Prognostic value of serum cytokeratin 19 fragments (Cyfra 21-1) in patients with non-small cell lung cancer

The role of serum CYFRA 21-1 level in patients with non-small cell lung cancer (NSCLC) remains to be defined. To re-evaluate the impact of serum CYFRA 21-1 in NSCLC survival, we performed this meta-analysis. Databases were searched to identify relevant studies reported after the publication of a meta-analysis in 2004. Totally, 31 studies with 6394 patients were included in this meta-analysis. The pooled Hazard ratios (HRs) indicated that high CYFRA 21-1 level was associated with poor prognosis on overall survival (OS) in patients with NSCLC (HR = 1.60; 95%CI = 1.36–1.89; P < 0.001). The pooled HRs were 2.18 (95%CI = 1.70, 2.80; P = 0.347) for patients at stage I–IIIA and 1.47 (95%CI = 1.02, 2.11; P < 0.001) for stage IIIB–IV. When stratified by surgical intervention, pooled HRs were 1.94 (95%CI = 1.42–2.67; P < 0.001) for studies with surgery and 1.24 (95%CI = 0.79–1.95; P < 0.001) for studies without surgery. Significant associations were also found in the patients treated with EGFR-TKIs (HR = 1.83; 95%CI = 1.31–2.58; P = 0.011) and platinum-based regimen (HR = 1.53; 95%CI = 1.18–1.99; P = 0.001). Meta-analysis of CYFRA 21-1 related to PFS was performed and pooled HR was 1.41 (95%CI = 1.19–1.69; P < 0.001). Our results indicate that high level of serum CYFRA 21-1 is a negative prognostic indicator of patients with NSCLC.

poor prognosis for NSCLC patients whatever the planned treatment. However, the result of the patients having undergone surgery only showed a trend of statistical significance. Furthermore, this metaanalysis had the following limitations: relatively short follow-ups, including a small number of accrued patients, and not providing detailed treatments for patients who did not undergo surgery.
Subsequently, many studies investigating the role of serum CYFRA 21-1 with more updated therapeutic strategies for NSCLC and larger numbers of accrued patients have been performed  . However, these studies yielded conflicting results. Therefore, we conducted an updated meta-analysis using data from these studies to reappraise the effect of serum CYFRA 21-1 on the prognosis in patients with NSCLC.

Results
Characteristics of eligible studies. After deleting the duplications, a total of 570 potentially relevant publications were collected after initial search. Among these, 313 articles related to the prognosis of lung cancer were reserved. Then 89 studies primarily researching the CYFRA 21-1 level were selected in full text for further screening after reviewing the abstracts. 49 studies were excluded for only investigating the clinical characteristics rather than specific OS, involving small cell lung cancer (SCLC), or other out of the scope according to the inclusion and exclusion criteria, consequently leaving 40 available for further review. After carefully reading, 31 studies, with a total number of 6394 patients were identified in this meta-analysis ( Figure 1). Some studies reported two endpoints, and they were analyzed separately. The main characteristics of the evaluable studies were listed in Table 1.
In the present meta-analysis, statistical calculations for OS were performed in 31 studies. Sample size of studies ranged from 48 to 1202. 17 studies were performed in Asian, 13 studies were in Caucasian, and one study was in mixed populations. 6 studied NSCLC patients at stage I-IIIA, 9 studied patients at stage IIIB-I, and 15 studied the wide range of stage I-IV. 12 were conducted prospectively, 17 were performed retrospectively, and 2 were not available. The numbers of studies reported all of the patients with surgery and without surgery were 9 and 9, respectively. 7 studies including 1061 patients were involved in PFS calculation. The detailed information of them was described in Table 1.
Serum CYFRA 21-1 was dichotomized into high and low levels, and different cut-off value was selected in each study. Most of the studies utilized the manufacturer's instructions, some applied the median or mean levels as cut-off values, and the remaining studied defined the value by themselves or a ROC curve.
We also performed sensitivity analyses by repeatedly deleting the single studies each time from pooled analysis to examine the stability and reliability of meta-analysis results. In our analysis, the omission of individual studies did not materially alter the results because the recalculated ORs and 95%CIs were not quantitatively changed, suggesting that the results were robust and convincing ( Supplementary Fig. S2, S3).
Publication bias. Publication bias was assessed by Begg's funnel plot and Egger's test. Publication bias was detected (P 5 0.144 for Begg's   (Figure 6). The recalculated ORs for OS did not change significantly (HR 5 1.27; 95% CI 5 1.075-1.493; P , 0.001), indicating the stability of the results. However, significant publication bias was not discovered among the studies regarding PFS (P 5 0.368 for Begg's test).

