Reference intervals for iron-related blood parameters: results from a population-based cohort study (LIFE Child)

Introduction

Iron represents in quantitative terms the most abundant essential trace element in people, and with enzymes dependent on it, is involved in important metabolic processes, such as oxygen binding and distribution via hemoglobin and myoglobin, transmitter production or cytokine induction. Inadequate iron intake can have many causes and manifests itself in a lack of iron, or even iron deficiency anemia, which is the most common form of anemia [1]. Clinically, it manifests itself in children early, unlike in adults, due to increased demand during growth. The symptoms are wide-ranging and may even include inhibition of physical, cognitive, mental and motor development [2, 3]. Contrary to outdated assumptions, iron deficiency anemia is detected not on the basis of a lowering of iron levels, but on the basis of a decrease in hemoglobin and ferritin concentrations, which are below the age-appropriate mean values by more than two standard deviations [4]. Latent iron deficiency may already be detected through a decrease in ferritin and an increase in transferrin levels while hemoglobin remains normal. A decreased reticulocyte count in addition to a drop in hemoglobin levels indicates a manifest iron deficiency [5, 6].

Reference intervals are used for the clinical interpretation of laboratory results and are thus the basis for clinical decisions. Particularly in pediatrics, age-related reference intervals are required to take adequate account of physical development, nutrition, growth, organ maturity, hormone and immune responses as well as diseases during specific age segments. After reference intervals had been used for iron-dependent parameters for a long time, which had been determined either several decades ago or on the basis of hospitalized test subjects, there have been a growing number of efforts in recent years to devise current reference intervals [7, 8]. These efforts have been informed by population-based data on children of different ages in the USA (NHANES, 1994; AACC, 2007) [9–11], Ireland (1997) [12], Canada (CALIPER, 2009/10/12) [13–15], Germany (KiGGS, 2009 [16]; Zierk et al., 2013 [17]), Denmark (2012) [18] and Sweden (2013) [19]. Among the total of eight studies mentioned, reference intervals were specified for hemoglobin in six, for ferritin in four, for reticulocytes in two, and for transferrin in four of the studies. Furthermore, only the German studies have so far determined reference intervals for hemoglobin and ferritin in a continuous manner on the basis of age, including reference percentiles, rather than using age categories. There are still no continuous reference intervals for transferrin and reticulocytes. However, these parameters have until now been considered parameters with poorly confirmed information regarding reference intervals, a situation due to standardization deficits in measurement procedures as well as due to a lack of or poorly verified basic data, particularly with respect to children [20]. As a result of these gaps described, it is the objective of this paper to determine age- and gender-dependent reference intervals, including percentile curves, for the parameters hemoglobin, reticulocytes, ferritin and transferrin. This is done on the basis of a large study population by means of current and highly standardized laboratory methods as well as in applying the Generalized Additive Models for Location, Scale and Shape-R (GAMLSS-R) package, which is still relatively unknown in laboratory medicine.

Materials and methods

Study design and participants

The data of participants in the LIFE Child study, aged 2.5–19 years and from the region of Leipzig, have been used. LIFE Child is a sub-project of the “Leipzig Research Center for Civilization Diseases” of the University of Leipzig with a longitudinal/cross-sectional study design. The study received a positive assessment from the competent ethics commission (Reg. No. 264-10-19042010). At the beginning of each visit, the test subjects and their parents received information from a study physician before they had to sign a consent form, which was a requirement for participating in the study. The data of the test subjects were pseudonymized to prevent any conclusions about the individuals [21]. The participants were not part of a clinical cohort, in other words, the computation of reference intervals was primarily based on healthy children and adolescents. Only those measurement time points have been factored into the data analysis at which full readings were available for hemoglobin, ferritin, reticulocytes and transferrin. This yielded 2778 measurement time points of 1779 test subjects representing 1346 families. Figure 1 shows the age and gender composition of the random sample at each first measurement time point. As regards weight distribution, the majority of children (78%) had a normal weight, while 7% were underweight, 7% overweight, and 8% were considered obese. The average age-adjusted BMI was 0.1, with a minimum value of –4.6 and a maximum value of 4.8. As the age adjustment is based on references according to Kromeyer-Hauschild [22], the minor increase in the mean value can be attributed to the increasing trend of recent years that has seen children become more overweight. It was impossible to rule out that the parameters ferritin and transferrin were affected by potential infections, which is why the same calculations were done for this with the measurement time points of a simultaneous increase in CRP of ≥5 mg/L (n=84) being excluded. In general, the CRP increase was ≥2.8 to 5.0 mg/L in 5%, and ≥5.0 mg/L in 3% of the study participants. For 92% of the participants, the CRP level was within the reference interval. Furthermore, a medication-based iron uptake was documented at 0.4% of the measurement time points (n=10). These were not excluded due to their small number.

