gms | German Medical Science

The concept of using markers for the monitoring of tumour growth and efficiency of therapeutic intervention is not new. The first attempts at monitoring tumour growth and/or activity were with monoclonal immunoglobulinopathies using electrophoretic techniques and endocrine tumours using hormones as markers. The chemical analysis of hapten hormones such as adrenaline, noradrenaline, serotonin in urine or plasma and their metabolites, homovanillic acid, vanillylmandelic acid and 5-hydroxyindole acetic acid in urine were used as differential markers for suprarenal neoplasia (neuroblastoma, phaeochromocytoma, carcinoid) [1], [2], [3], [4]. Tartrate-inhibited acid phosphatase (ti-AcP) was used as an indicator of benign or malignant prostatic disorders [5].

The introduction of immunoassays allowed the measurement of peptide hormones, at first mainly for pituitary disorders (corticotropin (ACTH), growth hormone (GH, hGH), chorionic gonadotropin (CG, hCG), prolactin (PRL), thyrotropin (TSH)), steroid hormones (cortisol, aldosterone) and thyroid hormones (thyroxine (T₄) and triiodothyronine (T₃)) [6], [7], [8], [9], [10], [11], [12].

The search for more specific markers for non-endocrine tumours led to the development of carcinoembryonic antigen (CEA), alpha foetoprotein (AFP) and the "CA-immunoassays", originally from Centocor. The latter include CA 19-9 (gastrointestinal tumours), CA 125 (ovarian cancer), CA 15-3 (breast cancer) and CA 72-4 (TAG 72-4, gastric tumours) [13], [14], [15], [16], [17], [18].

The use of prostate-specific antigen (PSA) as a more specific tumour marker became interesting after the discovery that PSA was present in at least three forms in serum (free and complexed with either α₁-antichymotrypsin (α₁-ACT) and α₂-macroglobulin (α₂-MG), the latter of which is not detectable by immunoassay, as the PSA is surrounded by the α₂-MG and is therefore "invisible" to antibodies). Assays were developed for total, free and complexed PSA [19], [20], [21]. The ratio of "free" PSA to complexed PSA (= PSA-α₁-ACT) has been propagated as being able to differentiate between benign (hyperplasia) and malignant (neoplastic) prostatic disease [22], [23].

With the advent of the polymerase chain reaction (PCR) - methods, the accent of diagnosis is shifting toward the use of genetic markers. Examples here are the BRCA-1 and BRCA-2 genes which are associated with breast and ovarial cancer [24], [25], [26].

This publication deals with the review of tumour-marker external quality assessment (EQA) - schemes offered by Instand e.V. with a critical appraisal of performance and method comparability and their influence on the estimation of the diagnostic sensitivity and specificity.

The large variation in kit-performance, as far as numerical values are concerned, raises the question as to whether the "cut-off" values generally accepted but not validated for each method/kit, are of any use in terms of setting limits for diagnostic sensitivity and specificity.

The data used in this publication was obtained from the EQA schemes run by Instand e.V. during 2003 and 2004.

Two samples - lyophilised processed serum for the proteohormones, thyroglobulin and α-foetoprotein; liquid processed serum for the CA-markers, carcinoembryonic antigen and prostate-specific antigen - containig different analyte concentrations were sent for analysis with each survey.

Participants were requested to return the analysis data together with details concerning the measuring device and method/kit used.

Data was used from EQA-schemes for the following analytes: calcitonin (CT, hCT), thyroglobulin, α-foetoprotein (AFP), chorionic gonadotropin, prolactin, corticotropin, prostate-specific antigen (free, total and complexed: f-PSA; t-PSA; c-PSA), carcinoembryonic antigen (CEA), CA-125, CA 19-9, CA 72-4 and CA 15-3.

Any statistical comparisons used were based on the calculation of the measure of central tendency (mean, median) and dispersion (standard deviation and coefficient of variation (normally distributed data)) defined confidence intervals as centiles (non-Gaussian data distribution) and mainly included box and whisker plots and data distribution curves for graphic illustration of group-results.

