The influence of undetected hemolysis on POCT potassium results in the emergency department

Introduction

Point-of-care testing (POCT) can be defined as the clinical laboratory tests conducted close to the site of patient care by personnel untrained in laboratory skills, that can lead to a possible change in the care of the patient with the ability to reduce turnaround time but without compromising the quality [1].

There are well-recognised advantages to the usage of these tests, particularly the rapid availability of results that allow for prompt clinical decision-making. The possibility of using unprocessed specimens with a small sample volume, preservation of sample quality, the possibility for more lean processes, and the ability to provide laboratory testing in a wider variety of sites or circumstances are other advantages of implementing POCT [2].

There are, however, some challenges associated with these types of tests, which could impact on patient care, including, in some cases, questionable reliability of POCT results [3].

Healthcare professionals running these tests are not necessarily qualified in laboratory discipline, but they need to be adequately trained and certified by laboratory professionals [2]. Nurses are the main users of POCT devices. During the last decades, an international shortage of qualified nurses has developed. Collaborative working is key to ensure that improvements made to the preanalytical process reflect users’ needs, and optimising preanalytical processes may enable best use of nurses’ time to focus on other clinical priorities [4, 5].

We can divide POCT errors like central laboratory (CL) errors into the three main phases of the process: preanalytical, analytical, and postanalytical. Similar to the CL process, most errors occur in the preanalytical phase [6]. Even though the whole preanalytical phase is much simpler in POCT, there are several steps prone to error like test ordering, patient and specimen identification, and specimen collection and finally, specimen evaluation or assessment of specimen attributes [7, 8].

POCT programs properly led and organized by the clinical laboratory can help to prevent these errors. Also, strategies that guarantee operator identification before analysis and adequate performance with quality assurance plans and data transfer (connectivity) are essential [9]. Development, implementation, and validation of performance using reliable key performance indicators (KPIs) is recommended. The evaluation of KPIs over time could determine a set of quality indicators and the implementation of improvement actions with POCT governance led by laboratory medicine, to achieve safer and better patient care [3, 10, 11].

Blood gas analysis (BGA) devices are one of the most commonly used methods for POCT testing in emergency departments (ED) and for acutely ill patients in hospital or hospitalised patients, and represents an essential part of the diagnosis and treatment of acute critically ill patients [12].

In vitro, hemolysis typically occurs during the preanalytical phase particularly during the collection process itself especially in those clinical scenarios with a high stress environment such as the EDs and CCUs, or while transporting the sample to the analyser [13, 14]. There are well-recognised sources of hemolysis that can cause falsely elevated potassium results. Some of them are inherent to the process of taking samples for BG analysis (venous or arterial) and how the staff use the phlebotomy equipment [7, 15]. Hemolysis can alter the result of different parameters, including potassium with abnormally high results or masking low levels (hypokalemia) when reporting normal concentrations. Hyperkalemia is conventionally defined as a serum or plasma potassium concentration >5.5 mmol/L or >5.0 mmol/L, respectively. Severe hyperkalemia, usually defined as plasma or serum potassium greater than 6.5–7.0 mmol/L respectively is associated with a risk of potentially fatal cardiac arrhythmia and requires emergency clinical intervention [16]. We can divide its causes into [7, 15], [16], [17]:

1. Decreased renal excretion or increased intake of K⁺ (renal failure, increased potassium intake, or administration of certain drugs such as β-blockers, digoxin and potassium-sparing diuretics among others)

2. Decreased cellular entry of K⁺ or increased exit of K⁺ from cells (metabolic disorders)

3. Preanalytical problems such as blood sample hemolysis, prolonged tourniquet placement, fist clenching, high speed of drawing blood into a syringe, sample contamination from infusive routes, prolonged storage of uncentrifuged blood, or leukocytosis or thrombocytosis

Unlike chemistry analysers in CLs, almost no BGA analysers can detect the presence of hemolysis in the sample or report the hemolysis index, and due to the characteristics of the sample, we cannot visually detect hemolysis either [18], [19], [20]. Some alternatives have been published to solve this problem using machine learning algorithms [21]. There is already an external option for a hemolysis detection available, the Hemcheck Helge H10 system, which is a CE-marked hemolysis test at the POC with sensitivity/specificity data of 80 and 95 % based on a free hemoglobin concentration cut-off of 0.5 g/L, as shown by Duhalde et al. [20]. According to Pradhan et al. clinicians should be aware of these circumstances and the authors recommend the confirmation of elevated potassium concentrations by CL methods [22].

