L, ±0.2 mmol/L	0.5 mmol/L or 7% whichever is greater
	≥4.0 mmol/L, ±5%
Ionized calcium	±7%	0.10 mmol/L or 10% whichever is greater
Glucose	<4.0 mmol/L, ±0.3 mmol/L	0.67 mmol/L or 12% whichever is greater
	≥4.0 mmol/L, ±7.5%
Lactate	<4.0 mmol/L, ±0.5 mmol/L	0.4 mmol/L or 15% whichever is greater
	≥4.0 mmol/L, ±12%

Statistical analysis

Deming regression analysis for determining the correlation of two analyzers was performed with EP Evaluator (Data Innovations, version 12). Results from the two analyzers were statistically identical when the 95% confidence interval (CI) of slope encompassed 1 and 95% CI of intercept encompassed 0. The two-analyzer comparison was also performed with EP Evaluator. Chi-square testing was performed on SPSS (IBM SPSS statistics, version 26) to identify the significance of failure numbers in the two-analyzer comparison between different groups. P < 0.05 indicated a significant difference.

Results

Regression analysis

More than 1000 results from each of the three shifts were collected and compared over a three-year period (Table 2). The total test numbers of the analytes at each shift were slightly different due to occasional unavailability of specific tests, or because the results were excluded when analyte concentrations were greater or less than reportable ranges. The results for each analyte were well distributed within the manufacturer's measuring ranges. Deming regression analysis showed a good correlation between the two blood gas analyzers at each shift. Correlation coefficients were greater than 0.96 (0.9622–0.9975) with the exception of the potassium at day shift (0.9265) which included one possible typo (see Supplementary Fig. 1). When this outlier was removed, the correlation coefficient of the potassium at day shift was 0.9900.

Table 2

Regression analysis of measurements between the two blood gas analyzers for nine tests at three work shifts.

Test	Shift	Test number	Result range	Correlation coefficient	Slope (95%CI)	Intercept (95%CI)
pH unit	Midnight	1085	6.95–7.62	0.9927	0.996 (0.989–1.003)	0.031 (−0.022 to 0.084)
	Day	1085	6.92–7.56	0.9789	1.007 (0.994–1.019)	−0.050 (−0.141 to 0.041)
	Evening	1059	6.92–7.68	0.9928	1.006 (0.999–1.013)	−0.042 (−0.096 to 0.012)
	All	3229	6.92–7.68	0.9884	1.003 (0.997–1.008)	−0.018 (−0.057 to 0.020)
pCO2mmHg	Midnight	1091	17–100	0.9945	1.000 (0.993–1.006)	−0.1 (−0.3 to 0.2)
	Day	1091	19–104	0.9957	0.999 (0.993–1.004)	0.2 (0.0–0.5)
	Evening	1065	14–116	0.9959	0.998 (0.993–1.004)	0.1 (−0.2 to 0.4)
	All	3247	14–116	0.9955	1.000 (0.997–1.003)	0.0 (−0.1 to 0.2)
pO2mmHg	Midnight	1080	6–352	0.9927	1.006 (0.999–1.014)	−0.5 (-1.0 to -0.1)
	Day	1079	11–265	0.9884	1.009 (1.000–1.019)	−0.3 (−0.8 to 0.1)
	Evening	1058	9–395	0.9931	0.996 (0.989–1.003)	0 (−0.3 to 0.4)
	All	3217	6–395	0.9922	1.003 (0.998–1.007)	−0.2 (−0.5 to 0.0)
Sodium mmol/L	Midnight	1085	110–161	0.9636	0.987 (0.972–1.002)	1.2 (−0.8 to 3.2)
	Day	1078	116–155	0.9685	0.989 (0.975–1.004)	0.8 (−1.2 to 2.8)
	Evening	1060	115–158	0.9653	1.002 (0.986–1.018)	−0.9 (−3.1 to 1.2)
	All	3223	110–161	0.9670	0.993 (0.984–1.002)	0.4 (−0.8 to 1.5)
Potassium mmol/L	Midnight	1086	0.8–9.2	0.9956	1.000 (0.994–1.005)	0.00 (−0.03 to 0.02)
	Day	1086	1.6–10.0	0.9265	1.009 (0.986–1.032)	−0.04 (−0.13 to 0.05)
	Evening	1059	0.8–10.3	0.9871	1.009 (0.999–1.019)	−0.04 (-0.085 to -0.004)
	All	3231	0.8–10.3	0.9712	1.005 (0.987–1.014)	−0.027 (−0.061 to 0.007)
Chloride mmol/L	Midnight	1086	82–132	0.9827	1.003 (0.992–1.014)	0.1 (−1.1 to 1.2)
	Day	1085	75–130	0.9735	1.016 (1.002 to 1.030)	−1.2 (−2.6 to 0.2)
	Evening	1057	82–136	0.9844	1.020 (1.009 to 1.031)	−1.6 (-2.7 to -0.5)
	All	3228	75–136	0.9804	1.009 (1.002 to 1.016)	−0.5 (−1.2 to 1.2)
Ionized calcium mmol/L	Midnight	1078	0.67–2.24	0.9695	1.005 (0.990–1.020)	−0.009 (−0.026 to 0.008)
	Day	1073	0.68–1.52	0.9622	1.047 (1.029–1.064)	−0.056 (-0.076 to -0.036)
	Evening	1048	0.24–1.64	0.9727	0.999 (0.985–1.013)	−0.002 (−0.018 to 0.014)
	All	3199	0.24–2.24	0.9689	1.013 (1.004 to 1.022)	−0.018 (-0.028 to -0.008)
Glucose mmol/L	Midnight	1087	1.3–36.6	0.9966	1.012 (1.007 to 1.017)	−0.08 (-0.13 to -0.04)
	Day	1084	1.2–33.0	0.9963	0.994 (0.989 to 0.999)	0.056 (0.013 to 0.100)
	Evening	1060	1.1–41.6	0.9966	1.000 (0.995–1.005)	0.028 (−0.019 to 0.074)
	All	3231	1.1–41.4	0.9965	1.003 (1.000–1.006)	−0.004 (−0.003 to 0.023)
Lactate mmol/L	Midnight	1082	0.4–17.0	0.9975	1.008 (1.004 to 1.012)	−0.018 (−0.029 to 0.006)
	Day	1083	0.4–14.2	0.9965	0.997 (0.992–1.002)	−0.01 (−0.02 to 0.00)
	Evening	1060	0.5–16.3	0.9961	1.026 (1.021 to 1.032)	−0.053 (-0.067 to -0.039)
	All	3225	0.4–17.0	0.9968	1.011 (1.009 to 1.014)	−0.027 (-0.034 to -0.021)

