Are commonly used paediatric reference intervals for water and electrolyte balance appropriate for clinical use?

Abstract

Background:

Paediatric reference intervals are less well characterized than in adults. An initiative for harmonization of pathology across the United Kingdom has recommended an interval for sodium of 133–146 mmol/L at all ages.

Methods:

To assess the validity of this, the laboratory database was interrogated for all renal profiles (sodium, potassium, urea and creatinine) for children presenting to primary care over a 13-year period. While the primary interest was in sodium results, sufficient current data were also available for potassium and creatinine and so these were included for study. The electrolyte results were filtered to include only normal renal function and the remaining data were analysed for age-related differences.

Results:

Sodium concentrations were observed to be lower for infants (1–5 years of age) with a mean of 138 mmol/L, increasing towards adult concentrations (mean 140 mmol/L) by teenage years. A similar pattern was seen for potassium results, and creatinine was seen to increase with age. At all ages, the distributions of sodium concentrations measured in this population were observably tighter than the interval of 133–146 mmol/L recommended by Pathology Harmony.

Conclusions:

We suggest that this interval is too wide, and more work is needed to establish more appropriate paediatric ranges.

Introduction

Discrepancies in reporting of results between different laboratories have long been raised as a source of possible confusion between clinicians working at sites using different laboratories, or for patients who move between sites regularly. ¹ A UK initiative for Pathology Harmony ² has provided useful leadership for the harmonization of reporting units and reference intervals, and recommended implementation of these harmonized intervals nationally by April 2011. However, a recent case, in which a child presented in Addisonian crisis, but in whom the only preceding biochemical abnormality four weeks earlier had been a mild hyponatraemia of 132 mmol/L (133–146 mmol/L), raised questions about the validity of these ranges in children.

Adrenal insufficiency classically presents with fatigue, weight loss, hyperpigmentation and hypovolaemic hyponatraemia with hyperkalaemia. ³ Its onset is usually slow and this, combined with the non-specific signs and symptoms, often leads to long delays in diagnosis. Of even greater concern is a missed diagnosis, which can eventually present as a full-blown adrenal crisis carrying with it a high risk of death.^4–6 Hyponatraemia is often mild or absent at diagnosis,^6–12 both in children^9,12 and in a variety of other patient groups; however, its presence in a patient with suspicious symptoms may be enough to prompt the clinician to consider the diagnosis.^4,6,13 Consequently, accurate reporting of sodium concentration by the laboratory is essential.

Hyponatraemia can also arise due to a variety of other, more common causes, and is not, in itself, a reliable indicator of adrenal insufficiency. ¹¹ Thus, although the presenting serum sodium in this case was outside the reference interval, it would not necessarily have triggered alarm, especially as it is often quoted that many reference intervals are based on 95% confidence intervals,^14,15 implying there is a 5% chance of a result lying outside the reference interval, even in a healthy individual. This prompted the authors to query the quoted reference interval.

While Pathology Harmony quote an adult range for plasma sodium of 133–146 mmol/L as being appropriate for all ages, ² no evidence is quoted in support of this. In contrast, data from the National Health and Nutrition Examination Survey (NHANES) are available for children from the age of 14 years upwards, and suggests reference intervals far tighter than those recommended by Pathology Harmony for sodium (136–144 for females, 139–144 for men) between the ages of 14 and 18 years. ¹⁶ In 2007, the National Patient Safety Agency (NPSA), prompted by a series of childhood fatalities caused by hyponatraemia, ¹⁷ issued an alert which defined hyponatraemia as a plasma sodium under 135 mmol/L and identified a plasma sodium under 135 mmol/L as a risk factor for the development of life-threatening hyponatraemia. ¹⁸ However, very little good quality evidence regarding electrolyte concentrations in the paediatric population exists, presumably because it is generally assumed that there is no reason for concentrations in children to differ from adults. High-quality, large-scale studies such as the Canadian CALIPER initiative ¹⁹ to establish paediatric reference intervals have therefore often not included electrolytes.

While guidelines exist regarding appropriate methods to define reference intervals,^15,20 methods used vary significantly in the literature, with many different approaches available for the partitioning of data^20,21 and for the derivation of the reference interval itself. ²⁰ Large amounts of data are routinely accumulated in the laboratory database and can be used to directly derive reference intervals.^22,23 Many renal profiles are requested in the absence of overt renal disease or conditions which cause a fluid or electrolyte imbalance, particularly in a paediatric population which is less likely to be on diuretics or other medications compared to adults. If appropriately filtered, laboratory data may be used to illustrate firstly how appropriate the current UK reference intervals are, secondly whether the assumption that there is no difference between adult and paediatric is valid, and finally how unusual it is to see a sodium concentration such as that seen in the above case.

