Cross-population validation of PESI-MS and machine learning models for pancreatic cancer diagnosis: Insights from an independent Japanese cohort

Pancreatic cancer remains a major challenge in oncology due to its poor prognosis as it is symptomatically dormant and is found in later intractable stages. This study aims to validate the robustness and versatility of a diagnostic framework combining Probe Electrospray Ionization Mass Spectrometry (PESI-MS) and machine learning models for early PDAC detection across different ethnicities. Building on our previous study that demonstrated high sensitivity and specificity in a Taiwanese PDAC high-risk and PDAC patient cohort, we used those prebuilt models on an independent cohort of 38 Japanese patients diagnosed with pancreatic cancer. The model assessed sustained its performance on the Japanese cohort, correctly identifying 97% of all cancer cases, indicating robustness to demographic differences. Moreover, enhancing the original model with dimensionality reduction achieved 100% sensitivity, correctly detecting every cancer case. These findings underscore the potential of PESI-MS and machine learning as universal diagnostic tools for PDAC, with significant implications for early detection that improves patient outcomes. The findings advocate for further validation in larger, multi-ethnic studies to confirm the clinical implementation of this approach and its integration into global screening strategies for high-risk populations.