Apart from a few exceptions (as highlighted in Table 2), the intercepts of the regression lines passed through zero. Similarly, the slopes were at unity except a few for chloride, glucose and lactate (as highlighted in Table 2). The causes of exceptions might be sample quality, mixing, or delay in testing (see Supplementary Fig. 1).

Results that failed the performance limits in the analyzer comparison

Table 3 lists the number of paired results that were out of the performance limits established by IQMH or the manufacturer in the analyzer comparison. Failures were expressed as numbers per 1000 tests. For each analyte, there was no significant difference in the number of failures among the three shifts regardless of which limit set was used. Glucose showed an average of 49.2 failures per 1000 tests when using IQMH performance limits (vs. 4.4 using manufacturer's limits). When using manufacturer performance limits, pO2 had an average 19.3 failures per 1000 tests (vs. 0.6 using IQMH limits). These differences in failure numbers were statistically significant (p < 0.05). The failure numbers were slightly different in the tests of pH, pCO2, potassium, ionized calcium, and lactate when determined by either IQMH or manufacturer limits but had no significant difference. The failure numbers of sodium and chloride were the same since the performance limits in both limit sets were the same.

Table 3

Failure numbers determined by different performance limits in the blood gas analyzer comparison.

Test	Shift	Test number	Failure number by limits from IQMH (per 1000 tests)	Failure number by limits from the manufacturer (per 1000 tests)
pH unit	Midnight	1085	0.9	0
	Day	1085	4.6	1.8
	Evening	1059	0.9	0
pCO2mmHg	Midnight	1091	0	0.9
	Day	1091	0	0
	Evening	1065	0.9	0
pO2mmHg	Midnight	1080	0.9	25.0*
	Day	1079	0.9	11.1*
	Evening	1058	0	21.7*
Sodium mmol/L	Midnight	1085	0	0
	Day	1078	2.8	2.8
	Evening	1060	4.7	4.7
Potassium mmol/L	Midnight	1086	1.8	0
	Day	1086	9.2	4.6
	Evening	1059	4.7	0.9
Chloride mmol/L	Midnight	1086	0	0
	Day	1085	1.8	1.8
	Evening	1057	1.0	1.0
Ionized calcium mmol/L	Midnight	1078	5.6	0.9
	Day	1073	5.6	0.9
	Evening	1048	6.7	0
Glucose mmol/L	Midnight	1087	49.7	3.7*
	Day	1084	45.2	3.7*
	Evening	1060	52.8	5.7*
Lactate mmol/L	Midnight	1082	2.8	0.9
	Day	1083	0.9	0.9
	Evening	1060	2.8	1.9

*P value<0.05 when compared to the failure numbers determined by the limits from IQMH.