Figure 1:

Histogram for age and gender distribution of the reference population from the LIFE Child cohort regarding the first visit (total: n=1779, boys: n=903, girls: n=876).

Statistical analyses

The choice of statistical methodology was based on the IFCC guideline for establishing reference intervals [25] and on the WHO recommendations on the estimation of percentiles [26], which were worked out for anthropometric parameters, but are also increasingly used for other parameters. As recommended by the IFCC, the lower and upper reference values were the percentiles 2.5 and 97.5 [25]. A modified LMS (λ-μ-σ) method according to Cole, Rigby and Stasinopoulos taken from the WHO recommendations was used [27, 28], which is also implemented in the GAMLSS package of the statistics R software [29–31]. This was used on the assumption of a Box-Cox power exponential distribution (BCPE), which can map normal distributions as well as steep and flat peaks, making it highly flexible. This way, it was possible to achieve a normalization of the age-based distribution, which is a requirement for determining reference intervals. These distributions are characterized by the parameters median (location parameter, μ, M), coefficient of variation (scale parameter, σ, S), skew (λ, L) and curvature (shape parameters) as a function of age [32, 33]. The approach of estimating the parameters as a continuous function of age was preferred, because it provided a better model for physiological development than a step-by-step consideration of different age intervals [27]. There were several measurement time points for each subject as a result of the longitudinal design. At the same time, the cohort consisted of 1346 families, some of whom participated with several children. To ensure the independence of the measuring points, as well as to be able to include all measuring points (and, thus, all information) in the model, a resampling technique was applied. Sub-samples were assigned in a first step by limiting the families to 600, followed by the selection of a measured value for each family. It was necessary to limit the number of families, because some families had only one measured value, and because the probability of factoring such a measured value into the calculation was to be <1. The sub-sample size of 600 values proved to be an acceptable subset in this case to allow for a maximum number of measured values while maintaining an adequate sampling selection. The sampling of a measured value per subject during multiple visits was congruent to the family sampling. The weightings were chosen in such way that all measured values were equally probable to be obtained. This was followed by the calculation of the respective models, done each time a thousand times on the basis of the sub-sample of 600 independent values. Figure 2 shows these calculations as gray individual curves. The average, estimated parameters (location, shape, scale) served as a basis for the calculation of the reference values, and are represented in Figure 2 as black curves. The smaller number of subjects in the outer age ranges made it necessary to limit the estimates to the age segment 3–16 years (total: n=2522, boys: n=1312, girls: n=1210). In order to be able to evaluate the effect that infections had on the parameters ferritin and transferrin, the calculations for these parameters were carried out also with an adjusted dataset that did not contain any measurement time points with a simultaneously increased CRP (≥5 mg/L) (total: n=2446, boys: n=1275, girls: n=1171). Moreover, PASW Statistics in Windows version 18.0 (SPSS Inc., Chicago, IL, USA) was used for further descriptive analysis.

Figure 2:

Smoothed percentile curves of age (3–16 years) based on the reference population of LIFE Child cohort.

The individual curves are gray and the averaged curves are black for the percentiles 2.5 (P 2.5), 10 (P 10), 50 (P 50), 90 (P 90), and 97.5 (P 97.5). (A) Hemoglobin curves (mmol/L) for boys. (B) Hemoglobin curves (mmol/L) for girls. (C) Ferritin curves (ng/mL) for boys. (D) Ferritin curves (ng/mL) for girls. (E) Reticulocyte curves (per 1000 erythrocytes) for boys. (F) Reticulocyte curves (per 1000 erythrocytes) for girls. (G) Transferrin curves (g/L) for boys. (H) Transferrin curves (g/L) for girls.

Results

Using the adjusted statistical method (resampling) coupled with the GAMLSS package of the statistics R software, it was possible to create for the parameters hemoglobin, ferritin, transferrin, and reticulocytes reference intervals (percentiles 2.5 and 97.5) and smoothed percentile curves (percentiles 2.5, 10, 50, 90 and 97.5) continuously for age and separately for gender. Biannual base points of the respective parameters were calculated for the upper and lower reference limits (percentiles 2.5 and 97.5), the median (M), the coefficient of variation (S) and the skew (L) (summarized in the Table included in the Supplementry material). As a result of the statistical approach, and due to the 1000 calculations of the respective models, there were just as many individual curves that provided an overview of the potential fluctuations in the individual parameters. Furthermore, an averaging of these individual curves made it possible to realize a stable curve assessment. These curves are described in the following for each parameter.

