AI detects gastric cancer with high accuracy in common blood tests

An algorithm based on 29 different blood parameters was able to predict early gastric cancer with high accuracy, high sensitivity, and low false positive and false negative rates. Therefore, the signature could be used to enhance cost-effectiveness and public participation rates in screening programmes for gastric cancer.

The lack of symptoms in the early stage of gastric cancer leads to delayed clinical presentation and low overall survival rate. Invasive screening, such as oesophagogastric duodenoscopy, is effective in decreasing gastric cancer mortality, but adherence is low [1,2]. Thus, there is a need for minimal or non-invasive gastric cancer screening, for example with routine blood tests. Algorithms based on routine blood tests were previously developed for the prediction of (colorectal) cancer, though low sensitivity, specificity, and positive predictive value limited their use [3,4].

Mr Tsz Chun Bryan Wong (Hong Kong University of Science and Technology, Hong Kong) and colleagues developed a deep learning AI algorithm using data from routine blood tests from more than 190,000 individuals who were prescribed medications for dyspepsia in the period between 2004–2015 [5]. Of these individuals, 4,790 patients were diagnosed with gastric cancer. The blood tests contained 29 parameters, including complete blood counts, liver function parameters, renal function parameters, and clotting function parameters.

Data from 2004 to 2009 and from 2011 to 2014 were used as a training cohort and the remainder was used as a testing cohort. A ‘gastric cancer’ signature in the blood was generated using deep-learning AI algorithms. After training and testing, the signature was able to predict gastric cancer with high sensitivity (96%), high specificity (100%), high positive predictive value (99%), and high negative predictive value (100%). In 2 large validation cohorts (n= 24,610 and n=17,058) the signature proved to be highly accurate (see Figure).

Figure: Validation of machine learning model to detect gastric cancer signature in blood [5]



FN, false negative; FP, false positive; NPV, negative predictive value; PPV, positive predictive value; TN, true negative; TP, true positive

Based on these outcomes, Mr Wong concluded that “the blood signature can be used to enhance cost-effectiveness and public participation rates in screening for early gastric cancer.”

1. Yashima K, et al. J Clin Med. 2022;11:4337.
2. Suh YS, et al. Cancer 2020;126:1929–1939.
3. Goshen R, et al. Br J Cancer. 2017;116: 944–950.
4. Kinar Y, et al. J Am Med Inform Assoc. 2016;23:879–890.
5. Wong TCB, et al. AI blood signature in common blood test for detection of gastric cancer in a cohort of 190,000 individuals. Abstract 1500, ASCO Annual Meeting 2023, 2–6 June, Chicago, USA.

 

Copyright ©2023 Medicom Medical Publishers



Posted on 2 August, 202327 March, 2024