# Crossing the Chasm of Quantified Health - Towards LLM-enabled Digital Twins _Date: 29 Aug 2023 , Tags:_ #LLM #digital-twins #digital-health #quantified-self #quantified-health #AI ![[Quantified Health.png|380]] ## Introduction > Benjamin Franklin, the eighteenth-century polymath and one of the Founding Fathers of the United States, kept track of how he spent his time and he upheld the moral virtues he had defined for himself. Beyond this apparently simple yet powerful act of self-tracking lies the future of health care. Indeed, the wide adoption of digital health technologies is changing how we manage our health. Since the 2010s individuals have started taking a more and more active role in taking care of their health. If the chasm of self-tracking has been crossed, with an estimated 21% of the US population using smartwatches or wearable fitness trackers, providing personalized insights at an individual level is a non-trivial task. It requires seamlessly, collecting, integrating, and analyzing various data sources (e.g. social, behavioral, EMR) to be used in clinical practice and delivering back to the individual the right information at the right time, through the right intervention. This process has already begun with the growing availability of chronic care and home care solutions, also propelled by the pandemic that skyrocketed the adoption of telemedicine solutions. As the amount of health data about ourselves increases, we are getting more resolution to (fully) defining a person from a quantitative standpoint (aka Digital Twin). Digital Twins, coming with Large Language Models, can democratize predictive access to health care by providing daily support to individuals and making predictive and preventive medicine a reality. However, the traditional challenges in healthcare remain (e.g. privacy, inclusion). ## The Promise of Digital Health: Proactive, Predictive and Preventive Medicine The variety of data generated and analyzed from individuals, health systems, and populations has the potential to transform the quality of healthcare delivery and improve outcomes. Already a decade ago, Daniel Kraft outlined in an influential article how digital health is set to profoundly change health care as we conceive it. Fig. 1. The evolution of healthcare ![[Reactive-to-Preventive-Medicine.png]] Source: re-adapted from Kraft (2021) Technology changes the place where healthcare happens. The effect of these new technologies will enable a paradigm shift in the healthcare system: moving away from a delivery of care that is reactive (focused on treating diseases when they appear and symptoms can no longer be ignored), to proactive (the person is empowered to self-manage his/her health through technology), and finally predictive and preventive (Ai systems/sensors will predict and alert the person of the risk of disease before the first symptoms appear). Early signs of this transition can be seen with the pandemic. Driven by necessity, necessity healthcare systems started adopting digital health solutions on a wide scale. While the adoption of digital health technologies increased, considerations on the real-world applicability of the data collected remain a topic of discussion. Fig. 2: The health data <> clinical utility gap ![[Health Data - Clinical utility gap.png]] Source : Chorghade et al., 2016 Indeed, new technologies can generate data that is not fit for purpose (e.g. does not follow the V3 framework of sensor-based data collection), meaning that the mere opportunity to collect a new but exceedingly narrow, window on biological complexity does not start from an unmet medical need, translating into an increasing gap between the data collected and clinical utility. Digital Health solutions have also become an integral part of many individuals, using sensors and apps (among more than 350k) to inform their health decisions. If the clinical utility, linked to established medical science, of many of these solutions has yet to be confirmed, what is underway in the healthcare system is a blurring between wellness and clinical care. ## The Missing Mile: From Quantified Self to Quantified Health In this scenario, while technology alone cannot solve the problems of the healthcare system ("technological solutionism"), it certainly represents an opportunity to improve it. As we move from reactive to proactive medicine, individuals take an active role in monitoring and trying to improve their health, via the use of sensors (e.g. fitness bands) or self-reporting it (e.g. mood). We see these activities more and more common. As an example, Apple is introducing mood tracking and journaling in iOS 17. The habit of measuring things and self-experiment is not new and history is rich in examples, for instance, Isaac Newton in an attempt to understand how the eyes work almost blinded himself by staring at the reflection of the sun. > Self-tracking refers to the practice of collecting and analyzing data about oneself, typically related to health and wellness, and has the ultimate goal of providing insights to empower people to make informed decisions about their health. The