Methods

Renal results (sodium, potassium, urea, creatinine) were extracted from the laboratory database at the University Hospital of Wales for all patients under 18 years of age. The laboratory at this hospital provides a service to a mixed primary and secondary care population of approximately 470,000 people. Plasma and serum samples are treated identically with no differentiation between the two in the laboratory database. Urea was not routinely provided as part of the profile for primary care during the later part of the data gather period, and so was used only for filtering of results. Two time periods were examined: a 13-year period from January 2000 to June 2013 (n = 318,408), and due to instrumentation changes over this period, a three-year period following the installation of current instrumentation (March 2010 to March 2013, n = 80,921). These were filtered by requesting location, including only those requested in primary care (n = 29,037 and n = 9292 for the 13- and three-year periods, respectively). Results were then stratified by age, and in the case of sodium and potassium, filtered for presence of a creatinine within the local reference interval as a surrogate marker of normal renal function. All samples with any record of haemolysis (haemolysis index 0.75 or greater, corresponding to a visual definition of ‘slight’ haemolysis) were excluded from the potassium dataset. Potassium and creatinine concentrations were log transformed. The frequencies of different analyte concentrations for each age were modelled to a Gaussian distribution to exclude outliers. Where necessary due to low patient numbers, consecutive age groups were combined if no clear difference could be seen when the frequency distributions were visualized. Mean and standard deviation (SD) were calculated. With the 13-year dataset, a bootstrapping approach was used to estimate error, whereby the dataset was divided into four groups with roughly equivalent numbers of samples. Each set was then processed as described above, and the SDs of results produced across the four datasets were used as an estimate of error. Calculations to determine partitioning for the final derived reference intervals were performed according to the distance method described by Lahti et al. ²¹

All analytes were measured on automated analysers. From March 2010 onwards, this was the Abbott Architect (Abbott Laboratories, Illinois); previously the Abbott Aeroset was used. Electrolytes were measured by indirect ion selective electrode, and creatinine by kinetic Jaffé as provided by Abbott Laboratories. Data were extracted from the routine laboratory database. These were processed using custom scripts written in the programming language Perl to clean the dataset, filter results and produce frequency distributions. Ages were treated as entire completed years of life at time of sample receipt, e.g. 20 months would be considered one year of age. Bin sizes for frequency distributions were 1 mmol/L for sodium, 0.2 mmol/L for potassium and 2 μmol/L for creatinine. Due to the proximity of the distributions to zero for both potassium and creatinine, these distributions were log transformed for modelling to Gaussian distributions. Sodium distributions were sufficiently distanced from zero for log transformation to be unnecessary. Modelling of frequency distributions to Gaussian curves with non-linear regression was performed in Excel using a sum of least-squares scoring system to optimize fits in a similar manner to that described elsewhere. ²⁴ Final reference intervals based on this population were created through combining age groups as deemed appropriate through the distance partitioning method and visual observation of distributions, followed by modelling to Gaussian distributions as described above.

Results

Sodium

Changes in monthly mean paediatric sodium of approximately 1 mmol/L were observed when changes of instrumentation occurred, and so the three-year dataset was used, to avoid complication of results by inclusion of more than one method. Patterns were confirmed by comparison with the full 13-year dataset, using both creatinine and urea as indicators of normal renal function. In total, 8164 sets of results met the filtering criteria. Insufficient data were available for children in the first year of life (n = 12), and so this age has been excluded from analysis. Distributions of sodium concentration fitted a Gaussian curve well for all age ranges, with R² values ranging from 0.986–0.999 (Supplementary Table 1). At each age, the SD of distribution of sodium results is broadly similar, but differences are seen with age. There is a clear trend for younger patients to have a lower sodium concentration than older patients (Figure 1(a)). With a SD of sodium distributions of approximately 1.5 mmol/L for most ages, a difference in reference intervals of more than 0.4 mmol/L (a quarter of the SD) potentially justifies partitioning, and a difference in intervals of more than 1.2 mmol/L (three-quarters of the SD) definitely justifies partitioning according to the distance method. ²¹ According to these criteria, the observed difference between younger ages (1–4 years of age) and teenage years (12–17 years of age) of 2 mmol/L for both upper and lower reference intervals would definitely require partitioning for sodium. While the population as a whole has a distribution of sodium concentrations not dissimilar to the Pathology Harmony reference interval, 95% age-specific reference intervals calculated from this population are significantly tighter (Supplementary Table 2, Figure 1).

