Multi-Cancer Early Detection Test Based on Multivariate Model Achieves 75% Sensitivity in Joint Clinical Validation by Two Hospitals

From October 11 to 13, 2019, the “ASCO Breakthrough Summit,” hosted by the American Society of Clinical Oncology (ASCO), was held in Bangkok, Thailand. As one of the most prominent conferences in the field of cancer diagnosis and treatment this year, the summit focused on the latest technological breakthroughs, scientific advancements, and clinical practices. Through exchanges and interactions among participants from around the world, it aimed to spark innovation and advance the development of cancer care.

Led by an industry enterprise, in collaboration with the Department of Hepatobiliary Surgery at Peking University Shenzhen Hospital and the Department of Clinical Laboratory at Sun Yat-sen Memorial Hospital, Sun Yat-sen University, this research employed low-pass whole-genome sequencing of cell-free DNA (cfDNA) combined with plasma tumor marker detection. By leveraging machine learning methods to construct a Multivariate Cancer Risk Score (MCRS) model for distinguishing cancer patients from healthy individuals, the study achieved significant results in early screening for multiple cancers. These findings were presented as an oral presentation at ASCO Breakthrough, selected as one of only four oral presentations in the Molecular Diagnostics section, and stood as the sole oral presentation from China.

It is understood that this study, launched in early 2018, explores methods for early screening of various cancers by integrating different types of cancer samples and clinical information to build, analyze, and validate predictive models. Long Guanghui, Director of the Hepatobiliary Surgery Department at Peking University Shenzhen Hospital, told VCBeat that clinicians hope to achieve early cancer screening to enable timely surgical resection, “as tumors are typically unresectable once they reach an advanced stage.” Duan Zhaohui, the principal investigator of this study from Sun Yat-sen Memorial Hospital of Sun Yat-sen University, stated that while conventional cancer screening in laboratory medicine already utilizes certain protein biomarkers, issues with specificity and sensitivity persist. Therefore, there is a need to develop DNA-based molecular screening methods to address the limitations of traditional protein-based approaches.

As a preliminary step in this study, researchers developed the MCRS algorithm model and established the cancer classification threshold using a training set comprising 465 subjects (39 cancer patients and 426 healthy individuals). Subsequently, they evaluated the accuracy of the MCRS algorithm model in an independent validation set consisting of 84 newly diagnosed Stage I–IV cancer patients and 338 healthy individuals.

Clinical studies have found that shallow whole-genome sequencing (sWGS) results can effectively detect copy number variation (CNV) changes in subjects, with a significant difference in the extent of CNV abnormalities between cancer patients and healthy individuals. By combining CNV analysis results with the detection of seven tumor markers and incorporating the MCRS algorithm model, cancer patients can be distinguished from healthy individuals. Results from the validation set showed that, with a specificity of 98.8%, the sensitivity of combining low-coverage cfDNA whole-genome sequencing with tumor markers reached 53.6%.

Director Long Guanghui pointed out that the most noteworthy and unique aspect of this study lies in its use of cfDNA fragmentation patterns as a novel informational dimension, further enhancing both specificity and sensitivity.

Studies have shown that circulating tumor DNA (ctDNA) fragments derived from tumor cells in the blood are shorter than cell-free DNA (cfDNA) from normal cells, and cfDNA fragment size can be assessed through end-to-end sequencing. Clinical data indicate that incorporating fragment size as an additional dimension into the original MCRS model resulted in a sensitivity of 75.0% at a specificity of 98.8% in the validation cohort, representing a 21.4% improvement over the original model. It was disclosed that the multi-dimensional, pan-cancer early screening technology employed in this study was provided by Shenzhen Siqin Medical, a genetic testing startup. Dr. Mao Mao, founder of Siqin Medical, stated that more novel tumor characteristics will be incorporated in the future to achieve more accurate and effective screening.

The results of this study indicate that the blood-based, non-invasive multi-cancer early screening test developed by Siqin Medical can accurately detect various types of cancer, achieving a sensitivity of 75% at a specificity of 98.8%. Furthermore, the accuracy of the MCRS model can be further enhanced by incorporating cancer feature variables from multiple dimensions. Director Long Guanghui pointed out that this study provides a non-invasive, cost-effective, molecular-level auxiliary diagnostic method for the early clinical diagnosis of multiple cancers.

In recent years, with the rapid development of next-generation detection technologies, a number of startups focused on early cancer screening have emerged both domestically and internationally. “The technical approaches and detection methods adopted by different companies vary significantly,” introduced Dr. Mao Mao. Currently, blood-based early cancer screening can be categorized into three main schools of thought.

First, regarding the detection of circulating tumor DNA methylation represented by the U.S.-based company Grail, Grail also released its latest experimental data at this year’s ASCO Breakthrough Conference.

Second, the technological approach exemplified by “CancerSEEK,” published in Science in 2018, enables early screening for ovarian, liver, gastric, pancreatic, and esophageal cancers by analyzing 16 cancer-related gene mutations in circulating tumor DNA and plasma protein biomarkers associated with cancer. According to the data presented in the article, both specificity and sensitivity are considerably high. Based on this technology, the relevant team established the startup company Thrive Earlier Detection.

Third, the early cancer screening technology adopted in this study. Unlike other technical approaches, this technique combines low-depth whole-genome sequencing of cell-free DNA (cfDNA) with plasma tumor marker detection, and employs machine learning methods to construct a Multivariate Cancer Risk Score (MCRS) model to distinguish cancer patients from healthy individuals, thereby enabling early screening for multiple types of cancer. “Due to tumor heterogeneity, genetic variations exist within cells of the same tumor type. Whole-genome testing allows us to obtain information from as many genomic loci as possible. By integrating and weighting data from multiple loci and dimensions, we can more accurately detect the onset and progression of tumors,” explained Dr. Mao Mao.

Finally, relevant officials from Peking University Shenzhen Hospital and Sun Yat-sen Memorial Hospital of Sun Yat-sen University, along with Siqin Medical, stated that they would continue to advance the translational medicine efforts based on these research findings. While conducting further clinical validation, they aim to promote the technological achievements in clinical practice. According to Dr. Mao Mao, clinical studies on tumor origin tracing based on the MCRS model have been underway for over a year. The team will promptly report progress once more data and further R&D enhance the accuracy of determination.