Discussion
To the best of our knowledge, we demonstrated, for the first time, that high serum CYFRA 21-1 level was an indicator of poor prognosis for NSCLC using updated data. Notably, our meta-analysis included three times more patients than the previously reported one 15 , and the studies employed more updated therapeutic regimen and patients with longer follow-ups. As a result, we were able to show a more plausible result. In this meta-analysis, we identified 31 eligible studies including 6394 patients and concluded that high serum CYFRA 21-1 level was a poor prognostic indicator for OS and PFS. For different pathological TNM stage, high serum CYFRA 21-1 level was associated with poor clinical outcome of NSCLC patients. Thus, high level of serum CYFRA 21-1 might be a negative prognostic biomarker in NSCLC patients with resected tumor or with unresectable disease. This link was observed in surgery subgroups. Notably, we verified the poor prognostic role of high serum CYFRA 21-1 level in the patients who had undergone surgery (HR 5 1.94; 95%CI 5 1.42-2.67; P , 0.001), while the previous meta-analysis only indicated a trend towards statistical significance (HR 5 1.41, 95% CI 5 0.99-2.03, P 5 0.055). However, for patients who did not undergo surgery, the HR was 1.24 (95%CI 5 0.79-1.95; P , 0.001) while the previous meta-analysis showed high serum CYFRA 21-1 level predicted poor survival (HR 5 1.78; 95%CI 5 1.54-2.07, P , 0.001) in the first year of follow-up 15 .
We also found that high serum CYFRA 21-1 level was a poor prognostic factor for patients treated with EGFR-TKIs or plat-inum-based regimen. Platinum-based chemotherapy could improve survival and has become the standard chemotherapy for years 48,49 , however, the efficacy of platinum-based chemotherapy varies among individuals, with a response rate of 26%-60% 50 . Various tumor markers have been studied in terms of their prognostic or predictive roles in NSCLC patients treated with platinum-based chemotherapy 8,51-54 . However, to date, no serum maker is currently recommended for routine clinical practice in prognosis of NSCLC patients treated with platinum-based chemotherapy. Based on our findings, serum CYFRA 21-1 might be a promising biomarker. EGFR mutations have been reported to be the most important prediction factor for patients treated with EGFR-TKIs 55 . Unfortunately, it is not feasible to obtain an adequate EGFR mutational analysis. Therefore, it is important to identify easily acquired clinical parameters that can serve as surrogate markers for EGFR mutants. Our results suggested that serum CYFRA 21-1 might play such a role in the prediction and prognosis of advanced NSCLC treated with EGFR-TKIs. Notably, this was the firstly pooled analysis to verify the prognostic role of serum CYFRA 21-1 in the NSCLC patients treated with platinum-based chemotherapy or EGFR-TKIs. In consideration of the relatively small sample size, large scale prospective studies are needed to further confirm the results.
The sub-group analyses by ethnicity and study design yielded the same results, further indicating the poor prognosis role of high serum CYFRA 21-1 level in NSCLC.
Attention must be paid to the relatively large heterogeneity among trials. Meta-regression was conducted by ethnicity, surgical intervention, chemotherapy, detective method, cutoff, study design, and sample size. However, none of them was found to be the sources of heterogeneity. In fact, many other factors may also be the potential source of heterogeneity, such as gender, age, and life styles of patients and so on. Due to lack of detailed data, we had to give up performing a meta-regression utilizing these variables. Publication bias is significant threat to the reliability of the results of a meta-analysis. Unfortunately, we did find publication bias by Begg's funnel plot and Egger's test. Then a trim and fill method was adopted to recal- culate the adjusted ORs and the results did not change significantly, indicating the stability of the statistic analyses 47 .
Several limitations have to be noted in relation to this meta-analysis. First, all our analyses were based on abstracted data and not on individual patient data (IPD). It was commonly acknowledged that an IPD-based meta-analysis would reproduce more reliable estimation compared with one based on abstracted data 56 , while an IPDbased meta-analysis is a time-consuming effort, especially when some studies without high quality 57 . Second, several HRs were obtained based on the survival curves, which might have biased our results. Third, cutoff values were different among the studies. Finally, the publication bias and heterogeneity in the meta-analysis may partly lessen its statistical power. However, these problems are almost inevitable in a meta-analysis.
In conclusion, our meta-analysis, based on an updated data, found a significant prognostic value of serum CYFRA 21-1. High serum CYFRA 21-1 level is correlated with poor OS and PFS in NSCLC, which might provide a simple and practical method to predict the outcome of NSCLC patients. Interestingly, serum CYFRA 21-1 is also related to the OS in the NSCLC patients treated with EGFR-TKIs or platinum-based regimen. But the sample size is relatively small, and more investigations are needed to identify its prognostic role in this part of NSCLC patients.

