| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Briefs |
1
Department of Clinical Biochemistry, Hillerød Hospital, Helsevej 2, DK 3400 Hillerød, Denmark
2
Department of Clinical Biochemistry, Odense University Hospital, 5000 Odense, Denmark
3
Department of Oncology, Herlev Hospital, University of Copenhagen, 2730 Copenhagen, Denmark
aauthor for correspondence: fax 45-48-24-00-67, e-mail geso{at}fa.dk
Clinical trials have suggested that the serologic cytokeratin tumor marker tissue polypeptide antigen (TPA) may be important as a monitor of metastatic breast cancer (1)(2)(3). The usefulness of this marker depends on the ability to identify, predict, and exclude tumor growth before the changes are detectable by imaging techniques and routine biochemistry (4)(5)(6). It is, however, difficult to compare results because the investigated patient populations have been heterogeneous. It remains unknown how TPA assessment criteria perform in situations with different (a) time intervals between measurements, (b) numbers of measurements, (c) rates of increase during progression, and (d) analytical quality. The time and expense required to investigate these issues in new studies will be considerable. Recently, computer-simulated marker data have been suggested to compare the diagnostic accuracy of assessment criteria used to interpret longitudinal cancer antigen (CA) 15.3 and carcinoembryonic antigen (CEA) data (7)(8). The computer-based approach was useful because the descriptive variables described above could be standardized and changed individually according to the simulated clinical and laboratory situation. The procedure-identified criteria must have a high ability to detect early CA 15.3 and CEA progression and at the same time maintain robustness against false-positive signals. Because TPA measurements among patients are frequently performed in combination with CA 15.3 and CEA, we consider it relevant to investigate whether computer-simulated data can be used to identify reliable criteria to interpret sequential TPA concentrations. However, because the typical background variability of TPA and the rate of TPA increase during progression are different as compared with CA 15.3 and CEA (7), we generated new data sets in the present simulation procedure to test our hypothesis in a clinically relevant environment. Progressive TPA concentrations that were clinically relevant were combined with representative TPA values for background steady-state variation in a computer-simulated model. Six criteria for assessment of longitudinal tumor marker data were obtained from the literature and programmed into a software package as computer functions (8).
The true-positive signals denoted TPA progression among events with continuously increasing concentration. The frequency of true-positive TPA progression signals was determined by the rate of marker increase (Table 1
). Events with slow rate constants were detected more easily if the required difference was small and the cutoff value was low. At fast and medium rate constants the increments were so large that any of the required changes were reached early. The time interval to true-positive signals of TPA progression also depended on the background variation because the rate of increase of progressing concentrations was stimulated by upward fluctuations and delayed by downward fluctuations. At slow rate constants, the increments delayed by increasing downward background fluctuations easily satisfied progression criteria based on a low cutoff, i.e., Fig. 1, A and B
, whereas criteria based on a high cutoff were met later (Fig. 1, E and F
). Marker kinetics with fast and medium rate constants were less influenced by background fluctuations.
|
|
The false-positive signals denoted TPA progression among events where TPA fluctuated in a steady state without any rate of increase (0.0000). The frequency and time point of the false-positive TPA signals were determined by the magnitude of the normal background fluctuations because larger fluctuations promoted the required critical difference. The criteria in Fig. 1, E and F
, did not give any false-positive progression signals when the concentrations fluctuated in the middle of the normal range. Their robustness against false-positive signals of progression was therefore further investigated in situations where TPA concentrations fluctuated closely below the applied cutoff value. Indeed, both criteria provided false progression signals if steady-state concentrations fluctuated high within the normal range. However, the criterion in Fig. 1F
gave false-positive signals at unusually large clinical fluctuations, whereas the criterion in Fig. 1E
also gave false signals at small fluctuations. The criterion in Fig. 1F
was robust against false-positive progression signals, owing to a higher cutoff value. In reviewing the data provided in Table 1
, we believe it is clear that the criterion by Sölétormos et al. (3) (Fig. 1F
), which adjusted the cutoff value and the critical difference to the higher background variation of TPA, performed better than criteria by other authors where this issue was left unconsidered (4)(5)(6) (Table 1
). However, the importance of adjusting the algorithm to the individual marker also applied to the criteria elaborated by Sölétormos et al. (8) because the criteria that sufficed for CA 15.3-CEA (Fig. 1, B and C
) gave false positives when applied to TPA data (Table 1
). Conversely, the criteria that gave delayed CA 15.3-CEA information was more suitable for TPA (Fig. 1F
; Table 1
) (8).
In conclusion, the elaborated computer-simulation models have considerable potential because criteria can be compared in a variety of simulated conditions of steady-state and progressive disease. Varying the background variation and the tumor growth according to realistic patient data, as well as the cutoff value, enabled identification of criteria that were robust against false-positive signals of TPA progression. When data that define steady-state variability and rates of increase become available, the model systems should be used to generate assessment criteria for other markers and malignancies and compare and optimize diagnostic performance of the criteria. However, it is important to emphasize that computer-simulation studies cannot replace clinical tumor-marker trials. Computer-based comparison of assessment criteria is a supplement to clinical studies and relevant only if these studies have provided reliable estimates of basic performance characteristics. However, by simulating a large variety of new trial conditions, the robustness of a criterion can quickly be investigated at low costs, and a determination can be made as to whether the criterion is of limited use or should be implemented in routine clinical practice for patient monitoring.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
significant denotes that the increment exceeded what could be explained by the analytical and biologic components of variation. *, recommended cutoff; **, recommended cutoff x 3.8; ***, recommended cutoff x 2. Further details may be obtained from the data supplement posted at Clinical Chemistry Online (