969P - Establishing a novel routine blood component signature for Hepatocellular Carcinoma (HCC) screening with big clinical data

Background

Alpha-fetoprotein (AFP) is widely used for HCC screening in at-risk populations, but its performance is suboptimal. Routine blood test panels provide insights on multiple cancer-related conditions and have shown to improve detection in other cancers. Hence, we explore the foundation for a new high-dimensional data approach taking multiple routine laboratory parameters and their intricate interactions into consideration with AI. In this study, a routine blood test signature for HCC derived from big clinical data is described

Methods

This is a population-based retrospective study. Patient records from 2000 to 2018 were retrieved from the Hong Kong Hospital Authority Data Collaboration Laboratory. CLD patients with and without HCC were identified based on ICD codes, antiviral drug history, virology test and radiology reports. Patients with decompensated CLD were excluded. Routine blood tests retrieved included CBC, LFT, RFT and clotting profile. Blood records within one month before the diagnosis of HCC and CLD patients were retrieved. Statistical analysis included descriptive statistics and Mann-Whitney U (MWU) test.

Results

The cohort yielded 223,862 patients including 31,149 patients with HCC (13.9%) and 192,713 without. Statistical test results showed a distinctive spectral signature for HCC patients compared to those without HCC. It is characterized by the concurrent presence of more significant liver function derangement (raised ALT, ALP, bilirubin, AST and decreased albumin), tendency for systemic inflammation (lower lymphocyte and RDW), tendency for bleeding (prolonged PT and APTT, low platelet) and suggestions of cachexia (lower albumin, creatinine and urea), all were statistically significant (P<0.05, MWU test)

Conclusions

This is the first study to describe a routine blood signature for HCC detection established by big clinical data. The spectral characteristics of HCC separated well from CLD controls. The novel spectrum provides solid clinical ground for the use of advanced machine learning to generate an interactive classification model to detect HCC. In a companion abstract, we describe the development of the proposed signature, which is shown to be superior to AFP in HCC screening.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The Hong Kong University of Science and Technology.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.