Sources of variation and establishment of Russian reference intervals for major hormones and tumor markers

Each clinical laboratory is expected to establish its own reference intervals (RIs) as recommended in the IFCC/CLSI guideline (C28-A3) [ 1 ], but most laboratories in Russia use RIs provided by the reagent manufacturers. They may not match to the Russian population due to a variety of population-specific factors. Therefore, we joined the global multicenter study on reference value (RVs) coordinated by the IFCC Committee on Reference Intervals and Decision Limits (C-RIDL) in 2013. We recruited 793 healthy volunteers from three major cities: according to the C-RIDL protocol [ 2 ] and analyzed biological features of Russian RVs for 34 commonly tested chemistry analytes [ 3 ]. As a result, it was revealed that the derived RIs for most chemistry analytes differed greatly from those shown in the reagent inserts, which underlined the importance of determining country-specific RIs for all laboratory analytes. However, unlike reports on RIs for chemistry analytes, there are not many reports of well-designed studies conducted for establishing RIs for immunoassay analytes. The only comprehensive report available is the IFCC Asian study conducted in 2008~9 involving 3,500 healthy volunteers [ 4 ]. The study revealed clear between-country differences for parathyroid hormone (PTH), adiponectin, folate, and vitamin B12 (VB12), but none for other analytes including most of commonly tested tumor markers and reproductive hormones. Another one is a Saudi Arabia study, conducted as a part of IFCC global multicenter study, where 826 apparently healthy individuals were recruited and RIs for 20 immunoassay analytes including five tumor markers, 12 hormones and three vitamins were derived [ 5 ]. There are other RI studies, targeting a smaller number of analytes, such as tumor markers [ 6 , 7 ] and thyroid hormones [ 8 , 9 ]. Besides, no comprehensive analyses of biological sources of variations have been performed so far except for a recent report from a Chinese group collaborating in the C-RIDL global study, which established RIs for eight male sex hormone-related analytes and seven thyroid hormones and analyzedtheir SVs [ 10 , 11 ]. In this second part of our RI study, we targeted 22 major immunoassay analytes. They include five tumor markers, eight reproductive hormones and related tests, five thyroid function tests, and four other hormones. By use of the same statistical methods as for the first part, we tried to establish the RIs specific to the Russian population in careful consideration of SVs of each analyte.

SVs of each analyte were evaluated by MRA and ANOVA as described in the part 1 of this report [ 3 ] and respective results are listed in Tables 1 and 2 . In the following sections, the findings of 22 laboratory parameters were divided into four groups according to their categories: tumor markers (AFP, CEA, CA19-9, CA125, CA15-3), the reproductive panel (PRL, LH, FSH, TβhCG, estradiol, progesterone, testosterone and SHBG), thyroid function tests (TSH, FT4, FT3, TT4, TT3), and miscellaneous ones (insulin, cortisol, GH and PTH). No apparent between-city differences (SDR city ) were observed for any analyte with the highest SDR city of 0.21 observed for TSH (data omitted). Therefore, all the data from the three cities were merged in the subsequent analyses. Sex-related changes with SDR sex ≥0.4 were not observed in any tumor marker as shown in Table 2 . However, by close look at S1 Fig for sex- and age-related changes, RVs of CA125 in females are appreciably higher than males until 50 years of age, but lower thereafter with SDR age of 0.35. This unmatched age-related change between the two sexes led to spuriously low SDR sex of 0.25 for CA125. While |BR| for between-sex difference is well above 0.375. Therefore, we chose to partition RVs by sex for CA125. Based on |r p |≥0.2 considered as a practically significant level of association ( Table 1 ), age-related changes in RVs were observed for the following analytes with their r p shown in the parenthesis: in males, CA19-9 (0.35), and CA15-3 (0.20); in females, AFP (0.33), CA125 (−0.32), CEA (0.30) and CA15-3 (0.22) in the descending order of |r p |. This female dominant age-related change of AFP and CA125 was clearly seen in Fig 1 and S1 Fig , respectively. However, the magnitude of age-related changes in terms of SDR age was all slightly below 0.4 except that of AFP in females (0.49). RVs of 11 representative analytes are shown subgrouped by sex and age (<30, 30~39, 40~49, 50≤ years). The box in each