Expanding surveillance eligibility through EMR and AI
Electronic health records (EHR) are increasingly used as real-world data to inform predictive AI models, which are now being created to identify people likely to develop challenging cancers. These models analyze extensive clinical records of larger populations, weighing an array of clinical measurements to generate criteria for high-risk surveillance cohorts beyond current monitoring programs, and continue to evolve as new information emerges for analysis.

The recent proliferation of advanced genetic tests is creating another data set that can be evaluated by AI to identify disease markers. Comprehensive EMR databases have enhanced the accuracy of genetic testing, and these tests are now commonly used for newborns, rare diseases, and cancer risk identification in families. AI can create more accurate risk profiles by combining common gene alterations associated with people who have developed certain cancer types with family history and clinical data.
A powerful duo – AI and surveillance
Predictive diagnostic tests are influenced by a person's prior probability of developing the disease, making the positive predictive value of these tests closely correlated with disease prevalence in a group. AI systems can detect those at risk for cancer using probabilities and predictions even when the available information is incomplete or uncertain, enriching disease prevalence in a surveillance cohort. Once individuals with a high risk of disease are identified, clinicians can use appropriate surveillance measures to detect early disease onset.

When individuals are initially surveyed, oncologists often find that they do not fit the criteria to enter active surveillance for all cancers. For instance, if someone is identified as a BRCA carrier in their 20s or 30s, it is unlikely they will be diagnosed with pancreatic cancer at that time. However, since breast and ovarian cancers are more likely to develop during this time, conducting regular, cancer-specific screenings is essential. As a person enters their 50s and 60s, pancreatic cancer in BRCA carriers is more likely to occur, and therefore, it may be appropriate to initiate surveillance around 35 to 45 years of age. The primary objective of predictive cancer diagnostics is to enable early-stage detection without burdening people with unnecessary testing.

In a recent study published in Nature Medicine, an AI model was developed and trained with the goal of predicting pancreatic cancer risk three years before diagnosis based on a sequence of specific conditions subjects experienced over time. Implementing AI on actual clinical records could aid the development of a more adaptable workflow that would enable larger populations to be identified for surveillance.

Business Affairs Homepage