chasm now lies in using data to get insights. The self-tracking lifecycle of data comprises four phases: (i.) collect, (ii.) synthesize, (iii.) analyze, and (iv.) reflect. Fig. 3: The self-tracking lifecycle of data ![[Self-tracking data information process.png]] Source : Watson, 2013 How we move from the act of collecting data to getting personalized insights will be the focus of healthcare systems for the next years and decades. Note: As these activities are conducted at an individual level, many considerations that are the foundations of digital clinical measures in clinical research & standard care are not fully addressed. i. Collect This phase is self-explanatory at first glance, but it brings considerations that have a profound impact on the subsequent phases. - What & Why: A variable of interest (e.g. number of steps) is selected. - How & the Way is collected: For instance, is it self-reported, or collected via a sensor? The selection of which device cannot be overlooked (e.g. security, usability). ii. Synthesize: The next point is all about integrating the data, combining or merging the data that come from different sources. It may be as straightforward as using the device maker's app, relying on other solutions (e.g. Apple's built-in Health App), or for the more data-savvy, exporting the data for further analysis. iii. Analyze According to the French philosopher Michel Foucault, self-awareness is not an end in itself but a means to living better. Analyzing is all about looking at the data. In this phase, challenges can arise from defining a normal (average) and healthy value. What is normal for a person may differ for another. Note that a healthy value does not always overlap with normal. Many digital health apps come with self-aggregated metrics (e.g. average number of steps in a week). iv. Reflect As Eric Topol puts it in the opening of his book Deep Medicine: >"Life can only be understood backwards, but it must be lived forwards" -SØREN KIERKEGAARD Reflection should be seen as a way to discover more about an individual's own health. Indeed, the applicability of self-tracking information may not have clinical utility, as "Meaning" for physicians is rooted in established medical science (double-blind, randomized, controlled clinical trial). Hence, data could become another element in clinical decision-making but not the most important one. Additionally, information alone is not sufficient to change one's individual behavior, and it needs support, coaching, or advice. Moving from quantified self to quantified health therefore requires models to effectively integrate disparate data sources into the clinical practice. So that the clinician can access individuals' data in real-time and reach out proactively thanks to an algorithm noticing alarming values. Whereas this is not technically possible, for instance, due to a lack of resources (e.g. shortage of healthcare personnel), the use of LLM-enabled digital twins could make healthcare more accessible and equitable by providing actionable insights to individuals. ## A glimpse into the future: LLM-enabled Digital Twins The core idea of self-tracking is that patients can ask and answer their own questions with their data. Data collected by individuals can be empowering, manageable, and informative but also overwhelming and confusing. Helping individuals to navigate their data in a supportive manner is therefore paramount. Here's where LLMs and Digital twins come into play. A digital twin, as envisioned by Eric Topol, moves from the n of 1 to the n of billions, and nearly fully defining an individual. By knowing more about an individual and aggregating at a population level we can better adapt treatments to particular individuals. Additionally, Digital Twins can be a supportive tool to guide patients whenever they need it, by providing precision wellness, prevention, and diagnostics. As an example, by training an LLM with established medical knowledge and feeding as a prompt the person's medical history + self-tracked data (including a personal diary), an individual can use the LLM as a reflection tool to get personalized insights. In other terms, we apply medical trials and cohort-based knowledge used to train the LLM and personalize it at an individual level. What is true for an individual may not be may not be true for others, especially in self-experimentation (“n-of-1” studies). While fully defining a person from a quantitive standpoint (e.g. including all the genomic, lab, and medical history data) may still be a utopian concept, an incremental approach will likely be the way digital twins are adopted. If we stick to this definition, a DIY approach to getting a personal digital twin is already technically possible. As an example feeding an individual's diary data + some self-reported metrics in an LLM can already uncover interesting insights. [In this experiment](https://every.to/chain-of-thought/does-gpt-4-know-me-better-than-my-girlfriend), a user showed that all LLM performed better than his girlfriend and his mother when answering questions about his personality. ## Conclusions >The