Balasundaram, Ponni; (2024) Improved detection of pancreatic neuroendocrine tumours through the development of a biomarker panel. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Ponni_Balasundaram_Final Thesis.pdf - Accepted Version Download (6MB) | Preview

Abstract

Pancreatic neuroendocrine tumour (pNET) patients present with non-specific symptoms leading to delayed diagnosis which in turn impacts survival. The current gold standard marker Chromogranin A (CgA) has limitations, including associated confounding factors such as proton pump inhibitor use and limited utility for insulinomas. In clinical practice, pNETs need to be discriminated from other pancreatic conditions such as pancreatic ductal adenocarcinoma (PDAC), acute pancreatitis (AP) and chronic pancreatitis (CP), but currently no test is available for this distinction to be made. A multianalyte biomarker approach for pNET detection, using seven previously explored GEP-NET markers was explored for their suitability for pNET detection. A training cohort of pNET patient and healthy control samples was used to develop machine learning (ML) algorithms across seven different marker combinations. Based on this work, a three-marker combination of CgA, VGF-nerve growth factor inducible peptide (VGF-NGF) and Angiopoietin-2 (ANG2) was identified as suitable across the algorithms assessed, and that the seven markers could be reduced to three markers without a large impact on performance, meaning a more cost-effective test using fewer markers. The suitability of this three-marker combination was also confirmed in internal validation. However, at external validation using an independent sample of pNETs, VGF-NGF was deemed not to be suitable and the performance of ANG2 and CgA algorithms was less than that seen using the training cohort. Models created based on the training pNET case and healthy control evaluation were also not found to discriminate between AP, CP, PDAC and pNETs. In summary, reduction to a two or three-marker pNET detection panel performed well in the ML model training stage but did not show strong performance in external validation. Further markers, perhaps identified in early pNET biological models are likely to be needed in combination for better identification in a point of care test.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item