Reference intervals for 33 biochemical analytes in healthy Indian population: C-RIDL IFCC initiative

Subjects

Under the aegis of IFCC’s C-RIDL, a cross-sectional observational study for defining RIs was conducted at Biochemistry department, Hinduja Hospital, Mumbai, India. A common protocol recommended by C-RIDL [4] was used by all the participating countries in the IFCC global study. The study design was approved by hospital’s Institutional Review Board (IRB) Committee, and throughout the study, adherence was made to the Declaration of Helsinki. A final sample size of minimum 500 healthy subjects was decided maintaining a gender ratio of 1:1 across all the different age groups. A priori approach was undertaken for selection of reference individuals who were mainly between 18 and 65 years of age and were mostly on no medications. Individuals above 65 years were also recruited, provided they comprise less than 20% of the study population. Further, the exclusion criteria were mainly those who were diabetic, had history of liver or kidney disease, donated blood as a donor in the past 3 months, HBV, HCV or HIV carrier and female volunteers who were pregnant or within 1 year after childbirth [4]. Primarily, employees working in Hinduja Hospital were recruited; however, their family members and those in healthcare business associated with the hospital were also recruited. Informed written consent was obtained from each volunteer prior to inclusion in the study. A health status questionnaire, which recorded information on the anthropometric measures, medical status, smoking, alcohol habits, physical activity and medications, if any, were obtained from each volunteer.

Sample collection and processing

Prior to blood collection, volunteers were requested to avoid excessive physical exertion/exercise for 3 days before the sampling and were also requested to avoid excessive eating/drinking the night before and fast overnight for 10–12 h. The sample collection started at 7:00 am till 10:00 am. To avoid variation due to postural influence and physical stress, the volunteers were requested to sit for 15–20 min prior to collection [5]. Approximately 8–10 mL of blood was collected in BD Vacutainers serum tubes, which after 45–60 min was centrifuged at 2000 g for 10 min. The serum was pooled, aliquoted (1.5 mL each) and immediately stored at −80°C until analysis.

Target analytes and measurements

The frozen serum aliquots were measured collectively for 25 analytes by Beckman Coulter Unicel DxC 800 autoanalyzer: alanine aminotransferase (ALT); albumin (ALB); alkaline phosphatase (ALP); amylase (AMY); aspartate aminotransferase (AST); calcium (Ca); chloride (Cl); creatine kinase (CK); creatinine (CRE); γ-glutamyl-transferase (GGT); glucose (Glu); high-density lipoprotein cholesterol (HDL-C); inorganic phosphate (IP); iron (Fe); lactate dehydrogenase (LDH); low-density lipoprotein cholesterol (LDL-C); magnesium (Mg); potassium (K); sodium (Na); total bilirubin (TBil); total cholesterol (TC); total protein (TP); triglycerides (TG); Urea; uric acid (UA), and measured for eight analytes by Beckman Coulter Immage 800 autoanalyzer: complement component 3, and 4 (C3, C4); high-sensitivity C-reactive protein (hsCRP); immunoglobulin G, A, and M (IgG, IgA, IgM); transferrin (Tf); transthyretin (TTR) using the manufacturer’s reagents, calibrators and controls as instructed. The Unicel DxC 800 autoanalyzer principle was based on ion selective electrode, conductivity electrode, glucose oxygen sensor, colorimetric, immuno-turbidimetric, enzyme immunoassay and rate end point, whereas the Immage 800 autoanalyzer was based on the principle of rate nephelometry and rate near infrared particle immunoassay (NIPIA).

