Urine biomarkers enable pancreatic cancer detection up to 2 years before diagnosis

Abstract

The poor prognosis of pancreatic ductal adenocarcinoma (PDAC) is mainly attributed to late diagnosis. We assessed the predictive performance of our previously reported urine biomarker panel for earlier detection of PDAC (LYVE1, REG1B and TFF1) in prediagnostic samples, alone and in combination with plasma CA19-9. This nested case-control study included 99 PDAC cases with urine samples prospectively collected up to 5 years prior to PDAC diagnosis and 198 matched controls. The samples were obtained from the Shanghai Women's Health Study (SWHS), the Shanghai Men's Health Studies (SMHS) and the Southern Community Cohort Study (SCCS). The urine biomarkers were measured by ELISA. Plasma CA19-9 was quantified by Luminex. Multiple logistic regression and Wilcoxon rank-sum and Mann-Whitney test were used for analysis. The internal validation approach was applied and the validated AUC estimators are reported on. The algorithm of urinary protein panel, urine creatinine and age named PancRISK, displayed similar AUC as CA19-9 up to 1 year before PDAC diagnosis (AUC = 0.79); however, the combination enhanced the AUCs to 0.89, and showed good discriminative ability (AUC = 0.77) up to 2 years. The combination showed sensitivity (SN) of 72% at 90% specificity (SP), and SP of 59% at 90% SN up to 1 year and 60% SN with 80% SP and 53% SP with 80% SN up to 2 years before PDAC diagnosis. Adding the clinical information on BMI value resulted in the overall improvement in performance of the PancRISK score. When combined with CA19-9, the urinary panel reached a workable model for detecting PDAC cases up to 2 years prior to diagnosis.

Keywords: early detection; pancreatic cancer; prediagnostic samples; urine biomarkers.

FIGURE 1

Performance of the biomarkers. (A) Mann‐Whitney t‐test between cases in IQ1, IQ2, IQ3, IQ4 and respective matched controls (the number of samples in each group is indicated in the brackets; ns, not significant); (B) Mann‐Whitney t‐test between cases in IQ1, IQ2, IQ3, IQ4 and the combined controls; (box and whisker histograms 5‐95 percentile); (C) distribution of cases (%) by year group in each IQ

FIGURE 2

Performance of the PancRISK alone and in combination with plasma CA19‐9. (A) AUC (no split, training and testing on the same data) and (B) AUC, leave‐one‐out cross‐validation up to 5 years prior to diagnosis. Receiver operating characteristic (ROC) curves of the PancRISK and CA19‐9 alone and in combination was used to evaluate their performance in distinguishing prediagnostic cases (C) 1 year and (D) 2 years before diagnosis (DX)