Establishment of reference intervals of clinical chemistry analytes for the adult population in Egypt

Background This is the first Egyptian nationwide study for derivation of reference intervals (RIs) for 34 major chemistry analytes. It was conducted as a part of the global initiative by the IFCC Committee on Reference Intervals and Decision Limits (C-RIDL) for establishing country-specific RIs based on a harmonized protocol. Methods 691 apparently healthy volunteers aged ≥18 years were recruited from multiple regions in Egypt. Serum specimens were analyzed in two centers. The harmonization and standardization of test results were achieved by measuring value-assigned serum panel provided by C-RIDL. The RIs were calculated by parametric method. Sources of variation of reference values (RVs) were evaluated by multiple regression analysis. The need for partitioning by sex, age, and region was judged primarily by standard deviation ratio (SDR). Results Gender-specific RIs were required for six analytes including total bilirubin (TBil), aspartate and alanine aminotransferase (AST, ALT). Seven analytes required age-partitioning including glucose and low-density lipoprotein cholesterol (LDL-C). Regional differences were observed between northern and southern Egypt for direct bilirubin, glucose, and high-density-lipoprotein cholesterol (HDL-C) with all their RVs lower in southern Egypt. Compared with other collaborating countries, the features of Egyptian RVs were lower HDL-C and TBil and higher TG and C-reactive protein. In addition, BMI showed weak association with most of nutritional markers. These features were shared with two other Middle Eastern countries: Saudi Arabia and Turkey. Conclusion The standardized RIs established by this study can be used as common Egyptian RI, except for a few analytes that showed regional differences. Despite high prevalence of obesity among Egyptians, their RVs of nutritional markers are less sensitive to increased BMI, compared to other collaborating countries.


Introduction
Egypt, being at the crossroads of Africa and Asia overlooking the Mediterranean Sea, had created a demographic melting pot of ethnicities over a long period of history. Thus, the study of the reference intervals (RIs) of common chemistry analytes and their sources of variations (SVs) in Egyptians in comparison to other nations is intriguing.
This study comes as the first Egyptian initiative to report RIs for 34 major laboratory analytes: 27 chemistry analytes, 6 immunoturbidimetric analytes, and parathyroid hormone. In 2013, the Committee on Reference Intervals and Decision Limits (C-RIDL), International Federation of Clinical Chemistry (IFCC) issued a harmonized protocol for establishing regional RIs by means of multicenter study (1). For standardization and harmonization of tests results, it provided a value-assigned serum panel to collaborating countries on request (2). Besides determining the RIs for common analytes, this study aimed at investigating the sources of variation that may impact the RIs, including age, gender, body mass index (BMI) and region.
Egypt is the third Middle Eastern country following Saudi Arabia and Turkey that adopted the C-RIDL protocol to derive country-specific RIs for nationwide use. The results of this study provide an opportunity to compare the RIs among Middle Eastern countries, and across other collaborating countries of widely different demographic profiles such as body mass index (BMI), alcohol intake and smoking.

Study design
This study was conducted by a collaboration of Cairo University Faculty of Medicine and Medical Research Institute of Alexandria University. The study was approved by the ethical committee of Cairo University (N- . Each participant signed an informed consent form before enrollment.

Recruitment of reference individuals
A total of 691 apparently healthy Egyptian volunteers were recruited from Cairo, Alexandria, Delta, Suez Canal, and Upper Egypt; Fayoum, Beni-Sueif or areas southern as far as Nubia The subjects aged more than 18 years and were stratified into 5 age groups: 18-29, 30-39, 40-49, 50-65 and ≥65 years. Subjects were selected so that at least 80% were from the age of 18 to 65 years, with equal gender mix and age distributions, except for individuals over 65 years of age.

Blood sampling
The volunteers were instructed to avoid excessive exercise and eating a few days before the sampling. The time of sampling was set from 8-11 AM, after resting in a seated position for approximately 20 minutes (3). In the resting period, each participant answered to the detailed health-status questionnaire that was adopted from the harmonized protocol. Ten mL of blood were drawn into a plain evacuated tube. Clotted samples were centrifuged, and the sera were divided and stored at -80°C.
Measurements were performed in a batch of 60~150 specimens/per day, after thawing at room temperature for at least 1 hour. Chemistry samples collected by Cairo University were analyzed on Beckman AU 680 (Beckman Coulter International, Nyon, Switzerland), while chemistry samples collected by Alexandria University as well as all samples for immunoturbidimetry and PTH from the two centers were measured on Cobas 6000 modular system (Cobas modular Roche, Indianapolis IN, USA), Roche, at Hassab Labs in Alexandria.
Both labs are ISO 15189 accredited. Samples were assayed within 2−3 months of storage at −80°C except for the following analytes: IgG, IgM, IgA, C3, C4, CRP and PTH, which were measured 8 months after the collection.