scattergram represents central 50% range and the vertical bar in the middle represents median RVs. On top of each panel, the magnitudes of between-sex and between-age variations are shown as SDR sex and SDR age derived separately for males (M) and females (F). No secondary exclusion was performed in plotting data. For GH, testosterone, and SHBG (marked by *), values of individuals with BMI≥ 28 kg/m 2 was excluded to avoid confounding of BMI on age-related changes. As SVs other than sex and age, smoking habit-related changes in RVs was noted by MRA in CEA as shown in Fig 2 . Another important factor as a SV was the Lewis blood group-related change in CA19-9. Although we have not confirmed it by actual analysis of the blood type, S1 Fig clearly showed a distinct cluster of data points below the detection limit of 0.8 KIU/L. Assuming them as representing Lewis negative individuals in Russia, its prevalence among healthy individuals are 10.5% (36/341) in males and 11.3% (45/396) in females. With these observations, we derived RIs for CEA after excluding individuals with smoking habits, and for CA19-9 after excluding individuals with values below the detection limit. RVs of 4 analytes found associated with BMI by multiple regression analysis are shown subgrouped by sex and BMI (<20, 20~24, 24~28, 28~32, <32 kg/m 2 ). In addition, RVs of CEA was partitioned by sex and the status of smoking habit. The box in each scattergram represents central 50% range and the vertical bar in the middle represents a median point. The magnitude of between-subgroup variation is shown on top of each panel as SDR due to BMI (SDR BMI ) or as SDR due to smoking habit (SDR smk ), computed separately for males (M) and females (F). From Fig 1 and S1 Fig as well as from Tables 1 and 2 , prominent sex and age-related changes were observed in all eight analytes in the reproductive panel. RVs of estradiol and progesterone in females showed an abrupt reduction at around 50 years of age (a peak time of menopause) with r p of −0.64 and −0.46 and SDR age of 0.98 and 0.63, respectively. It is notable that postmenopausal values are well below those of males. In contrast, RVs of estradiol in males stay unchanged by age, while RVs of progesterone in males decrease slightly with age (SDR age 0.43). For testosterone, between-sex difference is very prominent with female testosterone levels approximately 1/10 th of those of males. Interestingly, age-related reduction of testosterone is more prominent in females. It was shown that testosterone RVs in males were affected by BMI, but not by age (r p : −0.48 and −0.13, respectively). On the contrary, RVs in females were affected by age, but not by BMI (r p : −0.53 and 0.14, respectively). It was confirmed by a change of SDR after exclusion of patients with BMI>28, SDR age was 0.47 in females and 0.18 in males ( Fig 1 , S1 Fig ). In females, LH, FSH, and TβhCG showed an abrupt surge after menopause with SDR age of 0.88, 2.10, and 1.15, and with r p of 0.56, 0.82, and 0.62, respectively. On the other hand, in males, the age-related elevation of LH and FSH are slight and gradual with SDR age of 0.39 and 0.52, respectively. For PRL, the reduction by age is shown only in females with SDR age of 0.57. From these observations and using a criterion of SDR age ≥0.40, in females, partition of RVs by the status of menopause as self-reported in the questionnaire was essential for PRL, LH, FSH, TβhCG, estradiol, progesterone, and testosterone. For the age-related changes of FSH and progesterone in males, as a boundary value for partition, we chose 45 years of age, as roughly representing a mid-point of changes in RVs with age. Regarding BMI-related changes of the reproductive panel, in addition to testosterone, SHBG of males showed high association with r p (SDR BMI ) of −0.45 (0.35) (Tables 1 and 2 ). The trend is clearly shown in Fig 2 . For the two analytes, we examined the effect of excluding individuals with BMI≥28 on their RIs (see below). In the analyses of thyroid function tests, we first identified cases with subclinical autoimmune thyroiditis (AIT) by use of the criteria of TgAb≥4 KIU/L or TPOAb≥9 KIU/L, which are provided in the kit inserts. Prior to deriving RIs, we found the cutoff values were appropriate as a proximal point of tailing values in the distributions in S1 Fig (in the last two panels). The prevalence of individuals exceeding either of the cutoff values was 10.3% (36/350) in males and 24.5% (100/408) in females. The comparison of five thyroid function test results between individuals with and without the autoantibodies are shown