Abstract
BACKGROUND: Pancreatic cancer is highly aggressive and has a low survival rate primarily due to late-stage diagnosis and the lack of effective early detection methods. We introduce here a novel, noninvasive urinary extracellular vesicle miRNA-based assay for the detection of pancreatic cancer from early to late stages. METHODS: From September 2019 to July 2023, Urine samples were collected from patients with pancreatic cancer (n = 153) from five distinct sites (Hokuto Hospital, Kawasaki Medical School Hospital, National Cancer Center Hospital, Kagoshima University Hospital, and Kumagaya General Hospital) and non-cancer participants (n = 309) from two separate sites (Hokuto Hospital and Omiya City Clinic). The main inclusion criteria included a diagnosis of pancreatic cancer based on pathological or imaging examination, while multiple primary cancers were excluded. Extracellular vesicles were enriched using a polymer-based precipitation method, and miRNAs were comprehensively analyzed by small RNA sequencing. A machine learning model for pancreatic cancer detection was developed using a training dataset (n = 315) consisting of 99 pancreatic cancer participants (of which 33 were early-stage [I/IIA]) and 216 non-cancer participants, and validated with a test dataset (n = 147) consisting of 54 pancreatic cancer participants (of which 9 were early-stage [I/IIA]) and 93 non-cancer participants. FINDINGS: This method showed consistent performance, with areas under the receiver operating characteristic curves of 0.972 (95% confidence interval [CI], 0.928-0.996) and 0.963 (95% CI, 0.932-0.988) in the training and test sets, respectively. The sensitivities for pancreatic cancer detection were 93.9% (95% CI, 87.5%-97.3%) and 77.8% (95% CI, 64.9%-87.3%) overall and 97.0% (95% CI, 83.9%-99.8%) and 77.8% (95% CI, 44.2%-95.9%) for stage I/IIA pancreatic cancer, respectively. The specificities were 91.7% (95% CI, 87.1%-94.7%) and 95.7% (95% CI, 89.4%-98.5%), respectively. We also evaluated the sensitivity of CA19-9 for pancreatic cancer detection in 140 patients with pancreatic cancer, and it was 37.5% (95% CI, 23.5%-53.8%) for stages I/IIA pancreatic cancer. Performance in early-stage cancer detection was significantly higher for miRNA-based pancreatic cancer detection. Functional enrichment analysis of pancreatic cancer-associated urinary miRNAs revealed that the urinary miRNA signature reflects miRNA patterns of the pancreatic cancer tissue itself as well as those of the tumor microenvironment. INTERPRETATION: Urinary extracellular vesicle miRNAs may reflect signals from both tumor cells and their microenvironment, offering a unique opportunity for detection of pancreatic cancer from early to late stages. While this study has a limitation due to the relatively small sample size, our approach has the potential to contribute to treatment outcomes through population screening. Our primary goal is to make this assay more accessible to a broader population, particularly in areas with limited hospital access where cancer is often detected at a late stage, leveraging the advantage of using urine samples that can be collected at home. FUNDING: This research was supported by the Japan Agency for Medical Research and Development (AMED) under Grant Number JP24he2302007 and Craif Inc.