Comparison of glucose measurement between the blood gas analyzer and the chemistry analyzer

To investigate the failure in glucose testing, we compared glucose measurements obtained from each blood gas analyzer with the results from one of the two chemistry analyzers within the same three-year period. Table 4 shows that both blood gas analyzers correlated well to the chemistry analyzer respectively with a correlation coefficient greater than 0.997. The slopes and intercepts deviated slightly from 1 and from 0 respectively. One blood gas analyzer had slightly more failure numbers than the other, but the difference was not statistically significant (P > 0.05).

Table 4

Comparison of glucose measurement between the blood gas analyzer and the chemistry analyzer.

Glucose	Test number	Result range	Correlation coefficient	Slope (95%CI)	Intercept (95%CI)	Failure number by limits from IQMH (per 1000 tests)	Failure number by limits from the manufacturer (per 1000 tests)
GEM1 to Architect	339	2.5–29.6	0.9976	1.027 (1.019–1.035)	−0.076 (−0.135 to −0.017)	23.6	0
GEM2 to Architect	339	2.1–29.8	0.9974	1.042 (1.033–1.050)	−0.143 (−0.209 to −0.077)	47.2*	2.95

*P > 0.05 when compared to GEM1.

Weekly distribution of failed glucose comparisons between the two blood gas analyzers

Fig. 1 shows the distribution of failed glucose comparisons between the two blood gas analyzers over three years (160 weeks) when evaluated by IQMH performance limits. With glucose concentrations ranging from 3.5 to 27.1 mmol/L, 158 failures randomly occurred in 73 weeks. The observation of 1–2 failures within one week could be considered as the baseline due to occasional sample quality or sample handling errors. However, we also observed that 5 to 10 failures persistently occurred within one week in eight of the 73 weeks, which resulted in disruption (>2 persistent failures/year) in laboratory testing.

Fig. 1

Occurrence of failed glucose comparison between the two blood gas analyzers over time.

Discussion

The data collected from three daily shifts over a period of three years allowed us to evaluate the long-term result agreement and result interchangeability between the two of the same model of blood gas analyzers. Overall, each test such as blood gases, electrolytes, ionized calcium, and metabolites (glucose and lactate) measured on the two analyzers correlated well and were (or close to) statistically identical. However, the failure numbers for pO2 and glucose in the two-analyzer comparison was significantly different when using different total allowable performance limits. A persistent failure (lasting for up to 6 days) of glucose measurement in the two-analyzer comparison occurred randomly more than two times per year, which suggests recurring challenges with glucose testing.

When multiple analyzers of the same model are deployed in a Core Laboratory or as part of Point-of-Care Testing, they are usually considered as one system since they use the same reagents or cartridges [12]. During the initial validation of multiple analyzers, representative analyzers are validated comprehensively against a reference instrument, and additional analyzers are validated against the representatives with a simplified protocol [7,10,12]. This validation strategy also includes between-analyzer comparison to evaluate analyzer comparability. One published study reported the validation for 12 blood gas analyzers at installation [10]. Although the performance of all analytical parameters met the quality standards, the authors noticed that the slopes of one regression line for sodium and one for hemoglobin were greater than other analyzers. The differences of a few pairs of results were out of -2SD to 2SD range in the analyzer comparison. However, these exceptions did not change the clinical and biological interpretation of the results [10]. Another study investigated the comparability of 22 Point-of-Care blood gas analyzers. The results showed essential comparability across all devices tested, but larger imprecision due to analyzer variation was observed for metabolites (glucose and lactate) and the lowest quality control level of ionized calcium. Also, a couple of analyzers had larger limits of agreement for tests pO2, sodium, potassium, and total hemoglobin in the analyzer comparison [7]. GEM 4000's iQM is an automated quality monitoring module. It does not require any external quality control material for additional monitoring of the measurements [3]. Therefore, regular comparison with whole blood samples was used to monitor the measurement and the reproducibility of results between the two blood gas analyzers in our laboratory. Our investigation showed that in a three-year period every test on both analyzers correlated well and reported results were statistically identical.

The total allowable performance limits established by different methods have a direct impact on failure numbers in analyzer comparison and subsequent troubleshooting plans. At present, there is no consensus on the preferred method for establishing medically necessary analytic performance limits [13]. Limits that reflect clinical needs and medical care are rarely available [11,12]. Limits based on individuals' biological variation do not cover every test on blood gas analyzers, for example, pH and pO2 [14]. In current laboratory services, we use the total allowable performance limits established by the Proficiency Testing Program from IQMH, which combines test biological variation and achievable performance [11]. However, they may not be perfect for every test and the mean of peer group is not necessarily the true value for every proficiency testing specimen [13,15]. In this study, we also determined the failures in the two-analyzer comparison using the performance limits from the manufacturer [3]. There was no significant difference in failures among three work shifts when data was evaluated with IQMH or manufacturer's limits.