This study aims to highlight the importance of the influence of undetected in vitro hemolysis on the results of blood potassium analysed in POCT analysers and to demonstrate the possible impact it may have on an incorrect diagnosis. It was primarily designed to evaluate and quantify the discrepancies in potassium measurements between POCT and CL methods and to assess how these discrepancies could potentially impact patient management and treatment decisions in the fast-paced, high-stakes environment of the ED.

Materials and methods

Potassium and hemolysis gradient

To differentiate between the levels of K, results were divided into five classes: severe hypokalemia (<3.0 mmol/L), mild hypokalemia (3.0–3.4 mmol/L), normokalemia (3.5–5.0 mmol/L), mild hyperkalemia (5.1–6.5 mmol/L), and severe hyperkalemia (>6.5 mmol/L). This differentiation applied to both POCT as well as CL K (Figure 1).

Figure 1:

Potassium gradient as defined within the study showing the normokalemic, mild hypo-/hyperkalemic, and severe hypo-/hyperkalemic zones.

To differentiate between the levels of hemolysis, all three sites conducted individual analysis and set their own cutoff values according to their local clinical standard. Thus, H-I was divided into three categories: normal, mild, and severe hemolysis (Table 2).

Table 2:

Hemolysis gradient and normalization criteria.

	Madrid	Munich	Rotterdam
Equipment	Siemens Atellica	Cobas c8000	Cobas c8000
Type of sample	LiHep plasma	Serum	LiHep plasma
Normal	<100 mg/dL	<50	<42
Mild	100–200 mg/dL	50–150	42–100
Severe	>200 mg/dL	>150	>100

Results

Potassium hemolysis matrix

In clinical practice, hemolysis can lead to discrepancies between POCT-K and CL-K results which, in turn, might lead to erroneous over/undertreatment. We studied this impact utilising a POCT-K/CL-K concordance matrix as shown in Table 6 and Supplemental Tables 3 and 4. The matrix has been divided according to severity of alteration into three distinct zones/categories. The green zone/nominal where POCT-K corresponds with the CL-K result i.e., POCT-normokalemia = CL-normokalemia, the orange/subnominal zone where POCT-K corresponds with a step before or after the CL-K i.e., POCT-normokalemia = CL-mild hypo/hyperkalemia, and the red/severe subnominal zone where POCT-K shows severe discrepancy as compared to CL-K i.e., POCT-normokalemia = CL-severe hyperkalemia. From a clinical perspective, a change in category to abnormal might warrant further investigation or treatment, whilst a wrongful alteration into the normokalemic category might result in missing potential disease.

Table 6:

Summary of the concordance matrix.

1. The green zone/nominal where POCT-K corresponds with the CL-K result i.e., POCT-normokalemia = CL-normokalemia, the orange/subnominal zone where POCT-K corresponds with a step before or after the CL-K i.e., POCT-normokalemia = CL-mild hypo/hyperkalemia, and the red/severe subnominal zone where POCT-K shows severe discrepancy as compared to CL-K i.e., POCT-normokalemia = CL-severe hyperkalemia.

In the Munich dataset, 79 % (n=8,572) of the paired samples showed a good correlation in the green/nominal zone between POCT-K and the CL-K. The remaining 21 % (n=2,262) were in the orange/subnominal zone and 0 % (n=5) were in the red/severe subnominal zone. Those five patients in the red/severe subnominal zone might have received erroneous over- or undertreatment options. In the Madrid/Rotterdam location, 89.4 % (n=9,254) of the paired samples are in the green/nominal zone. Unfortunately, 9.9 % (n=1,018) and 0.7 % (n=73) were categorized in the orange/subnominal and red/severe subnominal zone respectively. This means that 1,018 patients’ diagnosis and treatment (9.9 %) might have been impacted by erroneous results and could get an expectant attitude (wait and see and eventually treat) while 73 patients (0.7 %) may have received erroneous over- or undertreatment.

Reference change value

The RCV was calculated as described above. Our data show that ±8 % of all paired potassium results as analysed on POCT and in the CL were outside the RCV ranges (Figure 2). This implies that these patients could have been given an incorrect diagnosis potentially resulting in a lack of the correct needed care.

Figure 2:

The RCV between POCT-K and CL-K. The graph depicts the percentage (%) of the paired K+ results outside the RCV.