Methods
Search strategy. To update the data, we intended to exclude the studies accrued in the previous meta-analysis published in 2004 15 . Therefore, only those studies published after January 1, 2002, were eligible. We conducted a computerized literature search of Embase, Web of Science, PubMed and China National Knowledge Infrastructure (CNKI) to identify all the studies that studied the association of serum CYFRA 21-1 lever and lung cancer. The combination of the following key words were used as search terms separately and in combination: ''CYFRA 21-1'', ''cytokeratin 19'', ''non-small cell lung cancer'', ''NSCLC'', ''prognosis'', ''survival'', and ''outcome''. Last search was updated on January 08, 2015. The references of all publications and reviews were manually searched to identify potentially relevant studies.
Inclusion and exclusion criteria. To be eligible for inclusion in this meta-analysis, studies had to meet the following criteria: (1) trials had to deal with NSCLC; (2) investigating the association between serum CYFRA 21-1 level and prognosis; (3) CYFRA 21-1 was clarified as ''high'' and ''low'' value; (4) published as a full paper in English for data extraction; and (5) including enough blood sample. To avoid duplication of data, only the most complete and recent study was included. Titles and abstracts of searching records were screened and full text papers were further evaluated to confirm the eligibility. According to the inclusion criteria, two reviewers Data extraction. Two reviewers (Xu and Qiu) determined study eligibility independently in order to avoid selection bias in the data abstraction process. The following information was culled from each study: the first author, publication year, country where the research was performed, ethnicity, number of patients, histology, stage, detective method, cutoff, study design, and survival data. The primary endpoint was OS. The studies utilizing PFS were also analyzed.
Statistical analysis. To evaluate the predictive ability of high serum CYFRA 21-1 level on survival of NSCLC, the hazard ratios (HRs) and their 95% confidence intervals (CIs) of OS and PFS were extracted from eligible studies. If these data were not available, the HRs and CI were calculated according to Tierney' methods 58 . The heterogeneity among studies was evaluated using the Cochran Q and I 2 test: for the Q test, a P , 0.1 was considered statistically significant 59 . For I 2 test, a value .50% was considered a severe heterogeneity 60 , then the random-effects model (based on DerSimonian-Laird method) was used 61 ; otherwise, the fixed-effects model (based on Mantel-Haenszel method) was applied. Meta-regression was performed to assess the sources of heterogeneity by ethnicity, surgical intervention, chemotherapy, detective method, cutoff, study design, and sample size (studies with less 200 participants were categorized as ''small'', and studies with more than 200 participants were categorized as ''large''). Between study, variance Tau-squared (t2) value was used to evaluate the degree of heterogeneity and describe the extent of heterogeneity explained 62 . Sensitivity analysis was performed to examine the stability of the pooled results.