The AI-Enabled Health History (AEHH): how generative AI and whole person health will refine the art of history taking in clinical practice
In medicine, getting to know a patient has always been as much an art as a science. This practice, the ‘art of history taking,’ is the cornerstone of exceptional care — it’s about finding the right balance between being thorough and discerning, all while building a therapeutic relationship with the patient.
The 20th century brought with it an astounding surge of medical advancements. Specializations emerged, laboratories and diagnostic testing became more sophisticated, and our knowledge of medicine grew exponentially. All these changes meant more tools at our disposal to diagnose and treat disease. Yet, with more data to consume for clinical decision-making, we also saw challenges arise. The reality of today’s healthcare system (particularly in the West) is an intricate paradox: we benefit from medical advancements but struggle with the fragmentation of care and a primary care system under immense pressure.
As a result, we have become experts in understanding and treating disease but distant from understanding the individual person and the many upstream factors that influence their health.
Historically, a skilled clinician had to have a deep understanding of the patient, serving as the diagnostician and care provider. Yet now, amidst the rapid diagnostic speed and the bounty of data provided by lab technologies, the role of astute clinical history taking finds itself at odds with technology. It’s at this crossroads that we face the future: the potential of generative AI and its implications for high-quality care delivery.
The rise of whole person health
The past decade has seen a growing fascination with the concept of whole person health (WPH). According to the National Center for Complementary and Integrative Health (NCCIH), WPH is defined as a comprehensive framework that acknowledges the multifaceted nature of human well-being. It involves looking at the whole person, not just individual organs or body systems, and can take into account a variety of factors that can either promote health or disease (such as their financial situation or what kind of food they have access to). This goes by many names, including holistic health, biopsychosocial health, or more broadly, the incorporation of social determinants of health (SDOH). Our comprehension of these health-impacting factors has certainly advanced over the last century. Simultaneously, the growing popularity of whole person health revives the age-old practice of good history taking, illustrating shared principles. Medical history taking traditionally focuses on gathering information relevant to a patient’s medical condition, while WPH expands upon this concept by considering the multidimensional aspects of an individual’s well-being.
What are these multidimensional aspects and how do we use WPH in clinical practice? Let’s start by looking at how it’s measured. Measuring WPH involves a detailed assessment that goes beyond traditional biomedical indicators (such as blood pressure control and weight). For example, the Riverside University Health System published the Whole PERSON Health Score (WPHS) assessment tool, which consists of 28 questions across six domains of health – Physical Health, Emotional Health, Resource Utilization, Socioeconomics, Ownership, Nutrition and Lifestyle. With there being an absence of a more detailed definition of WPH, many other questionnaires and instruments exist that fall generally into this category, such as WHODAS 2.0 and the SF-36, as well as the ICHOM Overall Adult Health protocol.
These tools however are not part of the typical clinical practice, and rather exist in the realm of research, quality assurance or in specific models of care delivery where compensation necessitates them (such as value-based care). Reasons for their exclusion from clinical practice are numerous, but big factors include the added administrative labour associated with their use (both to patient and an already resource-stricken health care system) and a lack of training and clarity as to how they would be used in a clinical setting. Much of the validation that exists for the use of these tools is restricted to controlled research settings.
Challenges in Integrating WPH into Clinical Practice
Each of us is likely to be generating an unprecedented volume of granular data within many of these domains discussed in WPH. A perfect example is data coming from wearable devices or wellness apps such a sleep trackers. This information is usually not available to clinicians. Electronic Medical Records (EMRs) represent the primary interface to healthcare data in the majority of community-based care delivery, serving as a repository of essential information about a person’s health. However, when it comes to many of the domains associated with WPH (such as financial health and resource utilization), traditional EMRs typically have not been required to have the room for this data in the necessary formats to make analytical and functional use.
Consider the domain of physical activity. While most EMRs may have a designated field to record physical activity, the available options are often limited to either a free text field or single select options such as low, medium, or high intensity, or by the type of activity performed. This overly simplistic approach fails to capture the rich and granular data that individuals now generate through wearable devices and fitness trackers. Wearable technology allows for the collection of detailed information about every step, movement, duration, intensity, and even physiological responses associated with physical activity (such as heart rate variability), and the number of linked wearable devices worldwide reached 1.1 billion in 2022. Importantly, being able to leverage this information would also introduce a path to automating it’s incorporation into a broader measure of health.