The hemoglobin level rose, on average, from 7.5 mmol/L at age 3 to 8.3 mmol/L at age 11. A more pronounced gender-specific distribution became apparent from that age forward. This was characterized by a stagnation of the curves among the girls and a marked increase in the curves for the boys. The girls exhibited a constant median of 8.3 mmol/L, while the boys experienced an increase in the median to 9.3 mmol/L at the age of 16. By way of example, the reference intervals (P 2.5–P 97.5) ranged from 6.7 to 9.3 mmol/L for the boys and 6.8–8.9 mmol/L for the girls aged 3, from 7.3 to 9.3 mmol/L for the boys and 7.2–9.2 mmol/L for the girls aged 11, all the way to 7.9–10.4 mmol/L for the boys and 7.0–9.1 mmol/L for the girls aged 16 (Supplement, Table 1). A look at the individual hemoglobin curves revealed blurred upper limits (P 90, P 97.5) for children younger than 5 years, with more pronounced fluctuations among the boys than among the girls, as well as blurred lower limits (P 2.5) for the boys older than 15 years.

Ferritin exhibited, for the curves of percentiles 90 and 97.5, a reduction of the concentration (P 97.5) from 88.0 ng/mL for the boys and 87.2 ng/mL for the girls aged 3 years to 78.9 ng/mL for the boys and 75.8 ng/mL for the girls aged 6.5 years. Then, these curves showed a continuous increase up to the age of 16 years, which was more pronounced for boys upon reaching 128.4 ng/mL than it was for girls upon reaching 112.7 ng/mL. The curves for percentiles 2.5 and 10 remained largely constant across the ages, with the girls showing a slight downward trend from the age of 8 years with levels (P 2.5) from 16.6 ng/mL to 8.2 ng/mL at age 16. The 50^th percentile curves for the boys increased slightly across the ages, from 29.7 ng/mL at age 3 to 45.7 ng/mL at age 16. For the girls, however, they remained largely unchanged between 33.81 and 35.91 ng/mL (Supplement, Table 2). The individual ferritin curves showed potential fluctuations in the upper limits (P 90, P 97.5) for both boys and girls across the entire age range. In addition, a sub-group (especially among the girls) began to take form. What manifested itself was a single curve grouping that exhibited across all percentile curves, but especially from the 50^th percentile, significantly lower concentrations at age 3, increasing concentrations up to the age of 11, and a brief drop in concentrations followed by a virtually constant curve in contrast to the averaged curves. As ferritin is an acute-phase protein, all measurement time points where CRP was increased (≥5 mg/L) at the same time were also excluded from the calculation of percentiles. The estimated reference intervals differed from the values of the entire dataset with respect to the lower reference limit by a maximum of 0.7 ng/mL, and with respect to the upper reference limit, by a maximum of 6.5 ng/mL.

The relative reticulocyte content was constant across all the years of childhood and adolescence for the boys. The boys’ median was consistently between 9.1 and 9.2 per 1000 erythrocytes, with the reference interval (P 2.5–P 97.5) ranging from 4.7 to 5.0 to 15.7 and 16.0 per 1000 erythrocytes. The girls exhibited greater variability overall: the median was between 9.8 and 10.1 per 1000 erythrocytes and the reference interval (P 2.5–P 97.5) between 4.9 and 5.3–18.2 and 19.0 per 1000 erythrocytes (Supplement, Table 3). The upper limits (P 97.5) of the individual reticulocyte curves were blurred across the entire age range.

The transferrin concentrations, too, were overall constant across the ages. There was a slight increase in the curves of the percentiles 50, 90 and 97.5 for the girls. In total, however, the margin was minor. The median was between 2.7 and 3.0 g/L and the reference interval (P 2.5–P 97.5) between 2.2. and 2.3-3.2 and 4.0 g/L (Supplement, Table 4). The individual transferrin curves exhibited minor potential fluctuations in the upper (P 97.5) and lower limits (P 2.5) of the boys for the outer age ranges (under 5 years and over 13 years); for girls, this was the case only in the lower age ranges (under 5 years). As transferrin is an anti-acute-phase protein, all measurement time points where CRP was increased (≥5 mg/L) at the same time were also excluded from the calculation of percentiles. The estimated reference intervals did not differ from those of the entire dataset.