Figure 1.

Graphical representation of mean (solid line with diamonds), two SDs from mean (dashed line with diamonds) and three SDs from mean (dotted line with diamonds) to indicate calculated points across different ages. Ages are completed years of life. Current reference intervals in the source laboratory (in line with Pathology harmonization recommendations) are displayed as a grey area on plots. (a) Sodium: values taken from Supplementary Table 1. Points for combined age groups are shown on the border between ages. (b) Potassium: values derived from Gaussian fits assuming a constant SD of 0.0793 from the natural logarithm of the mean, as described in text. (c) Creatinine: linear fit to values from Supplementary Table 1 shown instead of directly calculated values; individual points therefore not drawn for fit. Deviation from linear fit for male creatinine from 13 years of age upwards shown in grey.

Potassium

Seasonal changes in monthly mean potassium results were observed, with a winter peak of monthly mean potassium typically being 0.2 to 0.3 mmol/L higher than a summer trough. No other trends could be seen over longer time periods, implying no change in assay performance, and so the full 13-year dataset was used, along with crude bootstrapping to estimate error. Similar results were observed if the dataset was reduced to the three years used for sodium. A total of 24,879 potassium results were available from non-haemolysed samples meeting the filtering criteria. For the first year of life, data available were insufficient to estimate reference intervals (n = 33) and appeared too diverse to combine with other ages. Variation in the SD of the distribution was significantly higher at younger ages. The availability of additional years of data for this analyte enabled bootstrapping to be performed. This showed that this variation in SD of the frequency distribution was likely to be random noise, and implied a constant SD across all age groups (Figure 2). A consensus (achieved by calculating the mean of the SD for all subgroups) SD of 0.0793 ln (mmol/L) was therefore used to fit all Gaussian curves for potassium.

Figure 2.

SDs at each age for the four replicate fits calculated from subgroups of the main potassium dataset, showing the increase in variation of SD at lower ages. Different style points for each of the four subgroups.

All ages fitted a Gaussian distribution well with R² values from 0.953–0.999 (Supplementary Table 1). Data fitted better at older ages; above the age of eight years, all R² were 0.993 or more. A similar trend with age is seen to that observed for sodium, with younger children having lower potassium concentrations than teenagers; however, in infants from the age of three years downwards, there is a slight trend towards increased potassium concentrations (Figure 1(b)).

Application of partitioning criteria, as applied to sodium, shows that for the two SD upper limit calculated for a 16-year-old child, a difference in upper two SD limit from this of more than 0.1 mmol/L potentially justifies partitioning and a distance of more than 0.3 mmol/L definitely justifies partitioning. The differences observed in these data fall into the grey area where partitioning of subgroups may be justified. Calculated 95% reference intervals are generally comparable to a reference interval of 3.5–5.0 mmol/L as recommended by Pathology Harmony (Figure 1(b)). However, final reference intervals calculated from these data have a slightly lower upper limit between ages two and six years of age, and a slightly higher lower limit from the age of 11 years upwards (Supplementary Table 2).

Creatinine

Due to changes in creatinine methods over the time period of the larger dataset (a gradual reduction of approximately 20–30% in mean creatinine was seen over this period), the three-year dataset was used. As creatinine could not be filtered for renal function, a total of 9272 results were included. The R² of fits to Gaussians ranged from 0.942–0.994 (Supplementary Table 1). Clear trends for increasing creatinine with age were observed, and so partitioning calculations were not required. These trends fitted a linear function well (all deviations from fit were 5% of values or less) for females of all ages and for males until 12 years of age. From the age of 13 years upwards, differences were observed between male and female populations, with males having higher creatinine concentrations than females. Creatinine concentrations for males aged 17 years were not log-normally distributed; therefore, mean and SD were not calculated for this group. Final possible reference intervals (based on the linear fits for females and for males until the age of 12 years; directly derived from the population data for males aged 13 years upwards) are provided in Supplementary Table 2.

