Christian MünzenmayerIn the hopefully not too distant future, artificial intelligence (AI) and sensors will be present along the entire patient journey, both helping to gather and analyze individual patient data. At the same time, personalized medicine will benefit healthcare and care management in general. It allows healthcare professionals to make more informed decisions about prevention, diagnosis, and treatment, which in turn leads to improved patient outcomes and potentially reduces healthcare costs by optimizing treatment plans.
While data analytics and AI have an impact way beyond precision medicine, they are ultimately the key contributing factors to tailoring medical treatment and interventions to individual patients. So, how far along are we in this process and what are some of the challenges that lie ahead?
Tomorrow’s healthcare is data-driven.
Collecting robust and reliable medical data is the basis for informed decision-making throughout the entire medical care process. Especially when it comes to prevention, therapy, and aftercare, the use of wearable devices and built-in sensors for measuring bio-signals opens up new possibilities. Wearables, such as smart watches and rings from various commercial vendors, are widely accepted and worn by an increasing number of health-aware individuals. For higher accuracy, chest straps or smart textiles, such as CardioTEXTIL, a shirt with integrated ECG electrodes allowing for a 6-channel ECG, can be used. By integrating sensors into everyday objects, such as car seats, furniture or even into a toilet seat, a continuous stream of real-time health data becomes available. This data provides insights into an individual’s well-being and can assist both healthcare providers in their decisions and individuals in making informed choices about lifestyle and daily behavior.
Collecting data (especially high-quality, medical-grade data) is an essential component, but only one part of the equation. When recording bio-signals from wearables, reliable data interpretation is another prerequisite for making data-driven decisions – especially if the data is not as high-quality as we would need them to be in the healthcare context.
While measuring bio-signals through wearables is convenient for users, the presence of artifacts or noise in the data poses a significant challenge. These artifacts can originate, for instance, from incorrect sensor placement or excessive movement, leading to so-called motion artifacts. To address this challenge, robust algorithms can be implemented for analyzing bio-signals from mobile measurement systems, ensuring more reliable and accurate results – and bringing us to the next key factor of tomorrow’s personalized healthcare: AI.
Tomorrow’s healthcare is AI-based.
AI drives the digital transformation of health care in numerous application scenarios: Robotic systems assist healthcare professionals, AI is found in medical image analysis and diagnostics, and in developing personalized treatment plans. With regard to the patient journey, machine learning and AI help analyze large datasets to identify patterns and correlations, to build predictive models and assign these patterns to specific symptoms or diseases. A recent study in Science [1] shows, though, that these models have limits when predicting treatment outcomes (the study focused on the predicted success of different antipsychotic medications for patients with schizophrenia), depending on the data they were trained on. This touches a sore spot head-on: Where and how do we get relevant medical data?
CardioTEXTIL, for example, was developed to detect cardiac arrhythmias, a cardiological condition that affects about 2 % of the population in industrial countries and that can lead to heart failure or strokes. Long-term monitoring increases the probability of detecting arrythmias as these events occur sporadically and often go unnoticed because they are neither recorded by a 24-hour ECG (electrocardiogram) nor noted by the patient (cf., e.g., Reiffel et al. 2017) [3]. While sensor technology makes it possible to detect arrhythmias at an early stage thanks to the mobile multi-channel ECG of CardioTEXTIL, the challenge lies in the interpretation of the data: Most ECG algorithms are still rule-based and often restricted to certain sub-groups of the condition. This restriction leads to a low specificity of the algorithm, meaning it is less effective at correctly identifying non-target items (i.e., arrhythmias that are not yet included in the sub-groups). Unless the data is manually annotated by physicians (which is costly), arrythmias hence remain unseen. Therefore, we have developed state-of-the-art, high-accuracy algorithms that are able to detect more than 20 different types of arrythmia and that have been extensively tested on a database of over 43.000 ECGs (Oppelt et al. 2020) [2].
Moving forward, we therefore do not only need to develop robust algorithms for specific medical conditions but also need to develop standards and protocols for the underlying training data and for evaluating the robustness, safety, and security of these AI-based detection systems – if they are ever to be used in across-the-board healthcare contexts. Together with colleagues from various renowned institutions, we develop quality criteria and reference metrics for health data and AI, that help to ensure reliability of systems as well as compliance with the EU AI Act.
Tomorrow’s healthcare is personalized.
The biggest potential of wearables, integrated sensors, and personalized apps is that they can be used to monitor health parameters over time – before health issues develop. After early detection and diagnosis, they can play an integral part in managing health conditions, hopefully preventing a worsening of existing conditions, and in rehabilitation.
Breast cancer care is one example that highlights how personalized medicine can benefit patients. Precision diagnostics (through advanced imaging technologies, genetic testing, or bio-marker analysis) are the foundation for targeted therapies. Compared to standardized therapies, these personalized cancer therapies can be more effective and have fewer side effects but still can last over several years. The patient’s compliance is, therefore, a precondition for successful treatment. Innovative and – most importantly – convenient monitoring and communication tools help in that process. For this approach to work along the entire patient journey and across medical disciplines, though, we need
- tools that facilitate communication between patients, physicians, and health care providers (both for inpatient and outpatient care)
- data integration into various health platforms and workflows across different healthcare providers and disciplines
- standards and protocols to ensure data protection and regulatory requirements, and
- an infrastructure to use the generated real-world data to support science and R&D.
Establishing stringent data protection measures is a complex problem to solve in and of itself (here’s more on the issue of how to store sensitive patient data and how the GAIA-X project is working to find secure solutions for Europe). An equally complex challenge lies in integrating these tools and infrastructures into healthcare routine. May it be intelligent data analysis, may it be sensor technologies – we might excel at developing the technologies and transformative solutions for tomorrow’s health care, but we still have a journey ahead of us when it comes to implementing the entire potential of digital personalized healthcare into everyday practices. But there’s no doubt about it: The future of healthcare is digital.
References
[1] Chekroud, A. M., Hawrilenko, M., Loho, H., Bondar, J., Gueorguieva, R., Hasan, A., Kambeitz, J., Corlett, P. R., Koutsouleris, N., Krumholz, H. M., Krystal, J. H. & Paulus, M. (2024). Illusory generalizability of clinical prediction models. Science, 383(6679), 164–167. https://doi.org/10.1126/science.adg8538 [2] Oppelt, M., Riehl, M., Kemeth, F. & Steffan, J. (2020). Combining Scatter Transform and Deep Neural Networks for Multilabel ECG Signal Classification. Computing in Cardiology. https://doi.org/10.22489/cinc.2020.133 [3] Reiffel, J. A., Verma, A., Kowey, P. R., Halperin, J. L., Gersh, B. J., Wachter, R., Pouliot, E. & Ziegler, P. D. (2017). Incidence of Previously Undiagnosed Atrial Fibrillation Using Insertable Cardiac Monitors in a High-Risk Population. JAMA Cardiology, 2(10), 1120. 10.1001/jamacardio.2017.3180Copyright (cover image): iStock — Liuzishan
Grit Nickel
Markus Jechow
Dr. rer. nat. Christian Münzenmayer
Add comment