Within- and between-subject biological variation data for tumor markers based on the European Biological Variation Study

Sample collection and handling

In total, 91 healthy volunteers (53 females, 38 males; age interval, 21–69 years) from six centers in five different countries (Italy, Norway, The Netherlands, Spain, Turkey) were enrolled in the EuBIVAS (Supplemental Table 1). The study protocol was approved by the Institutional Ethical Review Board of San Raffaele Hospital and Ethical Board/Regional Ethics Committee of each participating laboratory in agreement with the World Medical Association Declaration of Helsinki. All volunteers provided written informed consent prior to the study procedure. Fasting blood samples were drawn weekly, between 08.00 and 10.00 am, for 10 consecutive weeks (April–June 2015). Serum samples were stored at −80 °C until all samples were collected and then sent, frozen in dry ice, to the coordinating center, San Raffaele Hospital in Milan, Italy and stored in frozen at −80 °C until analysis.

Analytical methods

All measurements were performed on the Roche Cobas e801 (Roche Diagnostics) at San Raffaele Hospital. Samples from the same person were measured in duplicate in a single run. The measurement system, reagents, calibrators and control materials used are detailed in Supplemental Table 2.

Data analysis

The Røraas method [13] was used to derive estimates of CV _I and CV _G as described in the following. Firstly, measurement results lower than limit of quantitation (LOQ) and thereafter, outliers as identified based on Dixon-Q test, were excluded for each TM. To verify that all participants were in steady-state, a linear regression was performed on the mean group value over the whole study period for each measurand. If the 95% confidence interval (CI) of the slope of the regression line included 0, the participants were considered to be in steady state, and if not, data were adjusted by applying the inverse of the regression formula to all measurement results at each week.

CV _I was estimated by using CV-ANOVA [13]. The homogeneity of analytical (between replicates) and within-subject variabilities were verified on CV-transformed data using Cochran and Bartlett tests, respectively. The normality of the residuals was verified using the Kolmogorov Smirnov test. The CV _G was estimated on natural-logarithmic transformed data using ANOVA after excluding outliers (Supplemental Table 3) and verifying the normality of data for all measurands.

We arbitrarily accepted 50 years of age as the cut-off value for premenopausal and postmenopausal women and performed separate analysis for women below and over of 50 years. We also estimated the CV _I and CV _G of CEA and HE4 for smokers and non-smokers, separately [14], [15].

The 95% CIs of all BV estimates were calculated as described by Burdick and Graybill [16]. The lack of overlap between the 95% CIs was used to identify significant differences between BV estimates.

Kruskall-Wallis and Mann-Whitney U tests were used to evaluate the difference between the median concentrations of TMs of different countries. The correlation between the concentrations of TMs and ages of subjects were analyzed by Pearson (parametric variables) and Spearman r (non-parametric variables) tests.

APS (based on BV data) for analytical imprecision (CV_APS), and bias (B_APS), RCV, II and the number of samples required to estimate the homeostatic set points (NHSPs) based on BV data were calculated using the equations given below. For most measurands, BV estimates derived from all subjects were used, but when CV _I estimates between men and women were significantly different, as judged by the lack of overlap of the 95% CIs, APS, RCV and II were calculated for each subgroup separately. APS for bias was not calculated when the CV _G estimate of the analyte was above 33% which indicates the skewed distribution of homeostatic set points of individuals [17]. Additionally, when the mean concentration of a TM in the subgroups of men and women were significantly different, the lowest CV _G estimate was used to calculate the APSs.

RCV were estimated using the equations given below [13]:

where SD _A,log is the analytical SD calculated from the back-log transformation of CV _A obtained from the duplicate measurement of study samples; the SD _I,log, is the within-subject SD calculated from the CV _I estimates; and the SD* is the combination of the SD _A,log and SD _I,log. For a significant unidirectional change, z value was accepted as 1.65 for the probability level at 95%.

NHSP were estimated using the following formula:

where D is the allowed percentage deviation from the true homeostatic set point (HSP). NHSPs were calculated for 5, 10 and 15% deviations from the true homeostatic set points, using CV _A estimates from the duplicate analysis of study samples.

Alpha‐fetoprotein

Increased AFP concentration has been associated with poor prognosis of hepatocellular carcinoma and also recurrence after treatment [24], [25], [26]. The marked individuality of AFP supports that RCV should be used in monitoring of patients rather than cut-offs such as popRI. Previously published BV estimates and associated RCV of AFP are based on studies in healthy [22], testicular cancer [21] and subjects with hepatic disease [20]. Erden et al. measured AFP in adult healthy subjects and reported a CV _I estimate with 95% CI of 26.7% (23.8–30.3) [22], i.e. significantly and substantially higher than our estimate of 4.1% (3.9–4.4). The lack of outlier and homogeneity analysis and the small number of samples (4 samples per subjects) in the Eden study [22] might explain this discrepancy. Trape et al. also have measured the BV of AFP in patients with testicular cancer [21] and hepatic disease [20] and reported CV _I estimates of 12.5 and 38%, and RCV of 33.3 and 88.9%, respectively. In our study, we found significantly lower CV _I s in males (3.4% [3.0–3.8]) than females (4.6% [4.1–5.0]) (Table 1), which indicates that different RCV should be applied, however, the difference between is quite small (RCV_increase males; 11.1%, females; 13.6%) (Table 2).