Statistical comparisons between methods for a given analyte were not made because the study was designed to present data in a visual way, rather than to compare the inter-method performance.

The results are given separately for each analyte. Samples 1 and 2 were the same pair in Tables 1-3 [Tab. 1], [Tab. 2], [Tab. 3]. Sample 1 was pooled adult male serum, Sample 2 pooled adult female serum. Both samples were spiked and filtered (0.2 µm pore size).

Generally accepted cut-off levels for the "CA-Markers", CEA and total PSA have been given in the text [27], (see also [28]). These "decision-limits" have usually been made many years ago for assay-designs (e.g. competitive radioimmunoassay) no longer existing. They have not been validated for immunometric methods and should be interpreted with care.

There are no such "dogmatic-limits" for the other analytes used as tumour markers.

The problems with 1-39 ACTH are due more to stability of the analyte than to analytical problems, although some kits react more "allergically" than others.

Table 1 [Tab. 1] shows the results from three samples sent in different EQA-surveys in 2004.

At present, kits are using both the 1^st IS, which is an extract from human pitutitaries, as well as the recombinant 22 kD 2^nd IS. The comparison between kits is so good that all kits using the same reference standard can be grouped together. The values obtained using the recombinant material are substantially lower than those from the extracted human pituitaries.

Table 2 [Tab. 2] shows the results for growth hormone assays from three samples sent in different EQA-surveys (EQAS) in 2004.

Prolactin kits were calibrated exclusively with the 3^rd international standard. Table 3 [Tab. 3] shows the results for prolactin results from three samples distributed in 2004.

The kits were divided into those measuring only holo-hCG, holo-hCG + free β-chain and only free β-chain. Results were given as method specific in the first two cases and as a single group in the latter case.

The calibration of methods for holo-hCG was with the 3^rd IS (NIBSC 75/537), for free β-chain with the 1^st IRP for immunoassay (NIBSC 75/551). Accurate calibration of kits which determine both holo-hCG and free β-chain is impossible, as there are two unknown variables. Both reference preparations are given in IU/ampoule, but 1 IU holo-hCG is not equal to 1 IU free β-chain. The specificity of the antibodies for holo-hCG is well documented in Table 4 [Tab. 4]. The cross-reactivity of the free β-chain in the "mixed" assays can be seen by comparing the values for samples 1 and 2 in Tables 4 [Tab. 4] and 5 [Tab. 5]. The results from three EQA samples sent in 2004 are shown in Tables 4-6 [Tab. 4], [Tab. 5], [Tab. 6].

The assays for calcitonin must be able to recognise very high values as such, so that a high-dose hook effect must be eliminated from hCT-assays. The concentrations of calcitonin seen in the serum of unoperated medullary thyroid caricinoma (MCT) patients is in the ng/l to mg/l range. The physician requires accurate pre- and post-operative values in order to assess the success of the operation. Values "greater than" are of little or no use here.

The results from three EQA-samples distributed in 2004 are shown in Table 7 [Tab. 7].

The kits must measure precisely at the lowest concentration possible (≤2 µg/l) if they are to be used as tumour markers after total thyroidectomy, where recurrence of thyroglobulin in serum or plasma indicates metastatic growth. The precision of the kits shown in Table 8 [Tab. 8] is acceptable at all three levels controlled. Kit 5b measures precisely, but consistently higher than the other methods. This could prove a problem for the physician responsible for monitoring thyroid cancer if the laboratory changed methods - for example from manufacturer 3 to manufacturer 5 - and this information failed to reach the physician.

CEA is generally used in assessing colorectal cancers and liver metastases. The generally accepted cut-off values for CEA are between 1.5 and 5 µg/l, depending upon the kit used.

Table 9 [Tab. 9] shows the results from different manufacturers for CEA Kits in 2 samples dispatched in 2002. Samples were liquid and were commercially prepared from pooled patient sera specially for Instand e.V.