Between-center method comparison and standardization
A panel composed of 50 healthy volunteers' sera with assigned values for 34 chemistry and immunoturbidimetry analytes (2,5) was measured by the two centers for merging and standardization of test results as described elsewhere For between-day quality control, "minipanel" was prepared comprising sera of 4 volunteers measure at each run of assay in both centers(2).

Sources of variation of reference values
Multiple regression analysis(MRA) was performed, separately for males and females, to examine sources of variation. Reference values (RVs) of each analyte was set as an objective variable, while set as explanatory variables were volunteers' region, age, body mass index (BMI), levels of cigarette smoking, exercise, consumption of soft drinks. For the region, we categorized the volunteers by their origin either as northern or southern (Upper) Egypt, where northern Egypt included inhabitants of Alexandria, Delta, Cairo, and Suez Canal, while southern Egypt included inhabitants from Fayoum, Beni-Sueif or areas southern as far as Nubia.
Northern Egypt was set as the reference category: [northern]=0, [southern]=1. The degree of association of each explanatory variable with the objective variable was expressed as a standardized partial regression coefficient (rp), which generally takes a value between −1.0 and 1.0. We regarded |rp|≥0.2 as an appreciable effect size of the association between small correlation (0.1) and medium correlation (0.3), specified by Cohen (6).

Partitioning criteria
To judge the need for partitioning RVs by sex, age-subgroup (<45 versus ≥45 years), and region (northern or southern Egypt), we primarily used standard deviation ratio (SDR) based on analysis of variance (ANOVA). SDR for a particular SV(SDRsv) was calculated as standard deviation (SD) due to a particular source of variation (SV) (SDsv) divided by SD due to "coarse" between-individual SD(SDbi) or SD comprising the RI(SD RI ). We regarded SDRs ≥ 0.3 as an appreciable level of between-subgroup variations, and SDRs≥0.4 as requiring partitioning of RVs by the factor concerned (7).
Because SDR represents the magnitude of between-subgroup differences at the central part of RV distribution, it may not reflect between-subgroup differences at lower or upper limits (LL or UL) of the RI. Therefore, as a secondary index of between-subgroup difference, "bias ratio" (BR) was computed at LL and UL. For example, the equation shown below for calculating BR of between-sex difference at UL (BR UL ), where subscript M, F, and MF represent male, female, and male + female, respectively, the same for the calculation of bias ratio for LL (BR LL ).
According to the conventional specification of allowable bias at the minimum level: 0.375× (8) where SD G and SD I represent "pure" between-and within-individual 2 + 2 = ( ) variations, we regard BR UL (or BR LL ) > 0.375 as a secondary criterion for partitioning RVs by sex, age, and region.
In both MRA and ANOVA, the RVs of analytes that showed highly skewed distributions were transformed logarithmically before the analyses to avoid excessive influence of deviated data in the periphery of RV distributions. Such analytes included TG, AST, ALT, LDH, GGT, CK, and CRP were performed using a statistical software, StatFlex Ver.7 (Artech, Osaka, Japan).

Derivation of Reference Intervals
The parametric method was used for computing reference intervals after transforming the distribution of RVs into Gaussian form using the modified Box-Cox transformation (7) to obtain mean and SD (final RI was calculated as the mean±1.96SD) which corresponds to the central 95% limits or LL and UL under transformed scale, then they are reverse-transformed to get the LL and UL on the original scale.
For improved precision of the RIs, the bootstrap method through 50-times resampling of the dataset was applied to obtain smoothed lower and upper limits (LL, UL), and mean of the reference intervals. This resampling procedure was also used for predicting 90% confidence intervals (CI) for the limits of the reference interval.
The LAVE method was applied for the secondary exclusion of individuals with conditions affecting nutritional, muscular, and inflammatory markers. In our study, the LAVE method was applied in two groups using a different set of reference tests. Group 1 consisted of nutritional,  reported regular exercise from one to six days a week. Participants with smoking habit were 27.5% in males, and 1.1% in females and those who did not deny drinking alcohol were 1.25%

Demographic profile of participants
in males and 0.6% in females.