Discussion

Bias from patient selection

Historical data

While these data are derived from historical data, and relevant intercurrent illness cannot be fully excluded, the filtering applied should have excluded populations of patients with significant renal dysfunction. Filtering for normal renal function with creatinine strongly reduced the variation seen for electrolyte distributions, confirming the importance of this step. Ideally, the dataset would have been filtered to include only those results where both urea and creatinine were within quoted reference ranges; however, this was not possible for recent data, as urea was removed from our laboratory's primary care renal profile in 2010. To determine the validity of using creatinine concentration alone to exclude significant renal dysfunction, we compared sodium distributions derived since the introduction of the Abbott Architect assay in 2010, filtered by creatinine alone, with those derived from 10 years of older data (2000–2009) when urea was routinely included in all renal profiles, filtered by both creatinine and urea (data not shown). The distribution patterns were identical for both cohorts, although assay changes over time made the direct use of the combined cohort inappropriate for defining a current reference interval for either sodium or creatinine.

Selection bias

Paediatric patients are not routinely bled without good cause. The population these results were derived from is therefore unlikely to be entirely healthy, and it may be that the age differences observed for sodium concentrations are simply the result of more significant illness in the younger cohorts, as in general, phlebotomy is not undertaken lightly in this group. Certainly, the lower number of younger patients in the laboratory database supports this assumption (Supplementary Table 1). However, we do not believe that the pattern of sodium distributions observed in children can be explained by illness, as not every illness will result in a similar disruption of electrolytes. Therefore, the distribution of results in a heterogeneously ‘ill’ infant population would be expected to have greater variance than a ‘healthier’ young adult population. This was not found to be the case if sodium distributions in older and younger paediatric patients are compared (Supplementary Table 1). Conversely, when paediatric sodium results from hospital inpatients were included, we found little difference in mean concentrations, but variance, and hence, SD increased; this was mainly due to an increased incidence of hyponatraemia (data not shown).

Preanalytical factors

No record of sample type (serum versus plasma) is made in the laboratory database. However, the majority of adult samples are serum, and the majority of paediatric samples are plasma. It is, therefore, possible that the age-related variation seen for electrolytes may well be related to the slightly reduced sodium and potassium measurements that occur with indirect ion selective electrodes for plasma samples compared to serum. Irrespective of the cause, the difference between sodium results observed for children below the age of five years, and for children above the age of 10 years, is clear-cut enough to be relevant when interpreting results in practice.

In the case of potassium, it is likely that preanalytical factors have an impact on the results obtained for the very young (under four years of age). Samples from younger patients show an increased concentration of potassium, in contrast to the trend towards higher potassium with age in children over five years of age. This may be due, at least partly, to difficulty collecting blood samples in younger children, which could result in greater release of potassium due to venous stasis or mild haemolysis.

This is likely to also be the cause of the increased variation in SD at younger ages (Figure 2), compounded by the reduced amount of data available for younger ages. At younger ages, there are varying degrees of outliers with higher potassium concentrations, an effect which is limited by the process of fitting to a Gaussian distribution. However, when high numbers of outliers are present and the number of results in the population is low, the outliers can merge with the main frequency distribution and artefactually increase the SD of the distribution. This effect should be negated with the use of a constant SD for all ages that was produced by bootstrapping. At ages over seven years, all frequency distributions fitted well to a Gaussian model, with minimal outliers, as would be expected with the additional robustness provided by increased numbers of patients in the datasets at older ages. Similarly, variation in SD was far lower for these age groups.

Reference intervals and age variations

Sodium

We calculated sodium reference intervals that included 95% of the reference population (i.e. Gaussian mean ± 1.96 SDs), and found them to be far tighter than any in current use around the United Kingdom.

For patients aged 17 years, the interval was found to be 137–143 mmol/L (Figure 1), which could pose some difficulty with interpretation as patients moved from paediatric to commonly used adult reference intervals. By extending this to mean ± 3 SDs, a range of 135–145 mmol/L was obtained. This is much more in keeping with existing adult reference intervals, and has the advantage of being more familiar to users.

Although there is little peer-reviewed evidence for paediatric sodium reference intervals, a lower sodium in younger children is consistent with other published reference intervals,^25,26 giving us confidence that these data are representative of a typical population, although it should be noted that at least one of these reference intervals is based upon hospital inpatient data. Furthermore, a recent Norwegian reference interval study did include electrolytes ²⁷ ; however, it included very few children below the age of seven years, which is unfortunate as our experience suggests that reference intervals in this group suffer the greatest variability. Thus, although the final non-parametric reference intervals were not separated by age, the youngest age group (5–6 years of age) did, in fact, produce a reference interval with a lower limit comparable to that seen here.

Hopefully future studies, using a priori selection of healthy patients, will consider including sodium as part of their reference interval investigations. This should help to clarify whether age-specific differences in sodium concentration are genuine.