The median values for each method/method-group varied by up to 85%.

CA 15-3 is used in the in-vitro monitoring of breast cancer. The generally accepted cut-off value for CA 15-3 is 25 kU/l, independent of the kit used.

Table 10 [Tab. 10] shows the results for CA 15-3 for the same samples as in Table 9 [Tab. 9].

The inter-kit comparison of the distribution of results was comparable for both samples (variation of the method-median: Sample 1 - 44%; Sample 2 - 42%).

CA 19-9 is often used in conjunction with CEA in monitoring intestinal cancer. The generally accepted cut-off value for CA 19-9 is 37 kU/l.

Table 11 shows the results for CA 19-9 for the same samples as in Table 9 [Tab. 9]. Figure 1 [Fig. 1] shows a box and whisker plot for Sample 2 in Table 11 [Tab. 11].

The inter-kit comparison was not comparable for both samples (variation of the method-median: Sample 1 - 116%; Sample 2 - 140%). The method comparison results for CA 19-9 was worse than for CA 15-3.

CA 125 was found to be useful in monitoring ovarian cancer. The generally accepted cut-off level for CA 125 is 35 kU/l.

Table 12 [Tab. 12] shows the results for CA 125 for the same samples as in Table 9 [Tab. 9].

The inter-kit comparison was again not comparable for both samples (variation of the method-median: Sample 1 - 162%; Sample 2 - 108%). The inter-kit variation was similar to that for CA 19-9.

CA 72-4 (formerly known as Tennessee Antigen (TAG)) is used in the follow-up of gastric cancer. The generally accepted cut-off level for CA 72-4 (TAG 72-4) is 4 kU/l.

Table 13 [Tab. 13] shows the results for CA 72-4 (TAG 72-4) for the same samples as in Table 9 [Tab. 9].

The inter-kit comparison was comparable for both samples (variation of the method-median: Sample 1 - 445%; Sample 2 - 576%). The variation of results from different kits was the greatest here, although one kit (5b) gave much lower results that the other kits. Even another kit offered by the same manufacturer (5a) gave results between 6 and 7 times higher than kit 5b. This reflects the absence of standardisation in the determination of the "CA"- tumour markers as a whole, with perhaps the exception of CA 15-3.

Alpha foetoprotein has mainly been used for monitoring primary hepatic cancer. The accepted range for healthy non-pregnant individuals is ≤7 kIU/l (≈10 µg/l) (calibrated with NIBSC 72/225).

Table 14 [Tab. 14] shows the results for AFP for the same samples as in Table 9 [Tab. 9].

The inter-kit comparison of results is much better than for the "CA"-markers, the variation of the kit-medians for Sample 1 being 25% and for Sample 2, 22%. This may be due to the fact that all AFP kits have been calibrated against the WHO Standard (NIBSC 72/225), which is no longer available.

The various forms of circulating prostate-specific antigen (PSA) have been widely used - both singly and in combination - to monitor and differentiate between benign and malignant disorders of the prostate. The generally accepted cut-off for t-PSA is 4 µg/l. The ratios between t-PSA and f-PSA or c-PSA and f-PSA are strictly method-dependent. t-PSA assays may be calibrated with the first international standard (NIBSC 96/670), a mixture of c-PSA (90%) and f-PSA (10%).

Table 15 [Tab. 15] shows the results for t-PSA, Table 16 [Tab. 16] for f-PSA and Table 17 [Tab. 17] for c-PSA in the same samples as for Table 9 [Tab. 9].

The difference in specificity of the antibody-pairs for PSA determination can be seen in Tables 15 [Tab. 15] and 16 [Tab. 16]. The material used for spiking was from seminal fluid, known to be mainly composed of free PSA. Whereas kit 5b differentiated well between "total" and "free" PSA, kit 6a measured more free PSA than total PSA! The results in Table 17 [Tab. 17] shows the amount of PSA-α₁-antichymotrypsin (PSA-ACT) in each sample, which is usually much more abundant than free PSA and reflects the non-physiological nature of both samples, due to the reason stated above.