Discussion

Our retrospective study conducted in EDs of three different European university hospital centers clearly shows that in vitro hemolysis is a common and potentially serious problem for POCT potassium results. From our big data analysis, we were able to describe the proportion of patients having a significantly altered potassium plasma/serum concentration (>5.0 mmol/L), presumably caused by sample hemolysis.

The normalized data sets from the Madrid plus Rotterdam (total n=10,245) sites were compared with those from Munich (n=10,842) with comparable total analysis numbers (10,245 vs. 10,842; Table 1). When focusing on the mild and severe hyperkalemia cases in Table 3, the percentages between Madrid plus Rotterdam, and Munich for the POCT measurements were comparable (6.23 vs. 4.8 %), whereas for the CL determinations, a significantly higher percentage of Munich vs. Madrid plus Rotterdam (10.91 vs. 7.01 %) can be stated. Since in Munich most of these cases are mild hyperkalemia (10.61 vs. 6.47 %), it can be presumed that the use of serum instead of plasma can explain this difference.

The gross differences in the sum of mild and severe hemolysis determined in the CLs between Madrid plus Rotterdam vs. Munich, described in Table 4, can be explained as effect of the sum of different complex conditions and circumstances connected with preanalytical issues, as described by Buño et al. and Schlüter et al. [7, 8], organizational and operational issues, the use of different specimens (plasma in Madrid and Rotterdam vs. serum in Munich), as well as selection and handling of admitted patients. Using a pneumatic tube system to transport the samples in Munich does not seem to influence these differences as previously demonstrated by Zanner et al. [23].

However, despite the above-described differences between the three study sites, the found percentages of hyperkalemia in hemolyzed samples, measured by POCT, are rather comparable (Table 5). These data impressively confirm the observations of other authors (see below) for hemolysis in BGA samples [12, 14, 20, 24], [25], [26], [27], [28], [29].

The data from the concordance matrix showed differences between Madrid plus Rotterdam vs. Munich, leading to an over- or under-estimation of the potassium level. Fortunately, a good agreement was found for the RCV analysis (±8 %) in the three sites of the paired potassium concentrations analysed either in the CL or at the POC.

Our findings are consistent with several other studies confirming the observation of hemolyzed patient samples as a common problem with BGAs under stressful and fast-paced conditions in the ED. Also important is the fact that the blood collection procedures and the preanalytical sample handling are undertaken by a variety of medical professionals with different training statuses [4]. It has already been pointed out that preanalytical POCT errors are inversely correlated with user experience [30, 31]. When assessing these studies, one must first determine which patient population and clinical setting the evaluation refers to in each case: outpatients, inpatients at an ED or ICU or on a normal ward.

We found five studies investigating the prevalence of hemolysis or hyperkaliemia in blood samples drawn in the ED. Singer et al. found 3.6 % hyperkalemic and 3.6 % hemolytic samples in 48,827 ED-patient visits [25]. The study of Tazmini et al. reports on 62,991 ED visits and found in 3.3 % of all samples a hyperkalemic status [26]. Another already mentioned trial found that a rate of 7.9 % hemolysis was found in 1,270 observed blood gas analyses from an ED [20]. Differences were seen between different sites in terms of hemolysis in a trial of Wilson et al. [27]. They observed 100 blood gas analyses and found a hemolysis rate of 13 % in the ED and 4 % in an ICU. Finally, Nigro et al. reported that in 472 arterial samples hemolysis was present in 12 % [28].

Casati et al. [12] found 5 % hemolysis in 1,244 observed BGAs, samples from different hospital wards and the ED, whereas Salvagno et al. reported a rate of 4 % hemolysis in 487 observed BGAs from routine and stat samples from a hospital [24].

An interesting study by O´Hara et al. deals with hyperkalemic specimens, showing that in consecutively collected 100 hyperkalemic samples, 40 % showed significant hemolysis [14]. Grieme et al. investigated the impact of various interferences on parameters in POCT and found a similar result: there was evidence of hemolysis in 60 % of samples with potassium serum concentrations from 6.0 to 6.9 mmol/L [29]. The discrepancy in the study of O’Hara is likely due to the application of different hyperkalemia cutoffs.

