Article Text
PDF
Abstract
Background Point-of-care testing allows rapid analysis of samples to facilitate prompt clinical decisions. Electrolyte and calcium abnormalities are common in acutely ill patients and can be associated with life-threatening consequences. There is uncertainty whether clinical decisions can be based on the results obtained from blood gas analysers or if laboratory results should be awaited.
Objectives To assess the agreement between sodium, potassium and calcium results from blood gas and laboratory mainstream analysers in a tertiary centre, with a network consisting of one referral and two peripheral hospitals, consisting of three networked clinical biochemistry laboratories.
Method Using the laboratory information management system database and over 11 000 paired samples in three hospital sites, the results of sodium, potassium and ionised calcium on blood gas analysers were studied over a 5-year period and compared with the corresponding laboratory results from the same patients booked in the laboratory within 1 h.
Results The Pearson's linear correlation coefficient between laboratory and blood gas results for sodium, potassium and calcium were 0.92, 0.84 and 0.78, respectively. Deming regression analysis showed a slope of 1.04 and an intercept of −5.7 for sodium, slope of 0.93 and an intercept of 0.22 for potassium and a slope of 1.23 with an intercept of −0.55 for calcium. With some strict statistical assumptions, percentages of results lying outside the least significant difference were 9%, 26.7% and 20.8% for sodium, potassium and calcium, respectively.
Conclusions Most clinicians wait for the laboratory confirmation of results generated by blood gas analysers. In a large retrospective study we have shown that there is sufficient agreement between the results obtained from the blood gas and laboratory analysers to enable prompt clinical decisions to be made.
- interpretation
- diagnosis
- clinical management
- errors
- metabolic/diabetes/endocrine
Statistics from Altmetric.com
Key messages
What is already known on this subject?
In studies investigating the agreement between laboratory and point-of-care (POC) test results, where samples were taken in a controlled environment, with a short interval between the laboratory and POC sample analysis, there has been a good agreement between the methods for electrolyte measurement.
What might this study add?
The current study, which involved gathering data in a natural diagnostic setting, shows sufficient agreement between the results obtained from the blood gas and laboratory analysers in a real-life situation to enable clinicians to make reliable and prompt clinical decisions.
Introduction
Severe disturbances in sodium, potassium and calcium are among the most important treatable medical emergencies. Late diagnosis and treatment of these abnormalities could have serious consequences. This is one of the reasons that the repertoire of modern ward-based blood gas analysers includes sodium, potassium and ionised calcium. In the presence of abnormal results, clinicians need to know whether the results obtained by such point-of-care (POC) devices are reliable and how they compare with those generated in the laboratory. This would facilitate appropriate and prompt clinical decisions. Previous studies have shown that only about one-third of clinicians rely on POC testing to guide their clinical decisions.1 Although the majority of clinicians find POC testing a useful method to obtain rapid results, less than half would rely on such results, and prefer to wait for laboratory confirmation before making important clinical decisions.2 Sometimes, this delays urgently needed treatment. The aim of our study was to assess the agreement between sodium, potassium and calcium results from blood gas and laboratory mainstream analysers in a natural clinical setting in a tertiary referral centre, with a network of one referral and two peripheral hospitals consisting of three networked clinical biochemistry laboratories.
Methods
The laboratory information management system (LIMS) database was retrospectively searched for results generated between 1 April 2008 and 15 January 2013. The search strategy extracted all the POC sodium, potassium and ionised calcium results from individuals, with their paired concurrent (within 1 h) laboratory analyses. The laboratory-based analyses also included albumin and adjusted calcium (adjusted calcium=measured calcium+(0.02×(40−albumin))).
POC analyses were performed using Radiometer ABL 700 and 800 analysers in several adult and paediatric units in three hospital sites. Blood gas samples were either arterial or venous blood collected into a heparinised syringe and analysed immediately after scanning the patient barcoded identity label, which was automatically uploaded into the LIMS.
Laboratory analyses were undertaken on Advia 2400 Clinical Chemistry analysers (Siemens Medical Diagnostics, Frimley, UK) in three laboratory sites using lithium heparin PSTII gel tubes (BD, Oxford, UK). The three laboratories were based in three hospitals in Oxford University Hospitals NHS Trust and electronically networked into a common LIMS.
To minimise adverse bias from preanalytical factors in either the POC or laboratory analyses, data were excluded if any of the following existed:
Potassium >6 mmol/L concurrent with calcium, magnesium and alkaline phosphatase below their reference intervals, suggestive of EDTA contamination. These results were excluded only from potassium and calcium comparisons, and not from sodium comparisons.
Potassium result on laboratory samples with significant haemolysis according to the laboratory analysers’ haemolysis index.
Ionised calcium of <0.5 mmol/L and potassium of >10 mmol/L were excluded from calcium and potassium comparisons, as these would be physiologically inconsistent with life.