Figures 2 [Fig. 2] and 3 [Fig. 3] show the results from Tables 15-17 [Tab. 15], [Tab. 16], [Tab. 17] in the form of a box-and whisker plot.

There were large variations in the median values from both total and free PSA kits as can easily be seen in Figures 2 [Fig. 2] and 3 [Fig. 3].

The results from the EQA-surveys show that the standardisation of assays used as tumour markers is in many cases far from being optimised. The data presented show that at the present time the hormone assays used for tumour monitoring are on the whole more precise and give rise to more comparable results than the less-specific mucin-carbohydrate "CA-markers" and the various circulating forms of PSA.

The problems of tumour marker measurement have both a physiological and methodological component. For example the determination of CA 19-9, a marker related to the Lewis-antigens [29], [30], [31] is not produced in patients who are both Lewis a and Lewis b negative [31]. This means, that such patients can have large intestinal tumours, which are "negative" for CA 19-9 in serum. Many tumour markers are influenced by renal function, so that dialysis patients and those with impaired renal function may have different (mostly elevated) concentrations when compared with patients with normal renal function [32], [33]. Others, such as AFP, are increased in pregnancy.

There are very few "specific" tumour markers. Examples are calcitonin (CT) in medullary thyroid cancer (MTC) and thyroglobulin (Tg) in thyroidectomised patients, although the latter may be masked by the presence of circulating antibodies to Tg. Elevated levels of chorionic gonadotropin (CG) is relatively specific in males as a marker of germinal-cell cancer.

The commercial interest in tumour-markers cannot be ignored, both from the side of kit-producers as well as from kit-users. The clinical use is far more restricted, many "tumour markers" not being able to fulfil their specifications with regard to both analytical and diagnostic sensitivity and specificity. Clinical decisions made purely on levels of tumour markers - with the exception of perhaps CT, Tg and CG in males - must be seen as irresponsible, especially in a decentralised health system with the free choice of analytical laboratory and methods used.

The EQA results clearly show that the comparability of results and continuity in monitoring patient progress is only possible when using the same method with the same components over a long period of time. The danger of "rationalisation" and "cost-saving" (= buying the cheapest kit offered) practiced by many administrators, coupled with the ignorance of the medical staff on the methodological problems mentioned above, further limits the quality of data obtained from analysing "tumour markers".

The effect of numerical values can be visualised in comparing the diagnostic sensitivity (the correct prediction of a positive (=tumour present) result) and diagnostic specificity (the correct prediction of a negative (=tumour not present) result). Figures 4 [Fig. 4] and 5 [Fig. 5] show the ideal case for a tumour marker, where healthy (red) and sick (blue) patients are clearly separated from each other (diagnostic sensitivity and specificity 100%). Figures 6 [Fig. 6] and 7 [Fig. 7] show the more common situation, i.e. where both groups overlap. In Figure 7 [Fig. 7], there are three ways of setting a decision point. At point A we have the lowest analyte concentration where all sick patients are correctly allocated (diagnostic sensitivity 100%). A number of healthy patients have however concentrations higher than this point (diagnostic specificity <100%) and would be classified as sick. At point B we have the converse situation - the highest concentration where all healthy individuals are correctly classified (diagnostic specificity 100%), but where some sick individuals (those under the blue curve to the left of B) are to be found (diagnostic sensitivity <100%) and are classified wrongly as being healthy. Point C represents the best compromise where the degree of false classification is minimised, but where sick and healthy individuals in the border region may be wrongly classified (diagnostic sensitivity and specificity in this case <100%).

In the case of t-PSA, if we leave the points A, B and C in Figure 7 [Fig. 7] where they are and measure patient sera with the methods in Table 15, it becomes obvious that the red and blue curves will be moved - either to the right or to the left - thus changing the diagnostic sensitivity and specificity according to the method/kit used. The points A,B and C must be evaluated for each kit.