Discussion

The application of the LMS method according to Cole, which is integrated into the GAMLSS-R software package, may be considered an adequate method for estimating pediatric reference intervals. By applying the resampling method described, it is possible to include several measurement time points per subject and family, while also representing potential fluctuations as single curves, as well as stable curves by averaging the single curves. The curves can be represented in a continuous fashion across the ages, which is not only recommended by WHO [26], but also helps avoid the arbitrary division of age intervals and yields a better representation of physiological processes. The pediatric reference intervals for the parameters hemoglobin, ferritin, reticulocytes and transferrin were not previously worked out by means of the analyzers used in this context – Sysmex XN-9000 and Roche cobas 8000, c and e modules. This satisfies both the requirements of IFCC [25] to determine reference intervals in connection with new analytical methods and those of the manufacturers [34] regarding verification on the basis of one’s own reference population.

With the increase in concentrations up to the age of 11 years, as described, and the subsequent start of a gender-based progression characterized by relative constant concentrations in girls and a strong increase in boys, hemoglobin follows a typical trend confirmed by previous studies [10, 12, 17, 19, 35, 36]. This can generally be attributed to growth during childhood development. With the beginning of puberty, the differences can be explained on the basis of gender, particularly as a result of hormonal factors (erythropoietin stimulation due to androgens, or erythropoietin inhibition due to estrogens) and the onset of menstruation (blood loss) in girls. The studies of KiGGS [16] and Zierk et al. [17] lend themselves to comparison due to geographic proximity, timeliness and the description of continuous results. When comparing the reference intervals computed in this study with those of KiGGS (P 3–P 97), one sees mean deviations from the lower reference limits of 0.17 g/dL (maximum deviation: 0.38 g/dL) for boys and 0.07 g/dL (maximum deviation: 1.66 g/dL) for girls. As for the upper reference limits, one sees mean deviations of 0.66 g/dL (maximum deviation: 1.65 g/dL) for boys and 0.11 g/dL (maximum deviation: 1.76 g/dL) for girls. When comparing the reference intervals computed in this study with those of Zierk (P 2.5–P 97.5), one sees mean deviations from the lower reference limits of 0.34 g/dL (maximum deviation: 0.39 g/dL) for boys and 0.25 g/dL (maximum deviation: 0.7 g/dL) for girls. As for the upper reference limits, one sees mean deviations of 0.12 g/dL (maximum deviation: 1.15 g/dL) for boys and 0.19 g/dL (maximum deviation: 0.7 g/dL) for girls. Thus, the deviations from the studies mentioned are minor only. Any of the deviations are most likely attributable to the different analyzers used in the laboratory. It is virtually impossible for the generated measured values to harmonize perfectly due to the technical particulars of the different analyzers (different manufacturers, but also different models from a single manufacturer) [37]. As has been demonstrated, especially in the case of hemoglobin, the differences between various analyzers are minor compared to other hematological parameters [38]. Furthermore, differences can also be attributed to different criteria of inclusion and exclusion, as well as to the different statistical methods employed in the various studies.

Ferritin tends to be fairly constant in the lower percentiles (P 2.5, P 10), while the upper percentiles (P 90, P 97.5) are characterized by an increase that starts at the age of 6.5 years. The values increased more significantly in young males than in girls. Ferritin stores iron in the liver, bone marrow, spleen and other tissues, such as muscles. The increase in ferritin from the age of 6.5 years and the gender-specific differences are likely due to growth, organ maturation processes and hormonal changes, as well as the onset of menstruation in puberty. The calculation of percentiles to the exclusion of measurement time points with increased CRP yields virtually no deviations from the calculation that includes all measurement time points. This suggests that the statistical approach involving the estimate of the averaged percentile curves can compensate for pathological changes to some degree. A decision, therefore, was taken in favor of a larger data volume, that is, to show the entire dataset in Tables and Figures. When comparing the reference intervals computed in this study with those of KiGGS (P 3–P 97) [16], one sees mean deviations from the lower reference limits of 3.07 ng/mL (maximum deviation: 5.39 ng/mL) for boys and 4.07 ng/mL (maximum deviation: 6.22 ng/mL) for girls. As for the upper reference limits, one sees mean deviations of 25.25 ng/mL (maximum deviation: 42.27 ng/mL) for boys and 18.81 ng/mL (maximum deviation: 31.33 ng/mL) for girls. Comparability with other studies is limited because of the difference in the approach of continuous interval determination and the general separation by gender in this study, while other studies employed reference intervals mostly for age groups that were not consistently gender-specific. By contrast, however, the lower reference limits of the CALIPER study [14] were lower than those in the present study or those of the KiGGS study. One possible explanation could be the recruitment of hospitalized patients, for whom metabolic disorders had been ruled out, but who did include orthopedic patients, for example. In other words, clinical components might play a role here. In a US study done in 2004 [39], ferritin levels exhibited a greater variability that was associated with higher upper reference limits than those of the present study. Our study shows similar variability in the percentiles 2.5–97.5 only for adolescents. This, too, might be attributed to the fact that the aforementioned study used hospitalized patients as reference individuals. It is highly likely that the deviations mentioned can be explained by the difference in the recruitment of study populations. There are also indications that the use of different analyzers is instrumental in the discrepancy of findings also where ferritin is concerned [40].