Quality control (QC) and traceability

Internal QC was run for each analytes at both normal and pathological concentrations prior to sample testing. Internal QC monitoring was undertaken by preparing a mini-panel of sera from five healthy volunteers, which were measured with each subsequent batch of sample testing in order to monitor the between-day stability of the assay (Supplementary Table 1) [4], [6]. Further, to determine the traceability and to make the test results comparable among countries participating in the IFCC global study, a panel of 40 sera were provided by IFCC C-RIDL, which were measured in batches along with the volunteers’ samples. The panel of sera were assigned values for 27/33 analytes (except ALB, Mg, Fe, AMY, LDH and ALP) by making it traceable to the reference measurement procedures [7]. The major axis linear regression between the assigned values and the observed values for panel of sera was used to assess the need for recalibration of the test results. When difference in the lower or upper limit of the RI before and after recalibration was more than 1/4 of standard deviation (SD) of the original RI, recalibration of test results was performed by using regression coefficients (intercepts and slopes). This criterion corresponds to Frasers’ allowable limit of bias [8], [9], [10].

Analysis of source of variation and partitioning

Multiple regression analysis (MRA) was performed to analyze potential SV such as age, BMI, drinking and physical exercise. The significance of each SV was evaluated as standard partial regression coefficient (rp), which corresponds to partial correlation coefficient and takes value between −1.0 and 1.0. We regarded 0.2≤|rp|<0.4 as slight to moderate association and 0.4≤|rp| as prominent association of test results with a given SV [6]. Nested analysis of variance (ANOVA) was performed to judge the need for partitioning RVs [4], [6], [11]. It separates the magnitude of variations attributable to each SV (age, gender, BMI, etc.) and expresses it in terms of SD. Derived were the variations of between-age (SDage), between-gender (SDsex), between-BMI (SD-BMI) components and net between-individual (SDindiv) component, which is adjusted for other SVs. We computed the SD ratio (SDR), which is the SD of a given SV over the SD-indiv. A ratio greater than 0.3 has been shown as a guide value for considering the need to partition RVs [6], [11]. The general-purpose statistical software, StatFlex for Windows version 6.0 (Artec Co. Ltd., Osaka, Japan), was used to carry out the MRA and nested ANOVA.

Derivation of RIs

For derivation of RIs, both non-parametric and parametric methods were used for comparative purposes. For the latter, modified Box-Cox power transformation formula [12] was used for Gaussian transformation of RVs, and values outside mean±2.57 SD were truncated once before computing the central 95% interval [12], [13], [14]. In the derivation of the final RIs, the effect of latent abnormal values exclusion (LAVE) method [6], [11] was examined. In applying the LAVE procedure, we allowed only one abnormal result (LAVE(+)Abn=1) among 12 reference analytes: Glu, TG, LDL-C, HDL-C, UA, ALB, AST, ALT, LDH, GGT, CK and hsCRP. They were chosen to detect individuals with metabolic syndrome, inflammation, insufficient fasting and muscular exertion prior to the sampling. Ninety percent confidence intervals for the lower and upper reference limits (LL, UL) were calculated using the bootstrap method with iteration of 100 times, for both parametric and non-parametric methods. Making use of this procedure, the limits of the final RIs were set as the averages of repetitively computed LLs and ULs.

Characteristics of study volunteers

A total of 512 volunteers were recruited primarily from Hinduja hospital employees (Table 1) with wide differences in their level of physical activities, right from managerial or hospital administrative personnel to physically demanding work profile such as that of laboratory technicians, nurse and hospital transporters. With respect to nutritional status, the BMI profile of volunteers was almost similar to that seen in general Indian population. Similar approach for volunteer recruitment was also adopted in the IFCC Asian study where more than 3500 volunteers from seven countries were recruited mainly from hospital personnel [15], [16].

Analysis of source of variation (SVs)

MRA was performed analyte by analyte to identify the potential SVs (Table 2). The values shown are standard partial regression coefficients (rp) with positive sign highlighted by a background color of yellow (0.2–0.39) and orange (≥0.4) and negative sign highlighted by a light blue (0.2–0.39) or blue (≥0.4) background color. Those analytes with non-Gaussian skewed distribution of test results, such as TG, AST, ALT, GGT, TBil, hsCRP and CK, were logarithmically transformed before performing MRA.

