Authors (including presenting author) :
Cheung KM (1), Seo M (2), Sung W (1), Lee SC (3), Liu HW (4), Yip ASM (5), Woo PYM (6), Chow JCH (1), Ng SKK (5), Wu LT (7), Kan DMY (5), Kao S (3), Yiu HHY (1), Lam DCC (2)
Affiliation :
(1) Department of Clinical Oncology, Queen Elizabeth Hospital (2) Department of Mechanical and Aerospace Engineering, HKUST, (3) Surgery, (4) Department of Medicine, Queen Elizabeth Hospital, (5) Surgery, Kwong Wah Hospital, (6) Neurosurgery, Prince of Wales Hospital, (7) Research Office, Queen Elizabeth Hospital
Introduction :
The high mortality of gastric cancer (GC) is largely due to late diagnosis, and a reliable non-invasive detection method is lacking. While existing tumour markers have low sensitivity, routine blood tests (RBTs) like CBC and LRFT can indicate hidden cancer through signs of anaemia, inflammation, and cachexia. The HKUST, HA clinical team (oncology, medicine, surgery, neurosurgery), and HADCL co-developed the RBT-GC model, which can detect GC up to 6 months prior to clinical diagnosis. Previously presented at ASCO 2023 and ESMO GI 2024, this study aims to validate RBT-GC's sensitivity in a hospital setting up to 6 months before the diagnosis date.
Objectives :
To validate the early detection performance of RBT-AI screening for gastric cancer prior to OGD diagnosis
Methodology :
This study was supported by KCC Research Grant with a research assistant. We retrieved patients diagnosed with gastric cancer in QEH in 2020 from CDARS. Patients with other cancers and with incomplete blood records were excluded. Routine blood tests, including CBC and LRFT, up to 6 months before diagnosis date were retrieved. RBT-GC model was applied to each record, which gave an output ranging from 0 to 1. The scores above 0.48 were classified as predictive positive, which was previously validated cutoff for early detection.
Result & Outcome :
RBT records (N=49) demonstrated an increasing trend in GC risk score over the six months before diagnosis, with medians rising from 0.36 at 6 months, 0.55 at 3 months, and 0.66 at 1 month prior to the GC diagnosis. Sensitivity improved from 0.29 (4 out of 14 cases) 6 months prior, to 0.58 (7 out of 12 cases) in 3 months prior, and 0.74 (29 out of 39 cases) 1 month prior to diagnosis. This indicates that GC risk score effectively reflects tumour progression over time. There were only 8 cases with CEA records available at one month prior to GC diagnosis, and none for three or six months. The median CEA level was 2.90 ng/ml, with only one case exceeding the conventional upper normal limit 5 ng/ml (sensitivity 12.5%). This study confirms the early detection capabilities of RBT-GC model, showing an increasing trend in GC risk scores and improved sensitivity as diagnosis approaches. The limited availability and low accuracy of CEA necessitates the better alternative for non-invasive means for cancer detection.
