In a recent study, the “C the Signs” AI-based test achieved 93.8% sensitivity and 19.7% specificity for early colorectal cancer (CRC) detection. The model identified 29.4% of CRC patients as high-risk up to five years earlier than primary care physicians.
These findings were presented as a poster at the recent ASCO Gastrointestinal Cancers Symposium. Seema Dadhania, consultant clinical oncologist, Imperial College London Department of Surgery & Cancer, is one of the study investigators. She was interviewed by the American Journal of Managed Care (AJMC), which reported the study (Steinzor, P, AJMC, Jan. 23.)
Its developers say that in under 30 seconds, C the Signs can rapidly identify which cancers a patient is at risk of and recommend the most appropriate test or specialist to diagnose their cancer. With such high sensitivity and strong negative predictive value, this tool should enhance early CRC detection and risk stratification in primary care.
For this study, “we were trying to understand if the C the Signs platform could pick up the signs or risk of colorectal cancer earlier than the patient was actually diagnosed,” Dadhania told AJMC, “so [we’re] trying to understand if there is a way to pick up earlier signals of colorectal cancer by using an aggregation of symptoms or other pieces of data which were present within the [Mayo Data Platform].”
The rate of CRC, particularly among younger people, is growing. In early 2023, the American Cancer Society (ACS) reported that 20% of diagnoses in 2019 were in patients under age 55, which is about double the rate in 1995, and rates of advanced disease increased by about 3% annually in people younger than 50. It predicted that, in 2023, an estimated 19,550 diagnoses and 3,750 deaths would be in people younger than 50.
This trend creates a growing public health challenge. Because young people are not usually screened, many CRC cases are still identified only after symptoms appear, often at later stages when outcomes are poorer.
This retrospective study looked at data from primary care settings to evaluate the performance of C the Signs. Using the Mayo Data Platform, researchers analyzed a comprehensive dataset of electronic medical records (EMRs) spanning 20 years, from January 1, 2002, to December 31, 2021. A total of 894,275 patients’ data was analyzed, and 7,348 patients were diagnosed with CRC during the study period. The model achieved a sensitivity of 93.8% and a specificity of 19.7% in identifying patients at risk of CRC. Notably, the model identified 29.4% of these patients as high-risk up to five years earlier than primary care physicians.
“In order to get a patient to have a colonoscopy, they have to have a set of criteria or a group of symptoms that warrant the appropriate referral,” explained Dadhania, in the AJMC interview. “I think what this tool is allowing is it’s capturing those patients who will already get through the system, but it’s also capturing a proportion of patients who won’t get through the current system as it stands, to get a colonoscopy.”