A critical issue for considering measured hyperkalemia as an indicator of hemolysis is knowledge of the incidence of true hyperkalemia cases (caused by acute or chronic renal failure, heart failure, potassium supplementation, certain drug effects, insulin deficiency, or metabolic acidosis) in patients admitted to the ED. Singer et al. found an incidence of hyperkalemia of 3.6 % in 47,089 patients [25], whereas Kuo et al. only found 0.92 % in 602 investigated patients [32]. Lemoine et al. summarized data from the literature with incidence rates between 1 and 10 % [33]. In our study, we found a tendency to hyperkalemia but even normokalemia in a significant proportion of samples with severe hemolysis analysed with POCT. Whilst hyperkalemia and hemolysis have been observed and discussed for decades, the high incidence of hemolysis resulting in “normal” readings might be a concern. Hypokalemia is a common clinical finding in the ED (for example in frail elder patient), and pseudonormalized results due to hemolysis seem to be common. In consequence, patients with clinically relevant hypokalemia might be missed and left untreated.

These clinical studies indicate that among hyperkalemic BGA samples, the percentage of falsely elevated potassium concentrations is alarmingly high, and this finding is disproportionally often in the ED. The underlying reasons were already mentioned above. Therefore, it is not surprising that Lippi et al. reported 1.2 % hemolytic samples in 1,228 BGAs from clinical wards with the exclusion of ED and ICU sites [34]. This also fits with the observation from Jose and Preller et al. that clinicians – in particular, intensivists – trust more the potassium results from the CL than their own POCT results [35]. Hemolyzed samples at the BGA pose also additional problems. First, pseudohyperkalemia may mimic low serum potassium concentrations. In this case, the BGA reports normokalemia whereas the patient is hypokalemic (Supplemental Tables 3 and 4). Hypokalemia is found often in hospitalized patients and is associated with excess morbidity and mortality. The lower the potassium serum concentrations, the higher the risk, starting at potassium <4.0 mmol/L, with a significant risk increment when the potassium levels are <3.5 mmol/L, as described by Ferreira et al. [36]. Secondly, spurious hemolysis affects also various other BGA parameters, such as pO₂, pCO₂ or Ca²⁺ [29, 37].

In the literature, the prevalence of unsuitable specimens referred for BGA is given at 1.2–3.7 % [24, 34, 37], with hemolysis accounting for 40–70 % of the cases [38]. The Clinical and Laboratory Standards Institute (CLSI) C46-A2 guideline [39] generically advises against processing unsuitable blood samples for BGA. However, this requirement can only be met by providing the best possible training for operating personnel to accurately identify and appropriately manage spurious sample hemolysis [34].

Regarding the limitations of our study, different points compromise the value of our data: first, the paired blood samples for BGA at Munich were sent to the CL while in Madrid and Rotterdam were analysed as POCT. Also, the time frame of 45 min between the POCT BGA measurement and CL is questionable, although were considered a pragmatic and realistic approach. Second, the majority of over 21,000 samples were venous samples. However, we do not know the minor portion of arterial samples. Third, the three sites used different chemistry analysers for the measurement of potassium and used also different cut-offs for the HIL indices (even between Madrid and Munich, due to the use of plasma or serum) defining the hemolysis rate. These cut-offs had been previously established for routine analysis in all three sites and had proved successful in daily measurements by providing an effective screening tool for checking specimen quality concerning hemolysis, hyperbilirubinemia, and lipidemia [40]. In addition, the cut-offs had been compared and harmonized with measurements of free hemoglobin in plasma (data not shown). Fourth, the design of the study is retrospective and can be more susceptible to bias and confounding variables compared to prospective studies. Fifth, we cannot assure that the patient population at the three hospitals were comparable in terms of the pathology presented at the ED.

Conclusions

This multicenter study is the largest to date investigating the true incidence of hyperkalemia in hemolyzed samples in POCT in the ED. The observed rate of 7–9% is in good accordance with previous reports concerning hemolysis in BGA specimens. However, this study has much more statistical power due to the enormous number of evaluated data sets and its multicenter design.

The hemolysis rates presented in this trial and the other published clinical studies shows that there is still a major need for improvements in the preanalytical phase of BGA [41, 42]. There are important demands for dedicated training and revalidation for ICU and ED personnel in terms of preanalytics [31]. Another mandatory requirement for future BGA devices is the implementation of an inbuilt hemolysis detection system [12, 43], a request that is sustainably endorsed by our study. Until then other alternatives could be used to minimize the problem such as the use of predictive analytics, or the Hemcheck Helge H10 system and the stringent application of a Failure Mode and Effects Analysis (FMEA) protocol by the POCT quality management team [20, 21, 44].