In excluding the above results investigators were blind to the clinical information.
In cases where there were multiple POC analyses within 60 min of one laboratory sample, only the POC specimen taken at the time closest to the laboratory result was accepted for comparison studies. Samples received from neonatal units were excluded from the study.
Statistical analysis was performed using the XLSTAT software. For each analyte, the two methods were compared using the Pearson's linear correlation coefficient, Bland–Altman and Deming statistical methods of agreement between two methods.3
As in blood gas analysers the calcium is reported as ionised calcium, we used ionised calcium×2 for comparison with adjusted calcium.4
In order to have an estimate of the comparability of the two methods, we used the concept of measurement uncertainty (MU) and least significant difference (LSD). From a statistical point of view the difference between two successive values has less than a 5% chance of being due to random variation if it is <2.8 SDs. This figure is arrived at by what can be thought of as an inverse application of t test. A 5% probability equates to a value of t (two-sided) of 1.96. When reported as the ratio of the mean, the SD is replaced by coefficient of variant (CV). Online supplementary appendix 1 shows the basis for this calculation.5 In that context we used our laboratory assays CV for quality control (QC) materials during the study period. Measurement of uncertainty focuses on identifying the dispersion of results that might have been obtained for an analyte if a sample had been measured repeatedly. To do this, it uses available data about repeated measurements from a given measuring system (in this case QC material) to define an interval of values within which the true value of the measured analyte is believed to lie with that method. In our study, this was not possible to directly implement, as the methods for the laboratory and POC tests were different. However, with an assumption that the two samples were analysed with the same method, 2.8 times (1.96×√2) CV value (of the QC material in that period) was considered the LSD between the two results, which was considered MU in our analysis. This gave an estimation of the proportion of the results with less than the LSD.6
Results
A total of 19 788 paired samples for sodium, potassium and 15 229 samples for calcium fulfilled the time interval inclusion criteria. After excluding the repeated POC samples per each laboratory sample and neonatal samples, 15 322, 15 304 and 11 343 pairs of samples were studied for sodium, potassium and calcium, respectively. Overall, after exclusion of potassium results of >10 mmol/L (n=51), based on high potassium, low calcium, magnesium and alkaline phosphatase, we observed low incidence of significant EDTA contamination for blood gas samples (n=3) and more in the laboratory samples (n=5). The data obtained have been summarised in table 1, which shows means of 140.1, 4.2 and 2.3 mmol/L for the laboratory and 139.5, 4.1 and 1.16 mmol/L for POC sodium, potassium and (ionised) calcium, respectively.
Descriptive statistics for laboratory and blood gas analytes
The estimated CV of the laboratory assays for sodium, potassium and calcium QC samples during the study period were 0.9%, 1.8% and 2.2%, respectively. The LSD for sodium, potassium and calcium were 2.5%, 5.1% and 6.2%, respectively. The distributions of percentage differences between the analytes with the LSD line are shown in figures 1⇓–3. The percentages of results lying outside LSD were 9%, 26.7% and 20.8% for sodium, potassium and calcium, respectively. Of these, 8% of sodium, 20% of potassium and 13% of calcium POC results were lower than the corresponding laboratory results, whereas the rest were higher (figures 1⇓–3 and table 2).
Number of paired results difference and LSD
Percentage difference of sodium measurements and normality plot. The dotted lines show the hypothetical levels of acceptance or least significant difference (LSD), assuming the methods were the same.
Percentage difference of potassium measurements and normality plot. The dotted lines show the hypothetical levels of acceptance or least significant difference (LSD), assuming the methods were the same.
Percentage difference of calcium measurements and normality plot. The dotted lines show the hypothetical levels of acceptance or least significant difference (LSD), assuming the methods were the same.
The Pearson's linear correlation coefficient between laboratory and blood gas results for sodium, potassium and calcium were 0.92, 0.84 and 0.78, respectively. The Bland–Altman analysis showed that the overall bias for sodium was −0.57 mmol/L (95% CI −0.61 to −0.53), and the 95% CI of the differences was −5.38 to 4.24. The overall bias for potassium was −0.08 mmol/L (95% CI −0.08 to −0.09), with the 95% CI of the differences −0.63 to 0.46 mmol/L. The overall bias for calcium was −0.015 mmol/L (95% CI −0.013 to −0.017) and the 95% CI for the difference between the methods was −0.27 to 0.24 (online supplementary figures 4–6).
