Derivation of sex and age-specific reference intervals for clinical chemistry analytes in healthy Ghanaian adults

Study population and recruitment

Healthy volunteers were recruited from Accra in the South and Tamale in the North, which represent two different ethnic groups. Recruitment was based on the C-RIDL protocol for harmonized implementation of multicenter studies [7]. A target sample size of 500 was set with equal sex and age (18–70 years) distributions. Through posters and media advertisement, volunteers were recruited from public institutions, health facilities, churches, etc. A questionnaire based on the protocol was administered to record volunteers’ health status.

Inclusion/exclusion criteria

The exclusion criteria by the protocol were individuals with known diabetes mellitus under drug therapy, recent (≤14 days) recovery from acute illness, injury, or surgery requiring hospitalization, known carrier state of hepatitis viruses or HIV, pregnancy or within one year after childbirth. Additionally, we excluded individuals with body mass index (BMI)>35 kg/m², alcohol consumption≥70 g/day, smoking habit of >20 cigarettes/day, under regular medication. Upon completion of the questionnaire, the eligible volunteers from Accra were scheduled for blood collection at Medlab Ghana, while specimens taken in Tamale were transported to Medlab Ghana in Accra within 6 h each day. Participants’ results were reported individually upon completion of measurements.

Laboratory analysis

Target analytes and measurements

The 34 analytes measured in serum were ALP, GGT, AST, ALT, LDH, CK, amylase (AMY), triglycerides (TG), total cholesterol (TC), HDL-C, LDL-C, Na, K, Cl, Ca, inorganic phosphate (IP), Mg, total CO₂ (TCO₂), anion gap (AG), total protein (TP), albumin (Alb), total bilirubin (TBil), direct bilirubin (DBil), urea, creatinine (Cre), uric acid (UA), glucose (Glu), HbA_{1c ,} CRP, complement component 3 and 4 (C3, C4), IgA, IgG and IgM. Six calculated parameters were globulin (Glb=TP − Alb), adjusted-Ca (aCa=Ca − 0.025 × [Alb − 40]), eGFR (CKD-EPI) [14], TC/HDL, HDL/LDL and non-HDL cholesterol (nonHDL=TC–HDL-C).

Chemistry analytes were measured in batches on AU480 auto-analyzer (Beckman-Coulter, US). While inflammatory markers were measured by immunoturbidometric assays using Centaur-XP auto-analyzer (Siemens, Germany). HbA_1c was measured by Beckman Coulter AU 460 using the latex agglutination inhibition method. Detailed analytical methods and precisions are summarized in Supplementary Table 1.

Quality control

All measurements were carried out in Medlab Ghana and SynLab, Germany, all ISO15189 accredited Medical Laboratories, according to the laboratories’ SOPs and Quality Manual to ensure reliable measurements. Additionally, a mini-panel of sera were prepared from healthy volunteers and measured each day in singleton to monitor the stability of the assay according to the protocol [7, 12].

Standardization based on value-assigned serum panel

Participating laboratories to the C-RIDL global study received a panel of 50 sera that had assigned values for 30 chemistry analytes [12]. The sera were measured to enable recalibration of reference values (RVs), where necessary. The test results also allowed for direct comparison of RVs among collaborating laboratories.

Ethical approval

The study protocol conforming to the Declaration of Helsinki was approved by the Ethical Review Committee of the College of Health Sciences, University of Ghana (CHS-Et/M.8-P4.14/2016–2017). Written consents were obtained from all volunteers. Participants’ confidentiality was maintained at all times.

Statistical analyses

Demographic profiles of participants (Table 1)

The total sample size of reference individuals included in data analyses was 501, comprising 54% (n=270) males and 46% (n=231) females. The mean age and SD was 41.3 ± 13.5 years with almost flat distribution. The volunteers were from Accra (77%) and Tamale (23%). Regarding ethnicity, the participants were grouped into Akan and non-Akan based on their common language and diet. The Akan constituted 40.3% (n=202) and non-Akans 59.7% (n=299). Mean ± SD of BMI was 24.9 ± 3.5(M), 26.9 ± 4.0(F) kg/m², systolic/diastolic blood pressures were 121.2 ± 11.5/78.9 ± 7.2 mmHg. Almost all participants (99.4%) recruited were non-smokers, which is the cultural norm in Ghana. It is notable that individuals with any allergy, such as allergic rhinitis or atopic dermatitis, was as low as 1.9% in males and 0.4% in females. In relation to the testing of HbA_1c, we retrospectively ascertained the lack of disorders that may have affected its value, such as sever iron deficiency anemia, based on our previous report of hematology RI study [18]. While possible influence of sickle cell anemia in adults in calculating the RI was unlikely judged from its new-born incidence in Ghana of 2% [19].