in Fig 3 . It is apparent that only RVs of TSH differed between the two groups with SDR of 0.58 (male) and 0.48 (female) for the status of AIT or SDR AIT . With the results, in the subsequent analyses including derivation of RIs for all the thyroid function tests, we excluded individuals judged as AIT as well as those under thyroxine replacement therapy. A presumptive diagnosis of autoimmune thyroiditis (AIT) was made by the criterion of either TPOAb≥9 or TgAb>4 KIU/L. Thyroid function test results were compared between individuals with and without AIT. The difference of two-group centers was expressed as SDR for the status of AIT (SDR AIT ), computed separately for males (M) and females (F). By MRA, age-related reduction in RVs were observed for FT3 (r p = −0.21 in males), and TT3 (−0.21 in females). While SDR age was only 0.16 for both tests. Therefore, we chose not to partition RVs by age for any of the thyroid function tests. In fact, S1 Fig showed that age-related changes in RVs of FT3 and TT3 were not conspicuous. As for sex-related change, it was observed only in FT3 with SDR sex of 0.45 (values higher in males) ( Table 2 ). MRA revealed a conspicuous BMI-related increase of insulin and moderate decrease of GH with r p of 0.66 and –0.24, respectively, in males, and 0.61 and −0.38 in females. These trends are clearly seen in Fig 2 , but in terms of SDR BMI , only that of insulin showed high values of 0.91 (males) and 0.82 (females) in Table 2 . Therefore, we examined the effect of excluding individuals with high BMI as described below in deriving RIs for insulin and GH. For age-related changes, RVs for GH and PTH in males showed an increase with age (r p for age: 0.31 and 0.24, respectively) as shown in Table 1 . However, in terms of SDR age , that of GH is well below 0.4, apparently indicating age-related increase of GH is counter-balanced by BMI-related reduction of GH (i.e., BMI increases with age). According to the scheme for partition or secondary exclusion of RVs described above in details, RIs for all the 22 parameters were derived and summarized in Table 3 . It is of note that RIs for TPOAb and TgAb were not determined with availability of the cutoff values for diagnostic use. The first column stands for distinction between parametric (P) method and nonparametric (NP) method for RI derivation, the 4 th and 5 th columns for partitioning by sex and age, respectively, and the 6 th column for exclusion criteria. P = parametric; NP = nonparametric; LL = lower limit; UL = upper limit; Me = median; CI = confidence interval; OC = oral contraceptives; PreMP = premenopausal; PostMP = postmenopausal; AIT = autoimmune thyroiditis. The accuracy of Gaussian transformation by use of the modified Box-Cox formula is shown in S2 Fig . In the comparison between P and NP methods, 90% CI of RI limits were almost invariably narrower and upper limits tended to be higher as was described clearly in the part one of this report (the data omitted with similar tendencies). Exception for this general rule were encountered in deriving RIs for three analytes: GH, progesterone of premenopausal females, estradiol of postmenopausal females. Their RVs failed to attain Gaussian distribution even after power transformation with presence of bimodal peaks (progesterone, and estradiol) or many values below the detection limits (10% of males for GH). Therefore, NP method was used for derivation of their RIs. Among analytes with BMI-related changes (insulin, GH, testosterone and SHBG), the effect of excluding individuals with BMI≥28 was found effective for insulin in lowering UL from 15.7 to 10.5 mIU/L, but not for the other three. BR UL after partition by BMI was −1.49, which far exceeded the critical value of |BR|>0.375 ( S4 Table ). Therefore, we adopted the BMI restricted RI for insulin. As described in the Methods, after the completion of data analysis for this study, new assay methods for TSH and TβhCG became available. Therefore, we re-measured remaining serum aliquots from the volunteers stored at −80°C using the new assays after confirming the stability of the analytes. Method comparison between the old and new reagents was performed using the major-axis linear regression after logarithmic and square root transformation for TSH and TβhCG, respectively. The results are as shown in S3 Fig . Accordingly, the final RIs for TSH and TβhCG listed in Tables 3 and 4 and S4 Table were recalibrated to the values of the new reagents by use of the linear equations. PreMP = premenopausal; PostMP = postmenopausal; IFU = instruction for use; * See the right for age groups.