Deming regression analysis showed a slope of 1.04 (95% CI 1.03 to 1.05) and an intercept of −5.7 (95% CI −4.5 to −6.8) for sodium, slope of 0.93 (95% CI 0.92 to 0.94) and an intercept of 0.22 (95% CI 0.17 to 0.27) for potassium and a slope of 1.23 (95% CI 1.20 to 1.25) and an intercept of −0.55 (95% CI −0.49 to −0.61) for calcium (online supplementary figures 7–9). The differences between laboratory and POC results were >5% for 1% of sodium, 27% of potassium and 30% of calcium results.
In order to evaluate the effect of difference in the two measurements in clinical decision making, we compared the POC and corresponding laboratory values in relation to the reference intervals and also to the critical values used in the laboratory for definition of severe abnormalities. (tables 3⇓⇓⇓⇓–8 and online supplementary figures 10–15).
POC and laboratory sodium results and critical cut-offs
POC and laboratory sodium results and the reference intervals
POC and laboratory potassium results and critical cut-offs
POC and laboratory potassium results and the reference intervals
POC and laboratory calcium results and critical cut-offs
POC and laboratory calcium results and reference intervals
When analysed according to requesting locations, the correlation between laboratory and POC results for sodium, potassium (n=2088) and calcium (n=363), were 0.87, 0.83 and 0.62, respectively, for emergency medicine and 0.94, 0.88 (n=10 087) and 0.74, respectively, for intensive treatment unit (ITU) (n=9068).
Discussion
Sodium
Measuring sodium by direct-measuring ion specific electrodes (ISEs) on a blood gas analyser offers a more accurate means of measurement that is not affected by high triglyceride or protein, which causes pseudohyponatraemia on indirect ISEs used by most mainstream laboratory analysers.7 This potentially helps in managing the acutely unwell patients more efficiently. However, the high correlation coefficient (r=0.92) suggests that this was not a common occurrence in our studied population and the two methods were reliably close. The medians were very close to the mean, suggestive of normal distribution of data. The Bland–Altman graph showed small systematic bias between the two methods at different levels of sodium, namely, −0.57 mmol/L. Deming regression analysis showed a slope of 1.04 (95% CI 1.03 to 1.05) and intercept of −5.7 (95% CI −4.5 to −6.8), suggestive of a high level of agreement. Even with acceptance lines of −2.5% to 2.5% in figure 1, which is a tight hypothetical cut-off based on our laboratory method CV of 0.9%, 91% of the results lie within the defined threshold.
Potassium
The issues around measurement and reporting of potassium results are more complex than that of sodium, especially with regard to preanalytical factors, such as haemolysis, EDTA contamination and delayed centrifugation. Haemolysis can affect both blood gas and laboratory samples. Although many modern laboratory analysers use a haemolysis detection and reporting system, this is not possible on a blood gas analyser, as it uses whole blood samples. Delay in transit and as a result delay in centrifugation of the sample can cause falsely elevated potassium in the laboratory results, due to leakage of potassium out of blood cells, especially red blood cells. This is not an issue in analysing samples on a blood gas analyser, as the analysis is done shortly after sampling. Contamination with EDTA can affect the potassium levels, which is less likely in blood gas analysis as it is done in pre-prepared tubes. Indirect ways of detecting this effect include measuring calcium and magnesium level or alkaline phosphatase activity, which would be falsely low due to the chelating effect of EDTA.8 After excluding the spurious results due to above-mentioned factors in our study, the correlation coefficient was 0.84, which suggests a good correlation between the two methods. The Bland–Altman method showed a small overall systematic bias of −0.08 between the two methods. Deming regression analysis showed a slope of 0.93 (95% CI 0.92 to 0.94) and intercept of 0.22 (95% CI 0.17 to 0.27), which also suggests a very high level of agreement. With acceptance lines of −5.1% to 5.1% in figure 2, 73.3% of the results lie within the defined threshold of LSD. It should be noted that the overall comparability of the two methods is less than the sodium methods, probably due to the effect of an hour delay and undetectable haemolysis in blood gas analysis.
Calcium
It is of note that in the majority of blood gas analysers, calcium is reported as ionised calcium, which can be confusing for some clinicians. Awareness of the relationship between ionised calcium and adjusted calcium is crucial in order to make accurate diagnosis and initiate appropriate treatment. Studies have shown that ionised calcium is about half of the total calcium and is a better physiological marker of homeostasis than the albumin-adjusted calcium, especially in cases of severe abnormalities in albumin, for example, in low albumin states such as nephrotic syndrome or acute sepsis.9 The main preanalytical factor affecting calcium is contamination with the anticoagulant EDTA, which gives falsely low results due to the chelating effect of EDTA. This is less likely in blood gas analysis as it is performed on pre-prepared tubes.