Sources of variation analysis by MRA

The results of MRA performed for exploration of sources of variation (SVs) of each analyte are summarized in Table 2. By setting |r_p|>0.2 as the threshold, age and BMI were the key factors associated with many of the analytes. The notable age-related changes were mostly found in females for Na, urea, TCO₂, Cre, TG, UA, Mg, ALP, and TC in the descending order of strength, and only for TCO₂ and IgA in males. While, age-related decrease (r_p<−0.2) was observed in IgM, TP, and Glb in female, and Alb, TP, and Ca in males. Since there is no association of age in adjusted Ca (aCa), the decrease in Ca is obviously confounded by age-related decrease in Alb. These age-related changes are shown for BMI and RVs of eight representative analytes in Figure 1.

Table 2:

Multiple regression analysis for sources of variation of reference values.

1. R=multiple correlation coefficient. Akan=dummy variable indicating ethnicity (Akan=1; nonAkan=0) Alcohol=habit of drinking alcohol (Yes=1; No=0); ExerLvl=days/week of regular exercise (0–7); HrStand=hours of standing/day. The width of bar in each cell represents partial correlation coefficient (r_p) with negative value shown by red bar. |r_p|≧0.2 was marked by bold font.

Figure 1:

Sex and age-related changes of representative chemistry analytes.

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 of each subgroup is shown at the right bottom of the age group labels. Because no secondary exclusion was done for RVs, the range of the scattergrams may not match to the RI, which is to be derived with inclusion of single mean ± 2.81 SD truncation step under the transformed scale.

With respect to BMI-related changes, analytes tested in males showed generally higher associations with BMI compared to females. Positive association of RVs with BMI (r_p>0.2) were found in males for C3, TC/HDL, UA, nonHDL, ALT, GGT, LDL-C, TC, TG, Glu, HbA_1c, in the descending order of influence; while, in females for C3, CRP, ALT, TG, TC/HDL, UA, nonHDL, HbA1, LDL-C, and GGT; Conversely, negative association of RVs with BMI (r_p<−0.2) were noted for HDL/LDL in both sexes, and HDL-C in males, and AMY in females. These findings are shown graphically for 12 representative analytes in Figure 2.

Figure 2:

Association of BMI with RVs of nutritional/inflammatory markers.

Associations of BMI (on x-axis) with RVs of 12 analytes (on y-axis) are shown separately for males (blue dots) and females (red dots) by 2D-scattergram. The least-square regression lines are drawn and Spearman’s correlation coefficients are calculated for both sexes.

None of the analytes, including GGT or ALT, showed any significant association with drinking, obviously attributable to infrequent drinking habits of our volunteers: 13.2% (male: 15.6% female: 10.4%). The level of regular exercise or hours of standing per day was not significantly associated with any analyte. We did not include smoking habit in MRA because only three subjects were smokers. Using the dummy variable of Akan or not to examine the effects of ethnicity, only AG and IP showed weak associations in both sexes: i.e., their RVs in Akan were lower than non-Akan.

Standard deviation ratio (SDR) as a guide for partitioning RVs

The magnitude of between-sex differences is shown in Table 3 expressed as SDR_sex that was derived by two-level nested ANOVA with age as covariate. SDR_sex≥0.4 was observed for Cre, UA, ALT, Alb, AST, CK, DBil, TBil, Cl, GGT, IgM, C3, and Ca, in the descending order of magnitude. The direction of gender difference in RVs was all higher in males except for Cl, and IgM. For age related change, SDR_age specific for each gender was derived as SDR_ageM or SDR_ageF by one-way ANOVA. For males, only SDR_ageM for Alb and eGFR were significant, while for females, SDR_ageF were significant for Na, eGFR, Ca, TG, and ALP. From Supplementary Figure 1, the direction of changes is increase with age for all analytes except for Ca and Alb.

Table 3:

SDRs representing between-sex, between-age differences.

1. The value of SDR ≧0.4 was marked by bold letter, and SDR ≧0.5 was highlighted by a gray shade. SDRage was computed for males (M) and females (F), seprately as SDRageM and SDRageF.

Derivation of RIs

RIs were calculated in four ways by parametric (P) and nonparametric (NP) methods with/without application of the LAVE method. The comparisons of RIs for all analytes are shown in Supplementary Figure 3. It was obvious that UL by NP method was biased to the higher side and its 90%CI was generally wider especially for analytes with skewed distribution such as ALT, CK, GGT, and CRP. Judged from BR of >0.375 with/without LAVE shown in Supplementary Table 2, the LAVE procedure had a notable effect in lowering ULs for AST, ALT, GGT, TC, TG, nonHDL, LDL-C, TC/HDL, CRP in both sexes, UA and Cre in females, and CK and IgM in males.