purpose of data-driven health innovation should be “to make the consumer the CEO of his own health" -Vinod Khosla Digital Twins could improve doctor-patient relationships by facilitating more informed decision-making, providing personalized insights to patients 365/7, predicting the most effective therapy, and becoming an effective prevention tool (an area for which it is hard to build a sustainable digital health business). The use of LLM-enabled Digital Twins in healthcare comes not without ethical, legal, and social challenges. To mention a few: regulations should ensure LLMs do not cause harm or compromise the data and privacy of medical professionals and individuals; clinical validity, hallucinations, and bias are significant key challenges to address. It is also crucial to ensure transparency in data collection, consent, and control. To bridge the chasm of Quantified Health, Digital Twins must also offer a user-friendly experience and implement behavior design strategies (e.g. gamification, actionable recommendations) to enact change and need to be effectively integrated with healthcare providers. If healthcare systems will properly tackle these challenges, LLM-enabled Digital Twin could pave the way for high-performance medicine and make healthcare more effective, accessible, and personalized. Meanwhile, individuals can already self-experiment a "mini" digital twin by feeding an LLM with their data (with caution). ## References - Apple. “Apple Provides Powerful Insights into New Areas of Health,” 2023. https://www.apple.com/newsroom/2023/06/apple-provides-powerful-insights-into-new-areas-of-health/. - Béchard, Deni Ellis. “Body Count,” 2021. https://stanfordmag.org/contents/body-count. - Cluitmans, Matthijs. “How Happy Is Your Virtual Brain? The Exciting Intersection of Digital Twins and Large Language Models,” 2023. https://www.transformingmed.tech/p/how-happy-is-your-virtual-brain-the. - “Consumer Health Apps and Digital Health Tools Proliferate, Improving Quality and Health Outcomes for Patients, Says New Report from IQVIA Institute,” 2021. https://www.iqvia.com/newsroom/2021/07/consumer-health-apps-and-digital-health-tools-proliferate-improving-quality-and-health-outcomes-for. - DiMe Society. “The Playbook: Digital Clinical Measures,” 2023. https://playbook.dimesociety.org/. - Dow Schüll, Natasha. “The Folly of Technological Solutionism: An Interview with Evgeny Morozov,” 2013. http://www.publicbooks.org/interviews/the-folly-of-technological-solutionism-an-interview-with-evgeny-morozov. - Holko, Michelle, Tamara R. Litwin, Fatima Munoz, Katrina I. Theisz, Linda Salgin, Nancy Piper Jenks, Beverly W. Holmes, Pamelia Watson-McGee, Eboni Winford, and Yashoda Sharma. “Wearable Fitness Tracker Use in Federally Qualified Health Center Patients: Strategies to Improve the Health of All of Us Using Digital Health Devices.” _Npj Digital Medicine_ 5, no. 1 (April 25, 2022): 53. https://doi.org/10.1038/s41746-022-00593-x. - Kraft, Daniel. “Paradigm Shifting: From ‘Sick’ Care to ‘Health’ Care,” 2014. https://www.linkedin.com/pulse/20140520203233-320304-paradigm-shifting-from-sick-care-to-health-care/. - “The Future of Health & Medicine: Looking Ten Years Back & Ten Forward...,” 2021. https://www.linkedin.com/pulse/future-health-medicine-looking-ten-years-back-forward-kraft-md. - LaBerge, Laura. “How COVID-19 Has Pushed Companies over the Technology Tipping Point—and Transformed Business Forever.” McKinsey, 2023. https://www.mckinsey.com/capabilities/strategy-and-corporate-finance/our-insights/how-covid-19-has-pushed-companies-over-the-technology-tipping-point-and-transformed-business-forever. - Meskó, Bertalan, and Eric J. Topol. “The Imperative for Regulatory Oversight of Large Language Models (or Generative AI) in Healthcare.” _Npj Digital Medicine_ 6, no. 1 (July 6, 2023): 120. https://doi.org/10.1038/s41746-023-00873-0. - Moore, Geoffrey A. _Crossing the Chasm: Marketing and Selling Disruptive Products to Mainstream Customers_. Third edition. New York, NY: HarperBusiness, an imprint of HarperCollins Publishers, 2014. - Mukund S Chorghade. “Drug Discovery.” _Drug Discovery World Winter 2016/17_, 2017. - Neff, Gina, and Dawn Nafus. _Self-Tracking_. The MIT Press Essential Knowledge Series. Cambridge, Massachusetts: The MIT Press, 2016. - Shipper, Dan. “Does GPT-4 Know Me Better Than My Girlfriend?,” 2023. https://every.to/chain-of-thought/does-gpt-4-know-me-better-than-my-girlfriend - Topol, Eric J. “High-Performance Medicine: The Convergence of Human and Artificial Intelligence.” _Nature Medicine_ 25, no. 1 (January 2019): 44–56. https://doi.org/10.1038/s41591-018-0300-7. - The Creative Destruction of Medicine: How the Digital Revolution Will Create Better Health Care_. 1st pbk. ed. New York: Basic Books, 2013. - Watson, Sara. “Living with Data: Personal Data Uses of the Quantified Self,” 2013. http://www.scribd.com/doc/172418320/Living-With-Data-Personal-Data-Uses-of-the-Quantified-Self. - Wolf, Gary. “‘The Data-Driven Life.,’” 2010. http://www.nytimes.com/2010/05/02/magazine/02self-measurement-t.html. --- _Copyright © 2024 - Adriano Fontanari_