In our study Bland and Altman comparison of ionised and adjusted calcium showed a small systematic bias between the two methods in different levels, that is, −0.015 mmol/L. Deming regression analysis showed a slope of 1.23 (95% CI 1.20 to 1.25) with intercept of −0.55 (95% CI −0.49 to −0.61). With the strict acceptance lines of −6.2% to 6.2% in figure 3, based on our laboratory method CV of 2.2%, 79.2% of the results lie within the defined threshold. The agreement of POC and laboratory results for calcium was less than that for sodium. This is partly due to the errors involved in measurement of albumin, which is used to estimate ionised calcium (adjusted calcium) and also due a widespread use of a relatively unsophisticated formula for estimation of ionised calcium, which does not consider some factors such as blood pH, which will have an effect on ionised calcium.
As shown in tables 3⇑⇑⇑⇑–8 and online supplementary figures 9–15, measurement of the samples in the laboratory did not change many samples from being critically abnormal to normal, or vice versa. By using laboratory results, 0.5% of all samples changed their status from being critically abnormal to non-critically abnormal for sodium, 0.7% for potassium and 1.1% for calcium. However, when the reference ranges were considered the number of samples changing groups from normal to abnormal was higher. The overall change in status from abnormal to normal was 16.7% for sodium, 7.4% for potassium and 14% for calcium in using laboratory results compared with blood gas analysers.
As previously mentioned, despite lack of widespread reliance on POC test results for clinical decision making, we have clearly shown that this perception needs to be changed as the two methods for each analyte agree very well.
In general, making a clinical decision for treatment of a patient with electrolyte and/or calcium abnormality is based on presenting symptoms and findings in clinical examination, for example, signs of dehydration, fluid over load and neuromuscular junction abnormalities. On the other hand, accurate interpretation of abnormalities also depends on pretest probability of the findings. In the clinical scenarios where the index of suspicion is clinically high for the abnormality detected on POC testing, treatment should be started instead of waiting for repeat test from the laboratory to arrive. This is supported by the high agreement we found in the results. As the level of agreement is high with whichever statistical method we use, we recommend that in occasions where the result, especially potassium, is clinically in doubt, repeating the analysis with blood gas would be the most efficient way of getting a result reliably close to the true value.
In fact, blood gas analysers give accurate and in several occasions more reliable results for all three analytes. For sodium because of a lack of effect of pseudohyponatraemia, for potassium because of a lack of delays in separation of the sample and the fact that the likelihood of EDTA contamination is far less, and for calcium as it is physiological and the calculation for adjusting the ionised calcium have their inherent errors. The only drawback is when the sample is haemolysed, which affects potassium. We should emphasise that the laboratory results are also prone to errors and should not be considered as true value; that is why we used different methods for their comparison. The main advantage is that laboratory findings can indicate haemolysis.
The higher correlation of results in ITU wards is probably due to the difference in practice in the two settings. This gap can be narrowed and the overall quality of POC reports enhanced by improvements made in training of all relevant staff, QC of equipment and performing audits to ensure that clinicians can make the optimal use of POC testing to make effective clinical decisions.
One strength of our study is the large number of samples in different groups.
Second, as both the blood gas and the laboratory samples were heparinised, the effect of sample type on the results was negligible. Third, the agreement between blood gas analyser and the main laboratory analyser results was assessed in a real-life clinical situation, which has not been reported before. Although between-analyser and within-analyser variability in the laboratory and also the blood gas analysers would have an effect on the agreements, our results show that in day-to-day clinical practice the POC results are reliably close to the laboratory results.
One limitation of the study is that patients might have had medical treatment between the collection of the two samples, thereby changing the physiological state and rendering the two results incomparable. Although we did not have the information regarding any intervention, we hypothesised that this was unlikely to have happened in that short period of time, as the majority of clinicians usually wait for laboratory results before treating their patients. On the other hand, there might be some variation in the association between the results in subgroups of the studied population with different clinical and demographic features, which this study was not designed to address. In addition, the comparisons were made between the platforms used in the laboratory and POC at the time of study. Local correlation studies will be needed if the results were to be extrapolated to other laboratory or POC platforms.
Conclusion
Our study suggests very good agreements between the laboratory and blood gas results for sodium, potassium and calcium in a real-life clinical scenario, despite some minor limitations. The correlation was generally better in the ITU setting than in emergency medicine, probably due to a more controlled clinical setting. We strongly support encouraging our medical community, especially the acute care clinicians, to rely more on the results of blood gas analysers. This will enable them to act more quickly and reduce any adverse consequence of delays for the patient.
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
- Data supplement 1 - Online appendix
- Data supplement 2 - Online figures
Abstract in Persian
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
- Abstract in Persian - Online abstract
Footnotes
Contributors All authors made substantial contributions to the paper. BS designed the data acquisition together with MM, AM and TJ and supervised the integrity of statistical methods used. MM performed the data analysis and interpretation of data, which was helped by BS, AM and TJ. Drafting the work or revising it critically for important intellectual content was done by MM and AM. Final version was approved by all authors.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.