ERS 2024: Artificial Intelligence to Help Decision-Making in Primary Care
Speaker: Gabriel Garcia
Artificial intelligence (AI) is increasingly playing a role in primary care decision-making, offering advancements in diagnostics, patient monitoring, and disease management. AI applications have expanded into various areas, including diagnosis, treatment planning, and exacerbation detection, particularly for chronic conditions such as asthma. The growing integration of AI into everyday life—across industries including healthcare—indicates its potential to revolutionize medical practices. For diagnostic purposes, AI has been successfully employed in assessing asthma using electronic stethoscopes and spirometry data. AI algorithms can evaluate lung function and detect disease patterns earlier than traditional methods. AI tools have also been applied to primary care for analyzing radiographic images and even detecting lung cancer long before symptoms arise.
AI is also significantly improving the identification of exacerbation risks. Machine learning models can assess constellations of factors such as obesity, tobacco use, low lung function, female sex, and poor asthma control to predict exacerbations in asthmatic patients. These predictive models, based on large-scale clinical trial data, have shown high correlation with real-life outcomes, enabling tailored treatment strategies. AI has also played a role in drug discovery, notably contributing to the rapid development of vaccines like the COVID-19 vaccine by Moderna, which was designed in just two days. AI’s core components include big data, fast computing, and machine learning models. In 2023, humanity generated over 140 zettabytes of data, a quantity that can be leveraged by AI to perform predictive analyses. Advanced computing systems, capable of performing trillions of calculations per second, further enhance AI's capacity to analyze vast datasets, providing precise predictions and aiding in clinical decision-making.
AI is defined as the simulation of human intelligence by machines, including processes such as learning, reasoning, and self-correction. In healthcare, AI functions primarily by utilizing large datasets and applying predictive mathematical models, offering significant benefits in clinical workflows. Despite its benefits, the transparency of AI's decision-making process—often referred to as the "black box" problem—remains a challenge, as clinicians must understand and trust the conclusions reached by AI models. Current developments in AI include applications in diagnosing atrial fibrillation through video photography and AI. While these innovations may seem improbable, they are grounded in emerging technologies that continue to push the boundaries of clinical diagnostics. Finally, AI is at the forefront of a technological singularity, a point at which the advancement of AI could outpace human understanding. The healthcare field must carefully navigate the integration of AI, ensuring that the tools used are reliable, ethically sound, and beneficial for improving patient outcomes.
The application of AI in healthcare raises critical questions about its role in evidence-based medical practice. While AI can support knowledge acquisition and decision-making, the primary concern lies in its ability to adhere to clinical evidence. Healthcare professionals, including physicians and nurses, rely on robust evidence to guide patient care and treatment decisions. Therefore, any tool used in clinical practice must demonstrate a clear alignment with this evidence-based approach. One of the limitations with AI models such as ChatGPT is the lack of explicit reference to the degree of evidence behind its responses. In healthcare, it is essential for AI tools to provide information grounded in validated clinical data, yet current AI outputs often fail to show the hierarchy of evidence, offering generalized answers without clarity on the supporting evidence level.
Without this transparency, AI may not be suitable for making clinical decisions. For example, when asked to recommend asthma management guidelines, an AI tool might list options without accurately distinguishing between guideline documents and consensus statements. In one case, the Global Initiative for Asthma (GINA) was incorrectly labeled as a guideline, when it is, in fact, an evidence-based strategy document. This highlights the need for AI tools to differentiate between evidence types and consistently deliver reliable medical information. AI has the potential to improve healthcare services by providing timely, personalized information that enhances decision-making. Studies, such as those published in JAMA, have demonstrated AI's capacity to improve efficiency, accuracy, and the personalization of patient care. However, it is critical to remember that AI should not replace professional clinical judgment. Instead, it should serve as an adjunct, helping healthcare providers make more informed decisions while upholding the principles of evidence-based medicine.
In conclusion, AI offers promising advancements in healthcare, but its current limitations, particularly in its handling of clinical evidence, must be addressed. There may be a future where physicians are divided into two groups—those who embrace AI as an aid in patient care and those who prefer traditional decision-making methods. Regardless, AI should always remain a support tool, complementing, not replacing, the expertise of healthcare professionals.
European Respiratory Society Congress 2024, 7–11 September, Vienna, Austria.