Figure 8 [Fig. 8] shows the distribution of data of the t-PSA Sample 2 in Table 15. Both the mean values and precision of measurement (seen by the different degrees of kurtosis) make it clear that the variability in measurement - here taking t-PSA as the example - must lead to different kit-specific cut-off values, thus nullifying a static point of decision at 4 µg/l. Both extremes are shown by the points A (Kit 11a) and B (Kits 8a, 10a and 15c).

Standardisation of methods - not with international antigen preparations but with defined antigenic epitopes recognised by defined monoclonal antibodies, as in the case of the tumour marker Cyfra 21-1 [34], can lead to more comparability between manufacturers and to an improvement in the interpretation of results by clinical staff, thus improving the benefits to patients subjected to such analyses.

Method-dependent differences in results, especially in the case of PSA, make it imperative that method-specific reference ranges and "cut-off" values must be established for each analyte. In the case of PSA, where ratios between free and total PSA or complexed and total PSA are used in differentiating between benign and malignant disease, kits from the same manufacturer must be used for free and total/complexed PSA, if the ratios given by the manufacturer are to have any clinical use. General reference-ranges and ratios often given in text books are of no use, due to the individual nature of results from different manufacturers, although some authors [35] point out the importance of using kit-specific reference ranges. The importance of this is shown in Tables 15 and 16 above, even though the samples were not physiological (concentration of free PSA much higher than complexed PSA). Even kits from the same manufacturer, but developed for different instruments with different measuring techniques may give rise to statistically different results from the same sample - here seen in kits 41 and 42 for t-PSA.

The use of generalised figures for the diagnostic sensitivity and specificity of tumour-marker kits must be discouraged, as seen in the results from this study (as an example, see Figure 1 [Fig. 1]). Each laboratory must therefore estimate its own diagnostic sensitivity and specificity for all tumour-markers from data obtained from the kit used and the group(s) of patients studied. The relatively low diagnostic sensitivity and/or specificity shown by many tumour-markers reflects their inability to be used for screening purposes. Despite this, medical staff still use tumour markers for patient screening - especially PSA - although no prospective randomised trial has been performed to validate or reject the use of PSA in screening for prostate cancer [36]. Recently, the value of PSA in diagnosis and monitoring of prostate cancer has again been questioned [37]. Even in review articles, the cut-off for t-PSA decision taking is set at 4 µg/l, despite evidence for method dependent reference ranges cited in the same article [38]!

Even in the case of post-operative follow up, it is possible, that when metastasis occurs, "tumour-specific-antigens" may be produced which are structurally different from those in the primary tumour, thus rendering the further use of the original marker for follow-up useless.

The future of detection and post-operative follow-up of tumours may very well lie in the developing field of gene-expression chips [39]. That synthesis of tumour markers can change during the development of neoplasms has been shown in immunohistochemical studies [40].

Even the use of genetical markers can at present only predict a predisposition for - in this case - the development of a certain tumour. The time of appearance of the lesion cannot be predicted from the presence of or mutations in certain genes - for example BRCA-1 [24], which usually lead to the development of breast and/or ovarian cancer before the 5th decade of life.

Classical methods of detection and control, such as x-ray, sonography, PET, MRT and computer tomography, will remain the "golden" standards of tumour detection and control of tumour progression/remission for many years to come, although "gene-chips" will play an increasingly important role in detection of those at risk, as well as post-operative follow-up of cancer patients [41], [42], [43], [44].

To conclude, the majority of analytes used as tumour markers are unsuitable for screening purposes. Exceptions are the measurement of calcitonin, chorionic gonadotropin and thyroglobulin when used in the situations stated above. The use of the other analytes reported in this article may be of use in longitudinal follow-up pre- and post-operatively, with the restriction that the method used for a given analyte must always be the same, at least until both an acceptable standardisation of methods/results and comparability of numerical values (concentrations) is achieved. The latter has not been realised and is not to be expected in the near future, at least for the majority of analytes. Until then, defined individual cut-off values must be established for each tumour marker kit on the market.