Purpose: The purpose of this study was to identify and standardize optimal decision criteria for maximizing the effectiveness of tumor markers in clinical use during the follow-up of patients operated on for breast cancer.
Materials and methods: The study was prospectively performed on 859 patients enrolled in 10 institutions. A total of 13,337 determinations of CEA and 14,330 determinations of CA15.3 were available. The median number of samples per patient was 16 for CEA and 17 for CA15.3. The median follow-up was 7 years. Receiver-operating characteristic analysis was used to evaluate the ability of CEA and CA15.3 to discriminate relapses from patients who had no evidence of disease. The diagnostic performances of the two markers were evaluated using decision criteria based on both dichotomic cut-off points and dynamic variations among serial samples.
Results: We selected decision levels corresponding to preset levels of 90% and 99% specificity. Patients with CEA and/or CA15.3 levels above the cut-off values were considered positive only if a 1.5-fold increase occurred among the last sample and the mean of the first three samples. According to the different cut-offs used, specificity ranged from 94% to 99% and sensitivity from 48% to 63%. We calculated predictive values using the prevalence expected with reference to the stage of primary tumor and the length of follow-up. Positive predictive values ranged from 1.6% to 93.7%, and negative predictive values from 88.9% to 100%, according to the clinical scenarios and the decision criteria used. The choice of the decision criteria significantly affected positive predictive values within each patient subset. Differences related to time from surgery were still remarkable for every decision criteria (i.e., positive predictive values ranged from 36.6% to 2.8% in node-negative patients according to the year of observation, although the same cut-off point was used).
Discussion: The results of the present prospective study show that different decision criteria may provide different diagnostic performances for the same tumor marker and in the same patient. Therefore, we suggest that different decision criteria be settled and used according to the clinical goals.