The identification of blood pattern for gastric cancer from routine blood tests in a territory-wide big clinical database.

person Ka Man Cheung Department of Clinical Oncology, Queen Elizabeth Hospital, Hong Kong, Hong Kong info_outline Ka Man Cheung, Tsz Chun Bryan Wong, James C.H. Chow, Serene J.L. Lam, Kelvin K.H. Bao, Peter Y.M. Woo, Therese Yue Man Tsui, H.L. Leung, Kam-Hung Wong, Harry H.Y. Yiu, David C.C. Lam

Authors person Ka Man Cheung Department of Clinical Oncology, Queen Elizabeth Hospital, Hong Kong, Hong Kong info_outline Ka Man Cheung, Tsz Chun Bryan Wong, James C.H. Chow, Serene J.L. Lam, Kelvin K.H. Bao, Peter Y.M. Woo, Therese Yue Man Tsui, H.L. Leung, Kam-Hung Wong, Harry H.Y. Yiu, David C.C. Lam Organizations Department of Clinical Oncology, Queen Elizabeth Hospital, Hong Kong, Hong Kong, The Hong Kong University of Science and Technology, Hong Kong SAR, Hong Kong, Departments of Clinical Oncology, Queen Elizabeth Hospital, Hong Kong, Hong Kong, The Chinese University of Hong Kong, Hong Kong, Hong Kong, Queen Elizabeth Hospital, Kowloon, Hong Kong, Department of Neurosurgery, Prince of Wales Hospital, Hong Kong SAR, Hong Kong, Department of Clinical Oncology, Queen Elizabeth Hospital, Kowloon, Hong Kong, Department of Oncology, United Christian Hospital, Hong Kong SAR, Hong Kong Abstract Disclosures Research Funding No funding received None. Background: Development of non-invasive means for early detection is key to reduce gastric cancer mortality. Artificial intelligence (AI) has to potential to accurately identify gastric malignancy related physiological processes including subclinical bleeding, systemic inflammation and coagulation disorder, through machine learning of pattern of change in routine blood profile panel, which constitute a blood signature for cancer. The pattern of laboratory values of various components in the blood panels between diseased and non-diseased were not systematically reported in the Big Clinical population cohort. Methods: This is a territory-wide healthcare database study using data from Hong Kong Hospital Authority data collaboration laboratory. All patient prescribed with medications for dyspepsia from year of 2004-2009, 2011-2014 were retrieved. Disease status (patients with gastric cancer and no cancer diagnosis) and blood test results (Blood tests within 3 months closest to the date of diagnosis of gastric cancer and any blood tests for negative controls) were retrieved. Blood tests retrieved included CBC, LFT, RFT and clotting function. The laboratory values were normalized according to normal limits. Components with low record counts were excluded. The differences in median values across groups were tested with Mann-Whitney U test. Results: 7,734 patients with gastric cancer diagnosis and 603,484 non-diseased patients were included. The record counts for basophil and eosinophil counts are low and were excluded. In the group of diseased patients, only median of Hb level, RBC count, RDW, hematocrit, albumin, ALT and AST were out of normal range. The pattern for laboratory values in pts with gastric cancer is distinct, which shows lower RBC, hemoglobin, hematocrit, lower MCH, lower MCHC, lower MCV, higher RDW which could suggest occult bleeding; higher platelet, higher neutrophil and lower lymphocyte inclining systemic inflammation; lower protein, albumin and creatinine, inclining impaired nutritional status; shorter APTT and PT inclining clotting disorders; and higher ALT and AST level suggesting liver irritation, which may be related to systemic inflammation and metastases. All of the above differences were statistically significant (p < 0.05, Mann-Whitney U test). Conclusions: This is the first population-based description of the pattern of routine blood test results between patients with and without gastric cancer. Statistical analysis showed that the variations in blood components of subjects with gastric cancer is distinct and separable from the variations of non-diseases patients. The finding opens the opportunity for discovery and identification of a spectral signature for gastric cancer in the Big Clinical population cohort using AI. The classification outcome using Big Clinical Data and AI are reported in companion abstract.