The Diagnostic Value of the CA19-9 and Bilirubin Ratio in Patients with Pancreatic Cancer, Distal Bile Duct Cancer and Benign Periampullary Diseases, a Novel Approach

Abstract

Distinction of pancreatic ductal adenocarcinoma (PDAC) in the head of the pancreas, distal cholangiocarcinoma (dCCA), and benign periampullary conditions, is complex as they often share similar clinical symptoms. However, these diseases require specific management strategies, urging improvement of non-invasive tools for accurate diagnosis. Recent evidence has shown that the ratio between CA19-9 and bilirubin levels supports diagnostic distinction of benign or malignant hepatopancreaticobiliary diseases. Here, we investigate the diagnostic value of this ratio in PDAC, dCCA and benign diseases of the periampullary region in a novel fashion. To address this aim, we enrolled 265 patients with hepatopancreaticobiliary diseases and constructed four logistic regression models on a subset of patients (n = 232) based on CA19-9, bilirubin and the ratio of both values: CA19-9/(bilirubin^-1). Non-linearity was investigated using restricted cubic splines and a final model, the 'Model Ratio', based on these three variables was fitted using multivariable fractional polynomials. The performance of this model was consistently superior in terms of discrimination and calibration compared to models based on CA19-9 combined with bilirubin and CA19-9 or bilirubin alone. The 'Model Ratio' accurately distinguished between malignant and benign disease (AUC [95% CI], 0.91 [0.86-0.95]), PDAC and benign disease (AUC 0.91 [0.87-0.96]) and PDAC and dCCA (AUC 0.83 [0.74-0.92]) which was confirmed by internal validation using 1000 bootstrap replicates. These findings provide a foundation to improve minimally-invasive diagnostic procedures, ultimately ameliorating effective therapy for PDAC and dCCA.

Keywords: CA19-9; CA19-9-bilirubin-ratio; benign periampullary diseases; biliary tract cancer; bilirubin; diagnostic biomarker; distal bile duct cancer; liquid biopsy; pancreatic adenocarcinoma; pancreatic cancer.

Figure 1

Evaluation of baseline CA19-9 and bilirubin levels of all patients. Box-and-whisker plots of CA19-9 and bilirubin levels in patients with PDAC = pancreatic ductal adenocarcinoma, dCCA = distal cholangiocarcinoma, and benign disease. The median is shown by a thick stripe, and box depicts the interquartile range (IQR) with 75th percentile + 1.5*IQR and 25th percentile-1.5*IQR (whiskers). Significant comparisons are shown with p values described below. p-values were calculated with the Kruskal–Wallis test followed by Dunn’s test for multiple comparisons. The false discovery rate was controlled with the Benjamini–Hochberg procedure.

Figure 2

Four logistic regression models adopted in this study.

Figure 3

Overfitting-corrected calibration curve using a loess nonparametric smoother and 1000 bootstrap repetitions.

Figure 4

ROC curves and AUC comparisons of the ‘Model Ratio’ with the models using only CA19-9 and bilirubin. Malign vs. benign; Model ratio vs. CA19-9: p = 0.002 (**: p < 0.01), Model ratio vs. bilirubin: p = 3.0 × 10⁻⁷ (****: p < 0.0001) Model ratio vs. CA19-9 + bilirubin: p = 0.071 (ns: p > 0.05) PDAC vs. benign; Model ratio vs. CA19-9: p = 0.0011 (**: p < 0.01), Model ratio vs. bilirubin: p = 3.3 × 10⁻⁷ (****: p < 0.0001); Model ratio vs. CA19-9 + bilirubin: p = 0.048 (*: p < 0.05) PDAC vs. dCCA; Model ratio vs. CA19-9: p = 0.0009 (***: p < 0.001) Model ratio vs. bilirubin: p = 0.0002 (***: p < 0.001) Model ratio vs. CA19-9 + bilirubin: p = 0.0004 (***: p < 0.001).

Figure 5

Overoptimism-corrected decision curve analysis using 1000 repeats of 5-fold cross-validation. Five-fold cross-validation was repeated 1000 times and results were averaged across all repetitions to obtain robust estimates of the standardized net benefit at each threshold. Briefly, the clinical usefulness of a diagnostic test is assessed by comparing standardized net benefit over a range of probability thresholds. Net benefit is defined as the number of true positives found at a certain cut-off penalized by the number of false positives found for that cut-off, where the weight of the penalty is defined by the relative importance of finding a false positive. For further information regarding decision curve analysis, the reader is referred to [33,34].