The MRA revealed that age-related changes were observed for Glu in both sexes: TP, ALB and Ca in males; TC, LDL-C, TG, ALP, Urea, CRE, UA, LDH, IgA and IgM in females. BMI-related changes were observed for HDL-C, hsCRP, GGT, UA and C3 in both sexes; LDL-C, TG, AST and ALT in males; and Glu, Alb and AMY in females. However, regular exercise and alcohol consumption did not show any practically significant associations with the test values.

Partitioning of reference values

To determine whether there is need to establish different RIs with respect to gender and age, nested ANOVA was carried out (Table 3). Actual differences in LL and UL of RIs before and after partition were rather small when SDR is near 0.3. Therefore, we adopted a criterion of SDR≥0.4 for both gender and age partitioning. However, some analytes (TC, TG, Urea, ALP and IgM) with SDRsex<0.4 were partitioned by sex when SDRage for either males or females was around 0.4 with noticeable difference in the UL between age groups.

Therefore, gender partitioning was required for 17 analytes, TC, HDL-C, TG, hsCRP, ALB, AST, ALT, ALP, GGT, TBil, Urea, CRE, UA, Fe, TTR, CK and IgM, whereas age partitioning required five analytes, Glu, ALB, TC, ALP and Urea. Sex- and age-related changes of RVs are shown as one-dimensional scattergram for eight representative analytes in Figure 1, and for all analytes in Supplementary Figure 1. Sex- and BMI-related changes are also shown for four representative analytes in Figure 1 and for total of 12 selected analytes with known association with BMI revealed by the MRA in Supplementary Figure 2.

Figure 1:

Distribution of RVs partitioned by sex and age or BMI for selected analytes.

For the first eight analytes, RVs of each analyte were first partitioned by sex and then partitioned by age into four groups at the boundary of 30, 40, 50 and above 50 years. For the last four analytes, RVs partitioned by sex and by BMI using boundary values of 22, 26 and 30 kg/m². The box and center vertical line in each scattergram represent the central 50% range and the median of RVs. Please note that all the RVs were plotted without any exclusion, and thus the distribution range by visual inspection may not match the RI of respective analyte.

Derivation of RIs

The RIs were derived using both parametric and non-parametric methods after applying secondary exclusion based on the LAVE method. RIs derived by parametric method and their 90% confidence limits were in general narrower than those observed by non-parametric methods as shown for representative analytes in Figure 2, and for all in Supplementary Figure 4. Therefore, we adopted RIs by parametric method for all analytes, except for hsCRP that failed to give Gaussian shape by the power transformation. Interestingly, the figures also show that, for majority of analytes, there was little or no difference in the RIs by applying the LAVE method (see Discussion for possible reasons). These findings are clearly shown in Supplementary Table 2, which presents the complete RI data derived by parametric and non-parametric method with 90% CI, and with/without LAVE method. The final list of RIs adopted in consideration of SDRsex and SDRage is shown in Table 4.

Figure 2:

Comparison of RIs and their 90% CI derived in four ways.

The RIs were derived in four ways: by parametric (P) or non-parametric (NP) method with or without application of latent abnormal values exclusion method (LAVE) method. Each horizontal bar represents the RI, and the vertical line in the center corresponds to the midpoint. The shades on both ends of the bar represent 90% CI for the limits of the RI predicted by the bootstrap method. The results for six representative analytes for males (M) and females (F) are shown in this figure. The same figures for all analytes are available in Supplementary Figure 4.

Quality control monitoring and traceability

All the five sera in the mini-panel analyzed with every batch (approximately 100 samples/batch) of sample testing had an acceptable between-day CV of all <10% (Supplementary Table 1), which were within the desirable limits [17]. For traceability, the panel of 40 sera was measured, correlated with the assigned values [7] and presented as correlation matrix graph (Supplementary Figure 3). RVs for 10 analytes (Glu, TC, LDL-C, CRE, AST, ALT, GGT, TBil, C3 and C4) exhibited bias in either LL or UL judged by the criteria described elsewhere [8], [9], [10], and thus, they were recalibrated using the major-axis regression line.