The effect of excluding individuals with high BMI (≥30 kg/m²) was examined for the analytes with high r_p by MRA as shown in Supplementary Table 3. Based on BR of >0.375 as a significant effect of BMI restriction, we adopted the BMI-restricted RIs for UA, TC, LDL-C, and CRP for both sexes and, for nonHDL and TG for females only.

In Figure 3, we illustrated the effect of LAVE procedure and/or BMI restriction for four representative analytes: ALT, GGT, LDL-C and CRP. Effect of LAVE in lowering UL of the RI was observed more or less in all four analytes, while additional lowering of UL by BMI restriction was noted only in LDL-C and CRP. In other words, for RVs of ALT and GGT, the LAVE procedure had nearly the same effect as the BMI restriction.

Figure 3:

Effect of LAVE and BMI restriction on RIs of nutritional markers.

RIs of four representative nutritional markers were calculated for each sex in four ways: (1) using all RVs [1: All (2 + 3)], (2) using RVs that remained after latent abnormal values exclusion (LAVE) procedure [2: LAV Ex], (4) using RVs from individuals with BMI<30 kg/m² [4: BMI<30], (5) using RVs that remained after BMI restriction plus LAVE [5: 4 & LAV Ex]. The RI determined for each category of RVs is shown as rectangular region drawn over the scattergram. The scattergram (3) in the third row of each sex represents the subset of RVs that were excluded by the LAVE procedure [3: LAV]. Hence the scattergram (1) is broken down into scattergrams (2) and (3). In applying LAVE, the following mutually associated reference tests were used for identifying individuals with latent abnormal values: Alb, Glb, UA, Glu, AST, GGT, CK, LDH, TC, nonHDL, TG, and CRP. It is of note that, there seems to exist excessive number of points outside the RI overlaid, but, in the derivation of the RI, values outside mean ± 2.81 SD were truncated once.

In Table 4, a list of RIs adopted for all analytes are presented. For the selection of RIs with/without partition by sex and age, we principally used SDR_sex and SDR_age, but we also considered the bias at the RI limits (B_LL and BR_UL), which are presented in Supplementary Table 2. For partitioning RVs by sex based on SDR_sex≥0.4, the analytes selected were as noted above. However, actual between-sex difference at RI limits for Cl and Ca was less than three reporting units, and thus their RIs were determined without partition by sex. Whereas RVs for eGFR, IP, LDH, ALP, AMY, nonHDL, TG, TC/HDL, CRP and C4 recorded SDR_sex<0.4, yet, their actual between-sex differences at LL or UL expressed as BR_LL or BR_UL were high (>0.57), and thus their RIs were determined for each sex. For partitioning by age, SDR_age≥0.4 was noted, in females for Na, Ca, ALP, and TG, in males for Alb, in both sexes for eGFR (Supplementary Table 3). However, for Alb, Na, and Ca, actual differences at RI limits after partition at 45 years of age were small (3 g/L, 2 mmol/L, 0.06 mmol/L, respectively), and thus no partitioning was done (Supplementary Table 4). For TG in females, despite SDR_age>0.4, BR_UL after partition by age was small (<0.57). Hence, the RIs was not partitioned by age.

Table 4:

List of RIs after recalibration.

1. M=male; F=female; MF=M+F; LL=lower limit; Me=mid-point; UL=upper limit; CI=confidence interval. LAVE=latent abnormal values exclusion. (−) LAVE denoted with color light green shows that the LAVE method was not applied. (+) LAVE with a green color shows LAVE method was used. Yellow color code depicts the reference interval derived for all age and Bmi. Violet color code represents both male and female (M + F) reference interval. light blue color shows male reference interval. pink shows female reference interval.

Standardization of RIs

Comparisons of our test results for 50 panel sera with assigned values are shown in Supplementary Figure 2. The need for recalibrating the derived RI (LL∼UL) was assessed by computing adjusted RI (LLa∼ULa) by using major axis regression line coefficients [6] shown in the Figure, and by calculating BR_LL and BR_UL.

By regarding BR>0.375 as exceeding allowable analytical bias, recalibration of RIs was performed for Alb, Ca, IP, AST, ALT, LDH, ALP, and HDL-C, but not for Na and Cl because of high imprecision of slope b [CV(b)>11%] of the regression line as described elsewhere [6].

Limitations

Our volunteers largely represented the urban population, whose socioeconomic status, nutrition and lifestyles are likely to be different from the rural population. Hence some of derived RIs may differ in the rural setting. Another limitation was that we relied on self-reporting of health status and might have inevitably included individuals with latent, metabolic or inflammatory disorders. However, the application of the LAVE method and BMI restriction, if appropriate, seem to have coped with those conditions or in compliance to the sampling requirements.