Potassium

Variations in sample quality are likely to contribute to uncertainty over the appropriate reference interval for younger patients; however, reference intervals can be calculated assuming constant variance for each age-specific frequency distribution. Pathology Harmony currently recommends a reference interval of 3.5–5.0 mmol/L for children aged between one and 16 years. These data indicated that some fine tuning for different ages may be appropriate, with partitioning calculations being inconclusive. Possible reference intervals have therefore been calculated for three age groups (Supplementary Table 2), although in practice the many preanalytical factors which can affect potassium concentration may overwhelm any age-related differences. Notably, seasonal variation has been previously shown to have a dramatic impact on measured potassium, ²⁸ and even within season, exceptional weather can cause further deviations in potassium concentration.^29,30

Creatinine

Frequency distributions show clear differences at each age of life which suggests age-related reference intervals are appropriate. Furthermore, from the age of 13 years, gender differences are evident, suggesting a need for further division into gender-specific ranges. However, creatinine assays are less standardized than those for sodium or potassium, and it may not be appropriate to extrapolate the data provided here to assays other than the kinetic Jaffé on the Abbott Architect. Notably, when changes over time were examined for the full 13-year dataset, a reduction in mean creatinine of approximately 25% was observed between 2001 and 2010, despite no known changes in the patient population. This change was attributed to assay changes that occurred over that time period. However, the methods provided here could be used by any laboratory with sufficiently large throughput of samples from primary care to identify whether our findings were replicated with their assay. Despite the lack of clinical filtering of results, frequency distributions fit to Gaussian distributions well, with discordance of fits attributable to random noise rather than any visible systematic variation in the population for almost all ages. The only exception to this is males of age 17 years, where a positive skew is seen, possibly corresponding to variations in muscle mass in this population which are seen in adulthood. Reference intervals have therefore not been calculated for this group.

Choice of reference interval definition

Both traditional (1.96 SDs from mean) and alternative (3 SDs from mean) reference intervals have been calculated from the data shown. It is important to note that a reference interval at three SDs from the mean will include 99.7% of the reference population rather than 95%. For sodium, particularly, the three SD reference intervals calculated represent the distribution of results seen extremely accurate, and, consequently, results outside of the intervals described should be investigated at a far lower index of suspicion. It is likely that a similar low index of suspicion would still be applicable for young adults, whether the 99.7% reference interval of 135–145 mmol/L was used, or even more so with the Pathology Harmony interval of 133–146 mmol/L.

The use of three SDs from the mean is more similar to existing screening decision limits than to traditional reference intervals. There has been some suggestion in the literature that use of traditional 95% reference intervals is not compulsory, and more stringent criteria may be appropriate when a greater specificity is required.^14,19 As the low proportion of results outside the typical frequency distribution indicates that renal profiles are being requested more as a screen than a genuine query for renal disease, a more specific interval than the 95% reference interval commonly used may be appropriate. Irrespective of any disagreement regarding the appropriate definition of a reference interval, we provide the mean and SD (natural logarithm transformed in the case of potassium and creatinine) at each age, to enable calculation of any desired interval (Supplementary Table 1).

Summary

We have provided data on the values typically seen for sodium, potassium and creatinine in a general practitioner paediatric population, and while this work was not intended to generate reference intervals for clinical use, possible potential reference intervals based upon these data have been provided as illustration that existing reference intervals for electrolytes may not be as robust as commonly believed. Our work indicates particularly that a reference interval for sodium of 133–146 mmol/L is unnecessarily permissive, and suggests that paediatric reference intervals for sodium should likely be significantly tighter than this. It is clear from these data that a sodium of 133 mmol/L is highly unusual in a child over 10 years of age, and equally that a sodium of 146 mmol/L is highly unusual in a child under five years of age. In fact, further analysis using the number of SDs from the mean has shown that, for a child above the age of 10 years, a sodium concentration of 132 mmol/L, as seen in the case which prompted this work, has odds of approximately a million to one against being a ‘normal’ result. Clearly, this is a rather different message to that which is implied with a combination of the classical perception of a reference interval (95% of the reference population) and the Pathology Harmony reference interval of 133–146 mmol/L. We would therefore suggest that there is significant value in laboratories knowing and indicating, when quoting a reference interval, how that interval was derived, to enable clinicians to interpret results correctly. We also believe that current Paediatric Pathology Harmony sodium reference intervals are too wide and there is a risk that cases of clinically significant hyper- or hyponatraemia may be missed. In light of the increasing evidence available for reference intervals, for paediatric patients in particular, it may be worth re-evaluating Pathology Harmony recommendations to include available evidence, and to consider whether further work is required where good quality evidence is lacking.