Compared to IQMH performance limits that were used in our service, the manufacturer's performance limits on pO2 were more stringent at low values. This suggests that the optimal performance of their analyzers might exceed our clinical requirements. The pO2 data from this study met the performance limits set by IQMH, so the difference in failure numbers between the two performance limits observed in pO2 likely has no clinical significance. Based on the manufacturer's performance limits, glucose results were similar between the two analyzers. There were less than 6 failures among 1000 sets of data. However, when IQMH's performance limits were used, we observed a significantly higher number of failures. This reflects the more stringent quality target set by IQMH that is beyond the performance capability stated by the manufacturer. The high failure rate in glucose testing also necessitated frequent troubleshooting and disruption of services. It is desirable to have a dialogue between manufacturers and accreditation agencies to establish appropriate test performance limits.

During the three years of service, a persistent discrepancy in glucose measurement on the two blood gas analyzers occurred two or more times per year, but no flags were displayed in iQM on the analyzers. The comparison failed for several repeats after we checked sample quality, reviewed the testing procedure, and retested at different shifts. The problem sometimes persisted for several days. In our troubleshooting, we also compared the whole blood glucose to plasma glucose of the same samples measured on the chemistry analyzers. For the majority of cases, results were acceptable within the allowable performance limits and testing was continued. Blood glucose on blood gas analyzers and plasma glucose on chemistry analyzers were compared over the three years, the regression analysis was acceptable and 95% of the data was within IQMH limits. When the comparison of glucose persistently failed, a new blood gas cartridge was installed and the failures could be corrected. However, this is an expensive corrective action. Following our troubleshooting procedure, the observed failures in comparison of glucose results between the two blood gas analyzers were corrected and did not impact clinical interpretation or patient management. The between-analyzer failures for glucose most likely are a result of minor cartridge manufacturing differences [7] or lot-to-lot variation [9,16]. Reagent lot-to-lot variation has to be minimized for laboratories to produce consistent results over time. However, large reagent lot-to-lot variation has been observed frequently on many chemistry tests in laboratory services. Laboratories should develop approaches to provide an acceptable level of statistical robustness based on specific needs of test performance [16]. Our hospital has a relatively low monetary limit set for each reagent purchase, which limits the size of each purchase. In addition, different lot numbers of cartridges were usually received with each purchase. Thus, the two analyzers usually have different lots of cartridge on board at any given time. We followed the manufacturer's recommendation in the verification of each new cartridge. However, the data points were inadequate to evaluate the lot-to-lot variation of the cartridges. Increasing the volume of each purchase to minimize lot-to-lot differences between analyzers may be a solution.

With the huge effort of the manufacturer in quality improvement, the iQM system on cartridge-based blood gas analyzers has greatly reduced the variation of quality control testing [5,6]. For laboratory users, especially in the setting of point-of-care testing, this has simplified the operation procedures. There are also fewer errors observed in the measurements [5,6]. On the other hand, a laboratory's capacity in analyzer maintenance and troubleshooting may be limited. If the results are questioned during troubleshooting, the most a laboratory can do is to replace the cartridge. Although the blood gas analyzers are equipped with advanced technology for quality control monitoring, we recommend that a quality assurance procedure such as patient result correlation between the analyzers at laboratory defined frequency is a necessity, especially when a new lot of cartridge is installed.

Conclusions

The overall measurements of blood gases, electrolytes, ionized calcium and metabolites between the two blood gas analyzers in our laboratory were statistically identical and interchangeable. However, occasionally a persistent incomparable glucose result between the analyzers could occur randomly and might not be identified at the time of installation. Although new technology for quality monitoring is applied on blood gas analyzers, laboratories have the responsibility to verify result interchangeability between multiple analyzers used in a single laboratory or an organization. In addition, the strategy for cartridge purchasing should be optimized to minimize the lot-to-lot variation.

Author statement

Yun Huang: Conceptualization, Methodology, Formal analysis, Writing-Original draft, Writing-Review and editing. Robert Dean: Methodology, Validation, Investigation, Writing-Review and editing. Yvonne Dubbelman: Methodology, Validation. Anne Vincent: Methodology, Writing-Review and editing. Michael Chan: Methodology, Writing-Review and editing.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declaration of competing interest

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.