Human epididymis protein 4

HE4 is widely used as a biomarker for ovarian cancer and elevated concentrations of HE4 have also been observed in lung, endometrial, breast, gastrointestinal and renal cancers [27].

Braga et al. [28] analyzed HE4 in serum in 14 premenopausal and 14 postmenopausal healthy women sampled monthly for 4 consecutive months and reported CV _I estimates for HE4 of 12.1 and 6.5%, respectively, and a common CV _G estimates for both groups of 16.4%. We found a significantly lower CV _I estimate in postmenopausal (4.5% [3.9–5.2]) than premenopausal women (8.1% [7.6–8.8]), but similar CV _G estimates as Braga (Table 1). The IIs, reported by both Braga’s (0.59) and this study (0.35) indicated marked individuality for HE4, which supports the use of this marker in monitoring of patients.

Braga et al. [28] found no difference in median concentration for HE4 in postmenopausal females as compared to younger females, whereas Urban et al. [15] reported increasing concentrations of HE4 with age. In our study, we found lower CV _I , but higher concentrations of HE4 in females above 50 years, as compared to those below (Table 1). Additionally, there was a significant, positive correlation between the median concentration of HE4 and age of all subjects (Figure 3). In consecutive measurement of HE4, we found that RCV_increasing is 21% for young females and 11% in postmenopausal women. The lower CV _I s in postmenopausal women reported both by Braga et al. [28] and this study indicate that smaller changes in HE4 concentrations may be considered significant in postmenopausal than in fertile women.

Higher HE4 concentrations have previously been reported in female smokers [15] and therefore, we analyzed BV components separately for smoking (n=11) and non-smoking (n=42) females, but no significant differences were observed (Table 1).

Cancer antigen 125

CA 125 is considered the most robust serum biomarker for ovarian cancer, but elevated concentrations of CA 125 are also observed in benign gynecological diseases and even in heart failure and liver cirrhosis [29], [30].

Braga et al. [28] have reported a CV_I estimate of 9.09% in healthy subjects, whereas Tuxen et al. [31] reported a much higher results, of 23.3%. The study of Tuxen et al. was conducted in 1999, used radioimmunoassay to measure CA 125 and did not perform analysis of homogeneity of variances. Trape et al. [23] measured CA 125 in two groups; surgically treated colon adenocarcinoma (control group, Astler–Coller classification stages A or B1) and non-small cell lung cancer patients (subjects with complete remission treated with chemotherapy and total tumor resection). They reported CV _I estimates of 21.1 and 22.5% for these two groups, respectively, and a common RCV of 53.1%. In our study, we found CV _I of 8.6% (8.0–9.3) and RCV_increase of 23%, which are compatible with the data reported by Braga et al. [28]. Additionally, we found that the CV _I for CA 125 was significantly lower in postmenopausal females (6.1% [5.3–7.3]) than in younger subjects (9.4% [8.6–10.2]) (Table 1). Braga et al. [28] reported similar results (9.11 and 9.07%) for the two groups, but Tuxen et al. [32] found CV _I estimates of 28.9 and 12.6% for young and postmenopausal women, respectively. Similar to HE4, smaller changes in serum CA 125 concentrations might be of significance in postmenopausal than fertile women.

Carcinoembryonic antigen

CEA is widely used in the management of gastrointestinal tumors. However, it is not specific to gastrointestinal system tumors and is also used in combination with other TMs such as CA 125 in the management of ovarian cancers [33]. Different results have been reported for the CV _I of CEA in healthy and cancer patients. In studies on healthy subjects, Erden et al. [22], Sölétormos et al. [34] and Dittadi et al. [35] have reported CV _I estimates of 30.9, 9.3 and 8.4%, respectively. In study of Eden et al. the lack of outlier and homogeneity analysis and the small number of samples (four samples per subjects) might explain the large discrepancy. In cancer patients (in remission), such as breast cancer [35], surgically treated colon adenocarcinoma and non-small cell lung cancer [23], CV _I estimates have been reported at 19.3, 9.9 and 11.9%, respectively. We found lower CV _I estimates for CEA in all subjects and subgroups than all previous publications (Tables 1 and 2), but similar CV _I and CV _G estimates in fertile and postmenopausal women, in line with that reported by Tuxen et al. [36]. The marked individuality, as also reported by others [34], [37], supports the use of RCV instead of popRI for monitoring of CEA concentrations by serial measurements.

