Role of serum cytokeratin 19 fragment (Cyfra 21.1) as a prognostic biomarker in patients with differentiated thyroid cancer

Differentiated thyroid cancers (DTC) account for up to 85% of thyroid cancers and generally display an excellent prognosis. However, in a minority of cases, DTC progress toward less differentiated phenotypes leading to distant metastases and even disease-related deaths. Circulating biomarkers are warranted to complement the gold standard DTC marker thyroglobulin (Tg) in identifying and monitoring such cases. We measured serum Tg and Cyfra 21.1 6 to 12 months after primary treatment in 473 DTC patients. A complete response of Tg was related to an excellent outcome in all cases. Among patients with incomplete Tg response Cyfra 21.1 levels <2.07 ng/mL were associated to favorable outcome while higher levels greatly increased the risk of disease related recurrences and deaths. Both markers retained independent prognostic values in multivariate analysis. In conclusion, Cyfra 21.1 is a tool available to independently predict survival of DTC patients not achieving excellent response after primary treatment.

Serum thyroglobulin and Cyfra 21.1 in detecting cancer relapse and predicting outcome over time. Serum Tg and Cyfra 21.1 levels were measured at first follow-up visit and compared with patients' outcome over time. At ROC curve analysis the most accurate Tg and Cyfra 21.1 cut-off values to discriminate NED and sREC patients were 0.35 ng/mL (Area Under the Curve 0.950) and 2.07 ng/mL (Area Under the Curve 0.860), respectively. Briefly, serum Tg and Cyfra 21.1 levels above these cut-off levels were found in 61 and 46 of 62 sREC cases and 76 and 4 of 376 NED cases, respectively. Sensitivity, specificity, accuracy, positive (PPV) and negative (NPV) predictive values were 98%, 80%, 82%, 44%, 100% for Tg and 74%, 99%, 98%, 92%, 96% for Cyfra 21.1. Positive Tg and Cyfra 21.1 values were found in 36/37 (97%) and 22/37 (59%%) patients with loco-regional recurrences and 25/25 (100%) and 24 Table 2. Comparison between NED and sREC patients' characteristics. ATA category ( Table 3). The model showed significant results (p < 0.001) with serum Tg and Cyfra 21.1 as the most accurate independent predictors of tumor relapse (p < 0.0001). Out of the other features pT, TNM stage, and age > 55 yr retained levels of significance higher than that the specific multivariate model, while pM1 and high risk classification were not correlated with disease recurrence.

Discussion
It was well demonstrated that DTC cancer cells express cytokeratin-19 8     serum Tg levels >0.35 ng/mL and high levels of Cyfra 21.1 confer an increased risk of distant spreading of the disease. This is likely due to a progressive development of more aggressive phenotype as the number of chromosomal abnormalities increases and disease-specific survival decreases in patients harboring distant DTC metastases compared to primary tumours and loco-regional recurrences [25][26][27][28] . On these basis, as the main result of the present study, Cyfra 21.1 levels exceeding 2.07 ng/mL significantly increased the risk of disease related recurrence and death over time while lower leves are associated to favorable outcome even in the presence of incomplete Tg response at early follow-up. Strengths of our study are the large number of patients, the homogeneous treatment and follow-up protocols with uniform diagnostic methods, risk-assessment and dynamic risk stratification criteria. Some potential limitations should be also addressed; first, while consecutive patients were enrolled and prospectively managed, the performance of Cyfra 21.1 measurement was evaluated by post-hoc analysis. However, Cyfra 21.1 levels were not taken into account in clinical management and relevant biases are unlikely. Also, the period of postoperative follow-up of our patients might not be considered as long enough; nevertheless, most relapses occur early during follow-up and recurrences have previously been reported to be rare in patients achieving excellent response after thyroid ablation. Then, no significant bias might be expected. Our study was not designed to test the diagnostic performance of Cyfra 21.1 in patients carrying positive TgAb. In these cases Tg is considered not reliable and TgAb disappearance kinetics after treatment are widely adopted as surrogate tumour marker 29 . Then, future comparisons of Cyfra 21.1 and TgAb kinetic in TgAb-positive DTC patients appear desirable.

Conclusions
In conclusion, our study proved that Cyfra 21.1 is a tool available to independently predict survival of DTC patients not achieving excellent response after primary treatment.

