Reference intervals established using indirect method for serum ferritin assayed on Abbott Architect i2000SR analyzer in Chinese adults

Abstract Background Serum ferritin (SF) test has been widely used in clinical practice. However, its reference intervals (RIs) vary depending on the analytical method and ethnic origin. This study was to establish the RIs using indirect method for SF in Chinese adults. Methods SF was assayed on Abbott i2000SR analyzer. The SF test results of all health examinees (8913 males aged 18‐93 years and 5397 females aged 18‐90 years) between December 2010 and April 2019 were obtained from our laboratory information system. After Box‐Cox transformation of raw data and exclusion of outliers, parametric and non‐parametric approaches were used to calculate 95% RIs. The correlation between SF levels and ages, and the differences in SF levels between subgroups were also analyzed. Results SF levels in females were significantly different from those in males (Z = 88.96, Z* = 23.17; Z > Z*) and showed a weak positive correlation with age (r = .466, P < .0001). The RIs based on parametric approach in males were 66.12‐561.58 µg/L, whereas in all females were 3.59‐269.59 µg/L, females aged <50 years 3.26‐148.02 µg/L and those aged ≥50 years 17.28‐303.27 µg/L. The RIs based on non‐parametric approach in males were 65.00‐571.37 µg/L whereas in all females were 4.00‐254.00 µg/L, females aged <50 years 4.00‐152.00 µg/L and those aged ≥50 years 16.00‐304.05 µg/L. Conclusions Our indirect RIs for SF were markedly different from the manufacturer's recommended RIs and might be more suitable for Chinese adults, which would be helpful in interpreting laboratory data and clinical decision‐making.

the "direct" method, a major problem that the laboratory faces is to obtain a sufficient number of specimens from healthy individuals representative of the reference population that the laboratory serves. 24 Thus, as an alternative method, the "indirect" method via large collections of laboratory data stored in laboratory information systems (LIS) and statistical program is widely used to establish the RIs in recent years. [24][25][26][27][28][29][30] Relative to the direct method, the "indirect" method is much simpler, faster, and cheaper 25 and therefore is a convenient way for the clinical laboratory to set up their own RIs. In previous studies, however, most of which focused on the data obtained from hospital patient records and a group of "healthy" subjects was selected as the reference population from these hospital patients based on some statistical criteria, 24,[27][28][29][30] very few of which determined the RIs for analytes using the data from hospital database of health examination records. Although these data may not represent the general population, subjects were those who visited the health examination center for regular check-ups, but not for treatment of a confirmed disease, which may lower the selection bias compared with previous hospital-based studies. Therefore, the objective of the present study was to establish the RIs for SF, which is assayed by chemiluminescent microparticle immunoassay (CMIA) on the Abbott Architect i2000 SR analyzer, by an indirect method using the data from our LIS records of health examinees.

| Data source
All raw data were obtained from the database of the LIS of Shaoxing

| SF measurement and quality control
Venous blood samples were obtained from each study participant after a 12-hours fasting period and collected into BD vacutainer tubes (Becton Dickinson). Serum was separated from the blood by centrifugation at 3000 rpm for 5 min at room temperature (18-25°C).
Specimens with hemolysis, icterus, and lipemia were excluded according to the rules of the laboratory. All specimens were quantitatively determined within 2 hours of collection by CMIA on Abbott Architect i2000 SR analyzer (Abbot Diagnostics) according to the manufacturer's specifications and following the laboratory standard operating procedures. The detection kits and calibrators were provided by Abbott Diagnostics. To ensure the accuracy and reliability of the analysis, assays were traceable to the 3rd international recombinant standard for ferritin (National Institute for Biological Standards and Control Code 94/572) and subject to participate in internal quality control (IQC) at different levels every day and interlaboratory external proficiency testing (PT) schemes from local external quality assessment (EQA) provider (Zhejiang, China) every year. Throughout this study period, the total analytical coefficient of variation (CV) was <10% and the PT/EQA score was more than 80% for SF, which met the quality assurance requirements as described in the Clinical and Laboratory Standards Institute's (CLSI) EP15-A3 guidance document. 31

| Evaluation of test results' distribution and transformation of data
Skewness-Kurtosis test was used to evaluate whether the given test results are distributed symmetrically or normally or whether the distribution is asymmetrical or skewed. 32 When the data were significantly skewed, Box-Cox would be used to transform the data to approximately normal or symmetrical distribution. The Box-Cox transformation is defined as: where λ is an undetermined parameter, which is obtained by the maximum likelihood method.

| Deletion of outliers
In this study, Tukey methods were used to check for multiple outliers at either side. In this method, an outlier value is defined as a value that is smaller than the lower quartile minus 1.5 times the interquartile range (IQR, the range from the 25th to the 75th percentile), or larger than the upper quartile plus 1.5 times the IQR. The calculation formulas are below: With Tukey's criterion, an outlier will be eliminated and this process is repeated on the remaining data until all outliers are eliminated. When the data are significantly skewed, the outliers are removed after Box-Cox transformation. Following the outlier removal, the remaining data are re-transformed and the outliers are removed again (iteration) until no outlier after the final data transformation.
Then, the data were back-transformed to the original values for presentation.

| Determination of RIs
Parametric or non-parametric percentile statistical approach as described in the CLSI guidelines EP-A3 guidelines can be used to calculate RIs in indirect method. 21,24,25,33 For parametric approach, the 95% RIs (double-sided) was defined mean ± 1.96 SD; for nonparametric approach, the values of the 2.5th and 97.5th percentile were used as lower and upper limits of RIs following the guideline EP28-A3, 23 respectively. Meanwhile, the 90% confidence intervals (CI) for lower and upper limits were also calculated. The RIs were also established for the various age-and sex-specific subgroups when there are significant differences between the subgroups. Rank correlation was used to analyze the correlation between the SF levels and age. The standard normal deviation test (Z test) was performed to reveal the significance of differences in SF levels between the subgroups, 23,34 and the calculated Z > critical Z (Z*) was considered to be different from each other between subgroups.

