Challenge A

„Detecting Real-Time Adverse Effects“

Proposed by Bayer

Background

Medical Devices are part of the Health Care Environment. Digital steering mechanism as well as digital data monitoring and output has become an essential part of Medical Devices. In non- interventional studies these medical devices could be used in a sensible manner to gain more insight in the ‘Real Life World’ setting.

Specific Setting

So far, adverse events generated within non-interventional studies have to be documented and approved manually by the investigator in an electronic data-capturing system as soon as the investigator becomes aware of the event. These reports are then forwarded to pharmacovigilance for further processing, e.g., causality checks.

But what happens if the heart rate of study patients is continuously measured by smart watches (classified as medical devices)? The heart rate might be influenced by medicinal products, physical activity, or other factors. Any occurring bradycardia or tachycardia has to be reported as an adverse event and has to be assessed regarding its origin in an end-to-end process.

These data, such as heart rate, atrial fibrillation episodes, and physical activity in general (including falls), are generated on an ongoing basis. So far, it is unclear how to use and handle these data.

Example: According to the current study standards derived from this information, the investigator does not become aware of these data or any potential ‘adverse event’. Furthermore, in a routine setting, the investigator will not be able to handle this information adequately for AE reporting and only in a limited way for treatment control.

Actual Challenge

Using existing data, find a way to identify adverse events and distinguish them from physiological reactions.

Possible challenge approaches might include the following components:

Screening heart rate, inhalation and activity data.
Identifying thresholds for the data generated by the medical device qualifying for AE-reporting.
Creating a machine-learning algorithm, testing and validating it.
Discussing the technical & regulatory solution with regulatory authorities.

2024-04-22T02:01:44+02:00

Background
Arjan, a 58-year-old teacher, has no symptoms but worries about his risk of prostate cancer due to his father’s history with the disease. His doctor recommends screening, but Arjan is unsure if he really needs it. Meanwhile, Blerim, a 65-year-old retired mechanic, was diagnosed at a late stage, despite having regular check-ups. Advances in AI and data science offer the potential to change these stories—by identifying high-risk individuals earlier and ensuring timely interventions.

Problem
Despite medical advancements, many men still face late diagnoses due to limited screening accessibility, resource constraints, and gaps in existing risk prediction models. Some models lack accuracy across diverse populations, while others fail to integrate smoothly into clinical workflows, making them less effective in real-world settings. A more inclusive, data-driven solution could help bridge these gaps and improve early detection efforts.

The Challenge
How can AI-driven risk prediction improve prostate cancer detection and prevention? Your task is to develop a robust and inclusive machine-learning model that analyzes patient data—such as genetic profiles, lifestyle habits, and clinical records—to identify individuals at risk. The model should prioritize accuracy, fairness, and usability while ensuring patient privacy. Can your solution seamlessly integrate into clinical settings, helping doctors make better-informed decisions? Design a tool that brings early detection within reach for more people, improving health outcomes one step at a time.