This is the first comprehensive Russian study for derivation of RIs for 22 major immunochemistry parameters consisting of tumor markers, thyroid function tests, vitamins, reproductive and other hormones. The study was conducted by use of the internationally harmonized protocol elaborated by C-RIDL, IFCC with recruitment of 758 well-defined, apparently healthy adults from three major cities in Russia. No regional differences among the three major cities were observed in any parameter. Careful assessment and exclusion of latent abnormal values of common occurrence was a crucial step. Close associations of BMI with RVs were observed for insulin, testosterone (M), SHGB (M), and GH (F) in that order of strength. For insulin, exclusion of individuals with BMI≥28 was effective in lowering the UL of RI, but not for others. In the derivation of the RI for CA19-9, individuals with apparent Lewis-negative blood type (M: 10.5%, F: 11.3%) were excluded. For thyroid function tests, individuals with AIT (M:10.3%, F: 24.5%) were excluded, but the procedure only affected the RI for TSH. Partition of RVs by sex was required for all reproductive hormones, CA125, CEA, and GH. Partition by age was required for AFP, for CEA (F), and for all reproductive hormones (F). A majority of RIs derived in this study differed from those provided by the manufacturers. This fact points to the importance of establishing the country specific RIs. A variable degrees of differences were noted from CDLs (or cutoff values) set by clinical guidelines for CA19-9, CA125, testosterone, insulin, and TSH. Although some of the differences are attributable to the lack of harmonization in test results, they are inevitable from theconcept of the RI as “health”-associated range, which is distinct from the concept of cutoff value that requires the case-control study for its determination.

The study protocol, including methods for invitation, provision of information about the study for volunteers, taking informed consent and questionnaire regarding current health status and lifestyle, was approved by the Ethic Committee of City Hospital #40, Saint-Petersburg. The scheme used for recruitment, sampling, and measurements was described in the first part of our report, which dealt with RIs and sources of variation of chemistry analytes [ 3 ]. In brief, 758 healthy volunteers (350, 46% men, 408, 54% women) of 18–65 year old were recruited from three regions: Sankt-Petersburg (North-West region: N = 506, 67%), Moscow (Central region: N = 117, 15%) and Yekaterinburg (Ural region: N = 135, 18%) ( S1 Table ). They were chosen according to inclusion and exclusion criteria stipulated in the C-RIDL protocol, and blood samples were drawn at basal conditions [ 2 ]. The part 2 of the report deals with RVs evaluated for a total of 24 analytes measured by immunoassays: carcinoembryonic antigen (CEA), alpha-fetoprotein (AFP), CA19-9, CA125, CA15-3, insulin, cortisol, testosterone (Testo), sex hormone-binding globulin (SHBG), estradiol, progesterone (Prog), luteinizing hormone (LH), follicle stimulating hormone (FSH), total beta human chorionic gonadotropin (TβhCG), prolactin (PRL), growth hormone (GH), thyroid stimulating hormone (TSH), free thyroxine (FT4), free triiodothyronine (FT3), total thyroxine (TT4), total triiodothyronine (TT3), anti-thyroid peroxidase antibody (TPOAb), anti-thyroglobulin antibody (TgAb) and parathyroid hormone (PTH). All were measured by using the UniCel DxI 800 immunochemistry analyzer (Beckman Coulter Inc., USA) according to manufacturer’s assay instructions and requirements. Characteristics of the analytes, including their full name, abbreviation, unit, and assay principle are listed in S2 Table . For TSH and TβhCG, after completion of the measurements, the new assays became available, and thus they were tested again using serum aliquots kept stocked at -80°C. The new TSH and TβhCG assays are standardized to the new highly purified WHO standards: i.e., 3rd International standard (IS) (81/565) and 5th IS (07/364) respectively; the properties of the previous and current reagents are compared in S3 Table . Quality control was performed in two ways. One was through twice daily measurement of 2 or 3 levels of QC specimens obtained from Beckman Coulter Inc. and Bio-Rad Laboratories, Inc. The other was through daily measurement of a mini panel composed of six sera from healthy volunteers (3 women and 3 men) as described in the common protocol [ 2 , 13 ]. Based on repeated mini panel measurements, between-day coefficient of variation (CV) was calculated for each analyte. The CV of any analyte did not exceed the allowable limit based on the criterion described in the protocol (i.e., ½ of CV I : within individual CV, presented in the EFLM website ( https://biologicalvariation.eu/meta_calculations ) ( S2 Table ). Additionally, as a part of the study aiming at worldwide comparison of RVs, the panel of 40 sera for immunoassays provided by C-RIDL were measured in four batches over a period of four days. Data analyses and statistical methods used were those recommended in the C-RIDL protocol [ 2 , 12 , 13 ]. Details are descried in Part 1 of our report [ 3 ]. The multiple regression analysis (MRA) was performed by setting RVs of each analyte as object variable and following factors as explanatory variables: sex, age, body mass index (BMI), the levels (see below) of cigarette smoking, alcohol consumption and regular physical exercise [ 14 ]. Standardized partial regression coefficient (r p ), which corresponds to the partial correlation coefficient. In reference to the Cohen’s guide [ 15 ] of “effect size” for correlation coefficient (r): r = 0.1 (small) and r = 0.3 (medium), as a middle point, |0.2| ≤r p was interpreted as a practically significant factor influencing the reference values. The levels of smoking, alcohol consumption and physical exercises were classified into 3, 5 and 8 categories, respectively, using the following criteria: none, ≤20, > 20 cigarettes/day; none, 50 g ethanol/day; none, 1–7 days/week. By use of 3-level nested ANOVA, between-sex, between-age, between-city, and between-individual variations were computed each as standard deviation (SD): SDsex, SDage, SDcity, and SDbtw-indiv (= SD G by common notation). Relative magnitude of each SD to the SD G was calculated as SD ratio (SDR): SDR sex , SDR age , and SDR city . After absence of between-city differences was confirmed by the criterion described below, SDR age specific for each sex was calculated by one-way ANOVA. As an additional analysis for analytes with obvious BMI-related changes, SDR for BMI (SDR BMI ) was computed for each sex by two-level nested ANOVA with age set as a covariate. The need for partition of RVs was considered by setting SDR≥0.4 as a primary guide [ 2 ]. However, SDR may be too sensitive when the width of RI that constitutes the denominator of SDR is narrow. Conversely, SDR may be insensitive when between-subgroup differences occur only at the periphery of distribution (LL or UL) because SDR represents between-subgroup bias at the center of the distributions. Therefore, we additionally considered actual difference (bias) at LL or UL as “bias ratio” (BR) using the following formula illustrated for a case of gender difference: BR LL = | LL M - LL F | ( UL M F − L L M F ) / 3.92 , BR UL = | UL M - UL F | ( UL M F − L L M F ) / 3.92 where subscript M, F, and MF represent male, female, and male+female, respectively. The denominator of each formula represents the standard deviation (SD RI ) comprising the RI, the width of which corresponds to 3.92 times SD RI . In accordance with the convention of allowable bias specification of a minimum level: 0.375 × SD G (= SD RI ) [ 16 ], we regard BR UL >0.375 as an auxiliary threshold for partitioning RVs when SDR does not match to actual between-subgroup difference at ULs (or LLs). In performing MRA and ANOVA, RVs of analytes that exhibited highly skewed distributions were transformed logarithmically. The corresponding analytes were marked in Table 1 . In that case, SDRs were computed by reverse transformation of each SD component that was calculated under the transformed scale as described elsewhere [ 17 ]. RIs were derived by both parametric and nonparametric methods. The former was performed after normalizing data by use of the modified Box-Cox power transformation formula [ 18 ]. The validity of the parametric method was confirmed by the linearly of cumulative distribution of RVs on probability paper plot [ 13 ] and by the Kolmogorov-Smirnov test. If the transformation failed, the non-parametric method was used. The 90% confidence interval (CI) of the lower limit (LL) and upper limit (UL) of RI was calculated by the bootstrap method through random resampling of the same dataset 50 times. Accordingly, the final LL and UL of RI both by parametric and nonparametric methods was chosen as the average of iteratively derived LLs and ULs.

(XLSX) Click here for additional data file. Distributions of RVs for all the analytes were shown after subgrouped by sex and age. No secondary exclusion was performed in plotting data. The box in each scattergram represents central 50% range and the vertical bar in the middle represents a median point. On top of each scattergram, the magnitudes of between-sex and between-age variations are shown as SDR sex and SDR age derived separately for males (M) and females (F). (PDF) Click here for additional data file. RIs were derived by both parametric and nonparametric methods. The accuracy of Gaussian transformation by Box-Cox formula can be assessed from theoretical Gaussian curves in two histograms shown on left top (before and after the transformation). The results of by Kolmogorov-Smirmov (K-S) test for normality of distribution were shown on right upper panel. The accuracy of the transformation can be also seen from the linearity in the probability paper plot on the right. The limits of the RI by nonparametric method corresponds to the points where red zigzag line intersect with horizontal 2.5 and 97.5% red lines of cumulative frequencies. (PDF) Click here for additional data file. Aliquots of volunteers’ sera stored at −80C° were tested in 2018 by use of new reagents for TSH and TβhCG after confirmation of the stability of the analytes. Recalibration of values by the old reagent was performed using the major-axis linear regression between new and old values after logarithmic and square-root transformation for TSH and TβhCG, respectively. (PDF) Click here for additional data file. (PDF) Click here for additional data file. (PDF) Click here for additional data file. (PDF) Click here for additional data file. (PDF) Click here for additional data file.