Our social lives have a significant impact on our overall well-being and offer crucial insights, yet the EMR typically lacks a comprehensive representation of this domain. Social media platforms and other digital channels offer a wealth of information about our social interactions, preferences, and emotional states. Studies have shown that changes in an individual’s song preferences or social media posts can provide insights into social satisfaction, contentment, and even loneliness. However, incorporating this nuanced information into the EMR poses challenges due to the unstructured nature of social media data, the need for sophisticated interpretation algorithms, and again, the need for there to be validated frameworks for how these can be used in clinical practice.
Emergence of generative AI
To bridge this gap between the extensive data available from various WPH domains and the limitations of traditional EMRs, the maturation of the domain of Artificial Intelligence (AI) holds great promise. In particular, the introduction of large language models (LLMs), a form of AI trained on vast amounts of data that can execute a variety of language functions, has led to the emergence of the ability to summarize information en masse into contextually appropriate formats. The capabilities of LLMs can be fine-tuned to process and analyze data from a variety of sources that were previously not effectively represented in the EMR. For instance, AI algorithms can interpret and extract meaningful patterns and insights from detailed physical activity data gathered through wearable devices, enabling a more accurate assessment of an individual’s exercise habits and overall fitness. Similarly, such tools could analyze and summarize the vast activity and “digital exhaust” that results from social media use (which has shown promise in providing valuable insights on their mental and emotional health), financial data, health insurance information, and so on.
Needless to say, the mere act of accessing and even considering linking such diverse data sources to derive conclusions regarding an individual’s health raises a variety of ethical considerations that need to be evaluated.
Before integrating AI within healthcare systems, medical ethics principles such as autonomy (patient’s right to self-care and self-determination), beneficence (acting in the best interests of the patient), non-maleficence (do no harm), and justice (reducing social inequality) should all be considered. Knowledge on the safe and effective use of this evolving technology must be incorporated into medical education for both new trainees as well as those in practice. Balancing the potential benefits of comprehensive information with privacy concerns and the potential for unintended consequences is a critical aspect that needs to be addressed. Ultimately, it should be the individual that decides which streams of information should be incorporated into their health profile on the basis of being appropriately informed of its implications.
A Vision for the AI-enabled Health History (AEHH)
WPH enabled by generative AI offers dual benefits. It not only enhances a physician’s capabilities but also enriches the diagnostic process through the incorporation of novel insights. This is particularly relevant as artificial intelligence continues to revolutionize clinical decision-making. Eric Brynjolfsson from the Stanford Digital Economy Lab spoke about the Turing Trap, which refers to the limitations of building AI that mimics human intelligence, which he terms as human-like AI (HLAI). Instead of striving to create systems focused on HLAI that automate human activities, Brynjolfsson argues that the real potential lies in building AI systems that surpass human capabilities and enable tasks that humans cannot perform in isolation.
In this context, we introduce the idea of the “AI-enabled Health History” or AEHH. This concept encapsulates the process of collecting and analyzing large amounts of otherwise uninterpretable health data associated with an individual’s holistic health and presenting it in a clinically pertinent and human consumable format.
Through the AEHH, leveraging advances in generative AI, healthcare professionals could access and effectively summarize vast amounts of information that were previously unavailable, under-utilized or in unusable formats. This could lead to vital insights, which may radically improve patient care.
Rather than concentrating the discussion on “automating intake” and “replacing physician history taking”, we redirect the narrative towards pushing the boundaries of possibility. We achieve this by merging the best of human abilities with the extended capabilities offered by artificial intelligence. The physician in this equation thus becomes defined beyond being a diagnostic entity, and further develops into an empathic expert, posing the ability to differentiate between the “red herrings” and “golden nuggets” in a patient history, validating the information at hand and exercising their expertise, experience, and ethical considerations in a shared decision-making process with the patient.
As clinical decision-making continues to evolve in both art and science, the rise of advanced artificial intelligence holds immense potential. It can catalyze improvements in history taking and data analysis in healthcare. WPH can serve as a framework to guide the modeling and design of how we leverage AI in clinical practice to enable more personalized and effective healthcare. On the part of the clinician, the value of human touch, in both its therapeutic and judgmental capacities, remains a vital cornerstone. The AEHH represents one way in which we can use AI to expand the capacities of healthcare providers. It isn’t just about doing the best we can – it’s about redefining the limits of what’s possible in the pursuit of health.
by Sukhman Tamber and Puneet Seth