Patients, samples, study design and reference standard.
The medical records of all DTC patients diagnosed, treated and followed-up since January 2005 in participating centers were reviewed to extract a standardized dataset (i.e, demographic data, surgical and pathological report, follow-up reports including imaging and laboratory data). Then patients were enrolled for the present study if: 1. they underwent a (near-)total thyroidectomy (and, on indication, central neck lymph node dissection) with a histologically confirmed DTC; 2. serum Tg and Tg autoantibodies (TgAb) measurements were obtained 6 to 12 months after ablation and 3. residual serum samples >0.5 mL were available; 4. serum samples were stored at −80 °C and undergone less than 1 freeze/thaw cycle, 5. coexisting diseases potentially causing falsely positive Cyfra 21.1 measurements were excluded [i.e. any type of lung cancer and mesothelioma, breast cancer, bladder cancer and squamous cell cancers; liver diseases;chronic kidney diseases and cutaneous systemic diseases (i.e. pemphigus, psoriasis). Follow-up visits consisted in a clinical examination with neck US and basal serum Tg determination on a yearly basis. Additional examinations were performed on indication in selected cases. A disease status was assigned by attending physicians for each follow-up visit basing on the longitudinal review of the available clinical, imaging, biochemical and cytological/histological data. Patients were classified as alive with no evidence of disease (NED) if there was no clinical, imaging, or cytological/histological evidence of disease and their measured basal Tg levels were undetectable (i.e. below the functional sensitivity of the locally employed assay) or, if detectable, they were less than 1 ng/mL and decreased or remained unchanged over time 30 . Patients who did not fulfill these criteria were classified as alive with disease (i.e. recurrence, REC) and further stratified into alive with incomplete biochemical response (bREC) or alive with structural recurrence (sREC), respectively. The overall survival (OS) was calculated from the date of radioiodine ablation to the date of disease-related death. The disease-free survival (DFS) was calculated from the date of radioiodine ablation to the date of last follow-up (NED patients) or the date of relapse detection (bREC and sREC patients), respectively.
Laboratory. Once the study was completed frozen sera aliquots were centralized at Department of Clinical Chemistry and Laboratory Medicine, Ente Ospedaliero Cantonale, Bellinzona (Switzerland). Serum Tg, Tg autoantibodies (TgAb) and Cyfra 21.1 were measured by specific Cobas ® assays on fully automated Elecsys ® platform (Roche Diagnostics, Penzberg, Germany). The Tg assay is standardized 1:1 against BCR© 457 international standard (BCR, Brussels, Belgium) and display a functional sensitivity of 0.1 ng/mL; the TgAb assay is standardized against the WHO 65/93 international standard and display a functional sensitivity of 20 IU/mL; the Cyfra 21.1 assay displayed a limit of detection (LoD) of 0.10 ng/mL as reported by the manufacturer in assays' insert packages. Given kits were used to measure all samples and measurements were done strictly following manufacturer's instructions.
Statistical analysis. Mann-Whitney U, Wilcoxon test, and paired or unpaired t-test were used to analyze differences between paired or unpaired variables in two groups of patients. The predictive tests, i.e. sensitivity, specificity, positive (PPV) and negative (NPV) predictive value, and accuracy, were calculated according to Galen and Gambino. Agreement between continuous variables was assessed by Passing and Bablok regression analysis. Continuous variables were dichotomized by receiver operating characteristics (ROC) curve analysis using the maximum value of Youden's index (J) as the most accurate cut-off point. Aging was analyzed in several fashions: as a continuous variable (comparison of medians) or by using the specific cut-offs of 45 years or 55 years (Links TP et al. 28 ). DFS was estimated by using the Kaplan-Meier method and differences between curves were analyzed by log-rank or Mantel-Haenszel test and expressed as Hazard Radio (HR).The association degree of specific parameters with cancer relapse was assessed by Odds Ratio (OR). Parameters with significant association with cancer recurrence were included to carry out a model for a multivariate regression analysis. Statistical significance was set at p < 0.05. All statistical tests were performed by MedCalc Statistical Software, version 15.8 (MedCalc software bvba, Ostend; Belgium).

Ethics. The protocol was approved by the Ethics Committee of Canton Ticino (Switzerland) and Clinical
Research Committees of Ente Ospedaliero Cantonale (EOC), Bellinzona (Switzerland). All patients gave their informed consent before participating in the study. All experiments were performed in accordance with relevant guidelines and regulations. Data availability. The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.