Smoking is one of the main reasons of lung cancer and also increases the CEA concentrations in healthy subjects [38]. In our study, we found higher concentrations of CEA in smoking subjects (Table 1), however, the difference was not significant. Furthermore, estimates of CV _I and CV _G were similar in smoking and non-smoking subjects.

Cytokeratin-19 fragment

Serum CYFRA 21-1 concentrations is used in the management of non-small cell lung cancers, particularly for squamous cell tumors [39]. BV for CYFRA 21-1 has previously not been studied in healthy subjects. Based on surgically treated colon adenocarcinoma patients Trape et al. [23] reported the CV _I and CV _G of CYFRA 21-1 as 22.5 and 38.1%, respectively, and similar estimates in lung cancer patients (in remission). We found a CV _I estimate (19.7% [18.79–20.7]), in line with these estimates, in our healthy study population, but a lower CV _G estimate (29.5% [25.3–35.0]). Similar to some of the other markers included in our study, the CV_I of CYFRA 21-1 was lower in postmenopausal women than in younger subjects (Table 1).

The CV_I estimate of CYFRA 21-1 was higher than those of the other TMs included in our study, thus resulting in higher RCV and a higher II. Both CEA and CYFRA 21-1 are used in the monitoring of lung cancer. Data of this study shows that CEA is associated with a higher individuality (II=0.11) and smaller RCV (RCV_increase=16.7%) than CYFRA 21-1. This indicates that smaller increases in CEA, than in CYFRA 21-1, might be of significance when monitoring patients. Additionally, unlike the other TMs, a marked asymmetry was evident in the RCV of CYFRA 21-1, where the RCV_increase was 57.9% and the RCV_decrease was −36.7%.

Cancer antigen 19-9

Serum concentration of CA 19-9 is mainly used in the management of pancreatic cancer, in addition to gastric and colorectal cancers. In the literature, the reported data of CV_I for CA 19-9 is controversial. Vestergaard et al. [40] recruited subjects with specific genotypes and reported the CV_I and CV_G as 15.8 and 102.2%, respectively. Erden et al. [22] recruited 38 subjects with four samples per subjects and reported CV_I and CV_G of 27.2 and 64.2%, respectively. Within the published data, for healthy subjects the lowest BV estimates were reported by Qi et al. [41]; CV_I and CV_G estimates of 6.8 and 49.6%, respectively. In our study, we found a CV_I estimate of 4.0 [3.7–4.2]), and the lowest results for II and RCV of CA 19-9 of all the TMs included in our study.

Pancreatic cancer is a highly aggressive cancer. In reflection of the low RCV found in our study, smaller increases in CA 19-9 concentrations should be considered as outside the expected, than for other TMs.

In this study, we observed higher mean concentration, but lower CV_I for CA 19-9 in postmenopausal women than in younger females (Table 1), which has previously not been reported.

Cancer antigen 15-3

CA 15-3 is used in the management of breast cancer [42], but is not specific, and elevated concentrations of CA 15-3 have been reported also in lung, ovary, colon, kidney and pancreatic cancers. Based on studies in healthy women, CV_I estimates for CA 15-3 have been reported by Söletormos et al. [43] and Dittadi et al. [35] as 6.2 and 6.0%, respectively. In addition, Dittadi et al. [35] and Hölzel et al. [44] reported CV_I of 17.3 and 11.2%, respectively, in brast cancer patients.

We found similar but lower CV _I for CA 15-3 at 4.4% (4.1–4.6) in our healthy population. A significantly lower CV _I for CA 15-3 was found in postmenopausal women compared with younger females, an observation that has previously not been published. Also CA 15-3 displays marked individuality (II=0.12) (Table 2), and the RCV should be used instead of popRI when monitoring patients.

Limitations of the study

The data of this study were obtained from healthy subjects and the derived BV of TMs represents the variation under physiological conditions, and may therefore differ from the BV of cancer patients. Additionally, subjects whose data were lower than LoQ were excluded and for some measurands, such as AFP, a relatively high number of results were excluded to fulfill statistical criteria (Supplemental Table 3). The exclusion of these data may lead to lower BV estimates and make our BV estimates less representative for the general population. This must be taken into account when using our data for RCV and other applications. Furthermore, our BV estimates were based on weekly samples collected over 10 consecutive weeks. These sampling intervals are likely different from the monitoring period applied in most cancer forms. Consequently, RCVs based on BV estimates derived from the healthy participants may need to be evaluated for fitness for the intended use in the monitoring of cancer patients.

Although the mean concentrations and CV_Is of most of TMs were found to be different in pre- and postmenopausal females, the number of postmenopausal females was only 10, and these estimates are therefore associated with large uncertainty. New studies with larger numbers of postmenopausal subjects are needed to verify our findings.