In June 2013 we searched MEDLINE (1980 to May 2013), EMBASE (1980 to May 2013), Web of Science (1990 to May 2013) and Cochrane databases. Although no language restrictions were imposed initially, for the full-text review and final analysis only English language articles were included. Additional articles were searched using the “Related articles” function in PubMed and by manually searching reference lists of identified articles and review articles. The following search terms were used: “pancreatic carcinoma” or “pancreatic cancer” and “chronic pancreatitis” and “carbohydrate antigen 19-9” and “diagnosis” or “sensitivity” or “specificity”. We contacted experts in the field to ask about studies that we may have missed in the databases. Conference abstracts and letters to the editor were excluded because of the limited data they contained.

A study was included when it provided both the sensitivity (true-positive rate) and specificity (true-negative rate) of using serum CA19-9 levels to diagnose pancreatic carcinoma or chronic pancreatitis in patients of any age. Studies were also included if they reported CA19-9 values in a scatter plot format that allowed patient-level data to be extracted. Studies had to involve at least 10 patients with pancreatic carcinoma or chronic pancreatitis in order to reduce selection bias due to a small number of participants. Patients had to be diagnosed with pancreatic carcinoma based on cytology and/or histology of pancreatic tissue, or diagnosed with chronic pancreatitis based on clinical information alone or in combination with histopathological resection, radiology (endoscopic retrograde cholangiopancreatography and computed tomography) and/or endoscopic ultrasonography. Two reviewers (Su SB and Jiang HX) independently determined study eligibility, and disagreements were resolved by consensus.

These same two reviewers independently confirmed the eligibility of the final set of studies and extracted the following data: first author, publication year, participant characteristics, assay methods, sensitivity and specificity data, cut-off values, and methodological quality. Serum CA19-9 values provided in scatter plots were extracted by placing scalar grids over the plots. A receiver operating characteristic (ROC) curve was calculated for each study.

To enable us to assess the methodological quality of the included studies, we extracted data on the following study design characteristics: (1) cross-sectional or case-control design; (2) consecutive or random sampling of patients; (3) blinded (single or double) or non-blinded interpretation of experimental and reference measurements; and (4) prospective or retrospective data collection. Su SB and Jiang HX independently assessed the methodological quality of studies using the Standards for Reporting Diagnostic Accuracy (STARD) guidelines[11] (maximum score of 25) and the Quality Assessment for Studies of Diagnostic Accuracy (QUADAS) guidelines[12] (maximum score of 14). Average inter-rater agreement on the methodological quality checklists was 0.96. If primary studies did not report information needed to assess methodological quality, we contacted the authors in an effort to obtain the data. If the authors did not respond, we changed the response for the relevant items from “not reported” to “no” on the assessment instruments.

Standard methods recommended for meta-analyses of diagnostic test evaluations were used[13]. Analyses were performed using Meta-DiSc for Windows (XI Cochrane Colloquium; Barcelona, Spain) and Stata 12.0 (Stata Corporation, College Station, TX, United States). The following measures of test accuracy were analyzed for each study: sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR) and diagnostic odds ratio (DOR). A summary ROC (SROC) curve[14] was generated for each study based on a single test threshold for sensitivity and specificity[13,15]. A random-effects model was adopted to calculate the average sensitivity, specificity, and other measures across studies[16,17].

To assess the effects of STARD and QUADAS scores on the diagnostic power of CA19-9, we included them as covariates in univariate, inverse variance-weighted meta-regression. We also analyzed the effects of other covariates on DOR, such as cross-sectional design, consecutive or random sampling of patients, single- or double-blinded interpretation of experimental and reference measurements, and prospective or retrospective data collection. The relative DOR (RDOR) was calculated to analyze the change in diagnostic precision in each study per unit increase in the covariate[18,19]. P < 0.05 was considered to show statistical significance.

The heterogeneity, or variability, across studies was assessed for statistical significance using the χ² and Fisher exact tests. Publication bias can pose problems for meta-analyses of diagnostic studies, therefore, we tested for the potential presence of this bias using funnel plots and the Egger test[20].

We identified 345 citations via electronic searches, and 106 were retrieved for detailed analysis (Figure 1). Of these, 47 studies were excluded for failing to satisfy the inclusion criteria, and another 17 were excluded because they failed to provide sufficient information for meta-analysis. Five studies were duplicate publications. Two articles were meta-analyses, and one was excluded for involving fewer than 10 participants. In the end, 34 publications were included in the analysis[21-54], involving 3125 patients with pancreatic carcinoma and 2061 patients with chronic pancreatitis. The average sample size of the studies was 153 patients (range: 24-941). Table 1 summarizes the clinical characteristics of participants in each study; the numbers of true-positive, false-positive, false- negative and true-negative results; and STARD and QUADAS scores.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Flowchart of study selection.

Of the 34 studies in the meta-analysis, 30 had STARD scores ≥ 13, and 29 had QUADAS scores ≥ 10. All studies collected data from consecutive patients using a prospective design. No study reported interpretation of CA19-9 measurements in which analysts were blinded to the corresponding reference measurements (Table 2).