If one uses the cut-off values for hemoglobin and ferritin issued by WHO [41] for diagnosing iron deficiency anemia as a basis for comparison, they harmonize with the results observed by us only partially, as they do with the results of the KiGGS [16] and Zierk et al. [17] studies. The WHO hemoglobin cut-off values are gender-specific only from the age of 15 years. But, as the current study shows, gender-specific differences manifest themselves already from the age of 11 years. Girls aged 12 and older generally remain below the WHO cut-off values according to the lower reference limits calculated in this study. When it comes to ferritin, this shortfall of the WHO cut-off values, when compared to this study and the KiGGS study, is even more significant and manifests itself already in girls aged 11. The WHO cut-off values for iron deficiency anemia diagnostics should, therefore, be updated.

The reticulocyte count remains largely constant across the entire age range, at 5–19 per 1000 erythrocytes, which is thus similar to a previously described margin of 5–15 per 1000 erythrocytes [35]. Previous data on reticulocyte counts have not been uniform. They are comparable only to a limited degree due to different representations as absolute and relative values. The literature has also described fluctuations due to different analyzers on multiple occasions [38, 42, 43]. Sysmex GmbH has described decreasing trends in adulthood [37], which partially matches the findings of this study. But, in contrast to the Sysmex references, this study has found lower lower reference limits and mostly higher upper reference limits, especially in girls. One Swedish study involving children aged between 8 and 18 years has demonstrated gender-specific differences, with higher concentrations in girls [19]. In the present study, this has been proved particularly with respect to the upper percentiles (P 50, P 90, P 97.5). The reticulocytes constitute the last immature precursor of the erythrocytes and will need to be re-created permanently to cover the breakdown of the erythrocytes. This is why a consistently equal share of reticulocytes is required under physiological conditions.

Transferrin is characterized by an overall continuous progression marked by an increase in variability among girls aged 12 and older. This has been confirmed by a large number of studies [13–15, 18, 19], while the subject has been controversial in a few others [39, 44]. However, it bears mentioning that the changes occur in such a small margin that one cannot assume any relevant gender-specific difference. In other words, the reference intervals of different studies harmonize well. Transferrin acts as a transport protein that allows for the exchange of iron between the intestines, iron stores and erythroblasts. It responds to iron deficiency by way of an increase, but does not appear to be influenced directly by physiological development processes.

The potential fluctuations described for the single curves in the outer percentiles and outer age ranges reflect a reduced stability due to a smaller number of incoming measured values (only the outer 2.5%) as well as a lack of data points outside the outer age ranges that could act as reference points. But these two factors can be compensated for well by way the mean estimate of the single curves. The grouping of ferritin single curves, which has also been described, suggests that there is a sub-group with a common characteristic. The most likely explanation in this context is that this involves a genetic component, such as the representation of a specific ferritin type (isoferritin) [45, 46].

The analyses can yield current reference intervals for hemoglobin, ferritin, reticulocytes and transferrin in children and adolescents, based on a primarily healthy, highly standardized sample population. This has been done for the first time for the devices used, Sysmex XN-9000 and Roche cobas 8000 (c and e modules). The statistical GAMLSS R software package has been confirmed as an effective means to calculate percentile curves and reference intervals. The progressions of the parameters have been represented continuously across the ages, for the first time for reticulocytes and transferrin, rather than on the basis of age groups. This approach maps physiological conditions (smooth transitions instead of stepwise approach) and prevents the problem of an arbitrary classification of age intervals. Furthermore, the customized statistical method has presented a way in which a longitudinal design can be applied to a cross-sectional study. In addition, it is possible to include family members, without violating the independence necessary for the calculations. The repeated model computations and averaging of the resulting single curves increase reliability and move the influence of confounders, CRP increases in this case, to the background. The individual curves further allow for potential fluctuations or special characteristics, such as the ferritin sub-group described, to be evaluated. By working out agreements and deviations with respect to other studies, factors, such as analyzers, criteria of inclusion and exclusion, type of age (continuous or grouped) and gender analysis (separated by gender or collectively), clinical aspects and statistical methodology, as well as the IFCC recommendation on the regular revision of reference intervals can be substantiated. Our study draws attention particularly to a possible revision of the WHO reference intervals for hemoglobin and ferritin.