Merging the datasets from the two participating laboratories
For the analytes with assigned values by the serum panel, (i.e. the standardized analytes), the common measurements of the serum panel in the two centers were used to compare the two participating laboratories (Suppl. Fig. 1 shows the comparison between the panel results from the two laboratories ). Among these analytes, obvious dissociation of panel results was observed between the two centers for Alb, Cre, Na, Cl, and Mg, and slight dissociations for HDL-C, AST, ALP, and GGT. This warranted recalibration of the results of each center independently to the assigned values of the C-RIDL panel, by use of major-axis linear regression lines. Then the results of the two centers were merged into one data set.
For the non-standardized analytes: TP, TBil, DBil, TIBC, UIBC; the test results from Alexandria University were aligned to those of Cairo University by use of the major axis regression lines constructed from the panel test results of the two centers.

Sources of variation and partitioning of reference values
Results of MRA are shown in  In females, age was positively associated with non-HDL, TG, LDL-C, TC, ALP, Urea, Cre, and GGT in this order of strength and negatively associated with eGFR. The association of BMI with nutritional markers was weak with |rp|<0.20 except for CRP, TC, and Na. As regards the regionality, RVs of southern Egypt were found lower (rp≤−0.2) for DBil, HDL-C, and Glu.
Further analyses of the SVs were performed using nested ANOVAs (  Between-sex differences with significant SDRsex (shown in the parenthesis) were observed and CRP (0.00).
Significant between-age differences for nonHDL (0.50) among females was noted affirming the already noted significant rp values derived by MRA, as well as an expected significant SDRage for eGFR for both sexes. Graphical representation of sex and age-related changes for selected parameters are shown in Fig. 1.

Fig 1: Sex and age-related changes of selected parameters
RVs were partitioned by sex (male: M, female: F) and age-subgroups (~29, 30~39, 40~49, 50~). The box in the center of each scattergram indicates the mid 50% range of RVs, and its central vertical bar represents the median. The data size is shown at the right bottom of the age group labels. Because no secondary exclusion was done for RVs, the range of the scatter plot may not match to the RI to be determined.

Derivation of Reference Intervals
RIs were derived by the parametric method with or without the LAVE. The calculations were done in two steps, first partitioning of RVs by sex, followed by further partitioning by age at 45 years (<45 versus ≥45 years). The RIs by the former calculation is listed for all analytes in Suppl. Table 2 and those by the latter calculation are listed in Suppl. Based on SDRage, RIs partitioned by age were found necessary only for eGFR (males and females) and nonHDL (female). While, based on significant BR LL or BR UL for between-age subgroup differences, age-partitioned RIs were also adopted in females for UA, TG, ALT, GGT, and CK, and in both sexes (without gender distinction) for Glu, TC, and LDL-C.
As for the need for applying the LAVE method, the decision was made based on BR LL or BR UL for the difference of RIs with/without LAVE. It turned out that the LAVE method was found effective in lowering the UL of the RIs in male for UA, TIBC, nonHDL, AST, TC, GGT, IgG, and CRP, and in female for TC, TG, AST, ALT, GGT, and CK.

10
The rationale behind choosing each RI are shown in Suppl. Tables 2 and 3. The list of RIs thus chosen are shown in Table 3.    Fig. 2 shows between-ethnic group comparison of RVs for four analytes that we identified as peculiar to Egyptian population from what reported in the C-RIDL's global study (8,11).

Discussion
Mean BMI in our study was 28.2±5.0. It is higher than those reported by most collaborating countries adopting the IFCC C-RIDL protocol namely (9) The interim report by C-RIDL on the global multicenter study (5,9) highlighted close association of BMI in terms of rp with the following analytes: C3, ALT, CRP, UA, GGT, HDL-C, TG, LDL-C, C4, and AST in that order of strength in males, and similarly in females. In this study, we observed a similar association of BMI with those nutritional/inflammatory markers, but at much weaker degree: i.e., values of |rp| for ALT, CRP, GGT, and TG were all below 0.2.
For TC, UL of Egyptians was 6.28 mmol/L for both sexes <45 years, comparable to other studies: Turkey 6.2(11), India 6.2(10), Saudi Arabia 6.36 (12) and China 6.16 mmol/L(13), while among individuals >45 years, the Egyptian UL (6.99 mmol/L) was the highest reported followed by Indian UL of 6.7(M) and 6.6(F) mmol/L. Applying the LAVE procedure to TC was effective in lowering the UL for <45 years of age, but not for UL of >45 years. This failure of LAVE was previously encountered in the Indian study (10) and was attributed then to a weaker association among reference test items used for the LAVE procedure. With regard to HDL-C, Fig. 2  response seen among inhabitants of equatorial and subequatorial terrains (14) or an underlying inflammation related to metabolic syndrome (15).
To further explore how consistent the BMI related changes were across different countries that collaborated in the C-RIDL global study, Fig. 3 was drawn to systematically evaluate the association of BMI and RVs of six nutritional markers namely LDL-C, HDL-C, ALT, GGT, CRP, and UA.