| RE SULTS
A total of 15 949 raw data for males and 8141 for females were obtained from our LIS database, respectively. Of which, 7036 data for males (44.12%) and 2744 for females (33.70%) were excluded either because of the SF levels exceeding the detection limit of the methodology or because of missing information or repeated results. Finally, 8913 males aged 18-93 years and 5397 females aged 18-90 years were included in this study for further analysis.
The statistical significance of the difference between SF in males and in females was tested by the Z test, and significant difference was found between males and females (Z = 88.96, Z* = 23.17; Z > Z*).
Thus, 95% RIs and their 90% CIs were established separately for males and females. Rank correlation analysis showed SF levels had a weak positive correlation with age in females (Spearman's coefficient of rank correlation (ρ) = 0.466, P < .0001) but no significant correlation between SF and age was found in males (Spearman's coefficient of rank correlation (ρ) = 0.0193, P = .0682) as shown in Figure 1. Z test revealed there was a significant difference in SF levels between females aged <50 years and those aged ≥50 years (Z = 34.71, Z* = 14.22; Z > Z*). So, we established the age-specific RIs and their CIs for females finally.
In this study, the SF in either males or females or their subgroups showed a significant skew distribution when Skewness-Kurtosis test was used. After Box-Cox was used to transform the data, all of x( ) = ln(x) when = 0.
Lower limit = P25 − 1.5 × IQR Upper limit = P75 + 1.5 × IQR them were approximately normal or symmetrical. The SF level distributions of sex-and age-specific subgroups are detailed in Table 1, and the representative histograms are shown in Figures 2 and 3.
After at least 2 iterations, the outliers in each group or subgroups were removed completely. The number of iterations and outlets in each group or subgroups is presented in Table 2. Then, we established 95% RIs for males and females and their age-specific subgroups based on parametric and non-parametric approach (Table 3).
Compared with the manufacturer's recommended RIs, the RIs established for all females showed a slight difference; however, the RIs established for the other subgroups, whether based on parametric approach or based on non-parametric approach, were markedly different (Table 3).

| D ISCUSS I ON
In this study, we established the RIs for SF in Chinese adults using the data obtained from our LIS and found that the RIs for SF in F I G U R E 1 Correlation between age and serum ferritin levels. A, Males; B, Females   RIs are calculated from hospital data using statistical approaches.
In the present study, we attempted to establish the RIs for SF in Chinese adults using the data available from our LIS records.
Unlike the data in most of the previous studies, the data in the present study were obtained from the subjects during their visit for the health examination center. The very low probability of disease makes these subjects obviously more desirable as the reference population than hospital patients since exclusion of extreme results, which are those most likely to be affected by disease, is a vital issue in establishing RIs. 24,[27][28][29][30] The present study demonstrated the SF levels had a significant difference between females and males, and had a positive correlation with age in females. Nevertheless, the manufacturer simply recommends 21.18-274.66 µg/L as the RIs for adult males and 4.63-204 µg/L for adult females but does not recommend the age-specific RIs. To gain insight into the effect of age on SF levels, we further tested whether there was a significant difference between various age-specific subgroups by Z test and found that the SF levels in females aged ≥50 years were significantly higher than those in those aged <50 years. These findings are in good agreement with those by Ogilvie et al 36   this, it is necessary to establish the sex-and age-specific RIs for SF in our own laboratory.

Number of outliers
In the indirect method, RIs can be calculated using various statistical approaches. 25 Of these statistical approaches, the standard parametric or non-parametric percentile approach is usually used in the direct method for RIs as described in the CLSI EP-A3 guidelines. 23 Jones et al 25 suggested that the standard parametric or nonparametric statistical approach can be used in the indirect method for RIs when the data were collected at a community screening project or similar. In fact, the parametric approach has been used in the analysis of data from NHANES, 21 and the non-parametric approach has been used in a Chinese and a Turkish study, respectively. 24,33 Of course, when the standard parametric approach is used, this will involve data transformation if the data are non-normal or asymmetric distribution. In this study, our data showed a significant skew distribution but were symmetrical or approximately normal distri-  38 As mentioned above, these findings might be related to considerable variation in SF levels in ethnic origin. 3,22,35 In fact, in multi-ethnic population studies in the USA, Adams et al 39 had proved that elevated SF levels are more frequent in Afro-Caribbean and Asian subjects than in whites or Hispanics. In a previous Chinese study on Cobas® 6000 system E601 (Roche) analyzer, Li et al 40 also found that Chinese adult males had higher RIs for SF.
However, compared with that study, 40 our study had slightly but significantly lower RIs for SF in Chinese adult males. This may be related to the different assay methods used, as reported in previous studies. [19][20][21] Taken together, we believe that the sex-and age-specific RIs for SF established in this study are appropriate.
In conclusion, the present study has provided evidence that the RIs for SF in Chinese adults are markedly different from the manufacturer's recommended Rls. In Chinese adult females, the age-specific RIs must be also taken into account. As a result, the establishment of the sex-and age-specific RIs by indirect method using the data from health examination records for SF in our own laboratory would give a better chance to aid the physician in differentiating or considering treatment of iron-deficiency anemia or a risk of iron overload than using manufacturer's recommended RIs or adopting RIs from other laboratories.

ACK N OWLED G M ENT
We thank Li-ming Mao and Hui Xie for critical reading of the study and numerous valuable comments.