As shown in Figure 2, a Forest plot of serum CA19-9 levels in all 34 included studies showed that the sensitivity of this biomarker to differentiate between pancreatic carcinoma and chronic pancreatitis ranged from 0.44 to 0.96 [mean: 0.81, 95%CI: 0.80-0.83; χ² = 77.23, P < 0.001), while the specificity ranged from 0.50 to 1.0 (mean: 0.81, 95%CI: 0.79-0.82; χ² = 111.98, P < 0.001). The PLR was 4.08 (95%CI: 3.39-4.91; χ² = 113.62, P < 0.001), NLR was 0.24 (95%CI: 0.21-0.28; χ² = 86.13, P < 0.001) and DOR was 19.31 (95%CI: 14.4-25.9; χ² = 94.02, P < 0.001). These χ² values and associated P-values indicate significant heterogeneity among studies.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Forest plot showing sensitivity and specificity of carbohydrate antigen 19-9 in the diagnosis of pancreatic carcinoma. The point estimates of sensitivity and specificity from each study are shown as solid circles. Horizontal error bars indicate 95%CI. Numbers between the plots refer to references. Pooled estimates for the serum carbohydrate antigen 19-9 assay were 0.81 for sensitivity (95%CI: 0.80-0.83) and 0.81 for specificity (95%CI: 0.79-0.82).

These measures of differential diagnostic power varied with different CA19-9 assays and cut-off values used to define CA19-9 levels as elevated or normal (Table 3). Data from the 11 studies that relied on the enzyme-linked immunosorbent assay (ELISA) method, involving 1396 patients, gave a sensitivity of 0.83 and specificity of 0.79. Data from the 17 studies using the radioimmunoassay method, involving 3074 patients, gave a sensitivity of 0.82 and specificity of 0.81. Data from the three studies that relied on an enzyme immunosorbent assay (EIA) gave a sensitivity of 0.75 and specificity of 0.79. Data from the 30 studies (4879 patients) using a cut-off value of 37 U/mL gave a sensitivity of 0.82 and specificity of 0.80. Data from the three studies using a cut-off value of 100 U/mL gave corresponding values of 0.69 and 0.85. These variations in sensitivity and specificity with CA19-9 assay and cut-off values did not achieve statistical significance (P > 0.05, Table 4), suggesting that high cut-off values such as 100 U/mL may better increase the specificity for differential diagnosis of pancreatic carcinoma.

Instead of assessing diagnostic power using the traditional ROC plot, we calculated an SROC plot to reveal the effect of varying thresholds on sensitivity and specificity within each study. In this plot, different studies appear as different data points, allowing SROC curves to provide a global summary of test performance and illustrate the trade-off between sensitivity and specificity. Figure 3 shows an SROC curve for rates of true- and false-positive results obtained with the CA19-9 assay in individual studies. From this plot we determined the Q value, which was defined as the point of intersection of the SROC curve with a diagonal line extending from the left upper corner to the right lower corner of the plot. The Q value indicates the highest identical value of sensitivity and specificity, thereby serving as an overall measure of the discriminatory power of a test. Our SROC curve was desirably positioned near the upper left corner, and the maximum joint sensitivity and specificity was 0.81. The area under the curve (AUC) was 0.88 (Figure 3A), indicating high overall accuracy. SROC plots differed based on the CA19-9 assay method and cut-off values, but all plots were positioned near the upper left corner with AUCs near 0.88 (Figure 3B-F), again indicating high overall accuracy.

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Summary receiver operating characteristic curves for carbohydrate antigen 19-9 assays for differential diagnosis of pancreatic carcinoma and chronic pancreatitis. Solid circles represent each study included in the meta-analysis, with circle size proportional to the number of participants in the study. SROC curves summarize the overall diagnostic accuracy for all included studies (A), studies using a cut-off of 37 U/mL carbohydrate antigen 19-9 (CA19-9) (B), studies using a cut-off of 100 U/mL CA19-9 (C), studies based on the radioimmunoassay method to assay CA19-9 (D), studies based on the ELISA method (E), and studies based on the enzyme immunoassay method (F). ELISA: Enzyme-linked immunosorbent assay; SROC: Summary receiver operating characteristic.

Quality scores based on the STARD[11] and QUADAS[12] guidelines were generated for every study on the basis of the title and introduction, methods, results and discussion (Table 1). These scores were used in meta-regression to assess the effect of study quality on the RDOR of CA19-9 in the differential diagnosis of pancreatic carcinoma and chronic pancreatitis. Studies of higher quality (STARD score ≥ 13; QUADAS score ≥ 10) produced RDOR values similar to those of lower-quality studies. In addition, RDOR values did not differ significantly as a function of blinding, cross-sectional or case-control design, consecutive or random sampling, prospective or retrospective design, CA19-9 assay method, or cut-off values (P > 0.05). These results suggest that study design did not significantly affect diagnostic accuracy and that the risk of detection bias was low.

The Egger test showed no significant evidence of publication bias in reports about CA19-9 assays for differential diagnosis of pancreatic carcinoma (P = 0.944).

Pancreatic carcinoma and chronic pancreatitis show similar clinical manifestations. Carbohydrate antigen 19-9 (CA19-9) shows promise for differentiating the diseases.

Differential diagnosis of pancreatic carcinoma and chronic pancreatitis remains a challenge, particularly in patients with pancreatic masses that may be benign (inflammatory) or malignant.

This is believed to be the first systematic review and meta-analysis of the utility of CA19-9 as a serum tumor marker and its sensitivity and specificity for distinguishing pancreatic carcinoma and chronic pancreatitis.

Available evidence suggests that elevated CA19-9 by itself is insufficient for differentiating pancreatic carcinoma and chronic pancreatitis. Nevertheless, elevated CA19-9 should increase suspicion of pancreatic carcinoma and therefore may complement other clinical findings to improve the accuracy of differential diagnosis.

This study evaluated the role of CA19-9 for differentiating pancreatic carcinoma and chronic pancreatitis and concluded that CA19-9 itself is insufficient, but elevated CA19-9 should increase suspicion of pancreatic carcinoma and therefore may complement other clinical findings to improve the accuracy of differential diagnosis. This result is reasonable and valuable for clinical practice.