Fig 3: Ethnic differences in regression line between BMI and RVs of nutritional markers
Comparison of least-square linear regression lines between BMI and RVs of six nutritional markers among ten countries: Middle Easterners: "Egypt, Saudi Arabia, Turkey" in graded red, Asians: "Japan, China, India" in graded blue, and Caucasians "USA, Russia, Argentina, South African Caucasians (ZA-Cau)" in graded green.
Notably, regression lines showed different slopes, very steep in some countries (implying slight change induces profound effect on the RVs) while moderate to gentle in others. Taking HDL-C as an example, it was previously (16) reported that the HDL-C concentration did not reduce linearly as BMI advanced >30 kg/m 2 . This could explain the wide differences in the regression line slopes among the different populations listed in Fig. 3, and the tendency of the  (19,20), while reduced in others (21). In our study, the IgA was positively correlated with smoking. Whereas, smoking habit-related lowering of IgG or HDL-C has been reported (9,22). Such a finding was not observed in this study despite the appreciable number of male smokers.

Conclusion
Standardized RIs for 34 chemistry analytes among Egyptians were established considering various sources of variations. They can be use in common in Egypt except a few analytes that showed moderate differences between northern and southern Egypt. Despite the high prevalence of obesity in Egyptians, the RVs of nutritional markers were found to be less sensitive to increased BMI, especially in females, when compared to other collaborating countries. From an international perspective, this study revealed that RVs of Egyptian feature low HDL-C and TBil, and high TG and CRP. RVs were partitioned by sex (male:M, female: F) and age-subgroups (~29, 30~39, 40~49, 50~). The box in the center of each scattergram indicates the mid 50% range of RVs, and its central vertical bar represents the median. The data size is shown at the right bottom of the age group labels. Because no secondary exclusion was done for RVs, the range of the scatter plot may not match to the RI to be determined.

Fig. 2: Side-by-side comparison of RVs for TBil, TG, HDL-C, and CRP between Egyptian and other population
RVs of ten countries were compared by box-whisker charts. The box represents central 50% ranges, the vertical bar at the box represents the median. The span of the horizontal bar represents central 95% interval. The ethnic groups were subgroups by ethnicity: Middle Easterners: "Saudi Arabia, Turkey" with the central box shaded in red, Asians: "Japan, China, India" shaded in blue, and Caucasians: "USA, Russia, Argentina, South African Caucasians (ZA-Cau)" shades in green. The full-range vertical line represents median RV of Egyptian male (blue) and female (red).

Fig 3: Ethnic differences in regression line between BMI and RVs of nutritional markers
Comparison of least-square linear regression lines between BMI and RVs of six nutritional markers among ten countries: Middle Easterners: "Egypt, Saudi Arabia, Turkey" in graded red, Asians: "Japan, China, India" in graded blue, and Caucasians "USA, Russia, Argentina, South African Caucasians (ZA-Cau)" in graded green.
Suppl. Fig. 1: Comparison of panel test results from two testing centers and assigned values Value-assigned panel of sera were tested in Cairo and Alexandria University and their test results were compared with the assigned values (11]. The linear regression was computed by major-axis regression line. Merging of volunteers' test results were done by aligning them to the assigned values.

Suppl. Fig. 2: Sex and age-related changes in RVs of all parameters
RVs were partitioned by sex (male:M, female: F) and age-subgroups (~29, 30~39, 40~49, 50~). The box in the center of each scattergram indicates the mid 50% range of RVs, and its central vertical bar represents the median. The data size is shown at the right bottom of the age group labels. Because no secondary exclusion was done for RVs, the range of the scatter plot may not match to the RI to be determined.

Suppl. Fig 3: Between-region differences in RVs observed in five analytes
RVs of five analytes, TBil, DBil, Glu, TG, and LDL-C that showed high SDR for betweenregion differences (SDRreg) were partitioned into four groups by sex and region (North vs. South Egypt). Values of BMI were also shown subgrouped by sex and region to prove that the regionality is independent of the levels of BMI. . |rp|  0.20 was considered significant and shown in bold letter. Negative |rp| was coded in red and positive one in green. Table 2: Standard deviation ratios (SDR) representing the magnitude of between-sex, between-region, and between-age variations of RVs.