Artificial intelligence has made its way to the front page of every industry, and healthcare is no exception. While much of the focus has been on automating operational tasks, attempts are also being made to impact consumer health—from access to personalized recommendations directly from Apple Health data to enterprise solutions like Thrive Global and OpenAI’s recently launched health coach, which aims to deliver “hyper-personalized” behavioral change. Research has shown that personalization can improve adherence by 52 percent in areas such as cardiovascular care and upwards of 13 percent in the metabolic health space. Given the recent frenzy around weight loss medications like Ozempic, metabolic health has become an area that’s particularly interesting to watch. The laundry list of core issues that have been exposed—including supply chain management, drug symptoms, adherence and lack of metabolic health specialists—has created a promising opportunity for new technology adoption.
Behavioral change—an aspect of care that has been heavily relied on in weight management—can be hard to improve. The burden of enacting behavioral change falls on the patients, and adhering to standard doctor recommendations has historically been challenging, in part due to serious biases overweight patients face in clinical settings. Large language models (LLMs) have significantly improved the accessibility of technology that allows for personalization and effectiveness of behavioral change. The constant availability and ability to mimic human reasoning make integrating such technology into existing experiences, like health coaching, easier. Training, fine-tuning and guardrails can also help these technologies avoid stigmatizing language that prevents effective treatment for many individuals.
It’s important to remember that the clinical outcomes for behavior change may not be significant in many cases. For example, some behavior change programs for weight loss or diabetes management have been shown to result in only 3 percent weight loss over a 52-week period. Research so far has been conflicting. Early data has shown that patients can lose a similar percent of their weight in under a third of the time using fully conversational A.I., while another study indicated that patients interacting with a health coach and A.I. lost nearly double the weight of using A.I. alone.
In part due to the challenging nature and lack of efficacy of behavioral health strategies alone, A.I.’s greatest potential within metabolic health is clinical decision support. LLMs can now ingest large quantities of personalized medical data to help providers tailor patient treatment plans, including the right behavioral and clinical recommendations. LLMs can help recognize patterns that may be difficult to detect, like responsiveness to a GLP-1 medication or symptoms thereof, suggest novel treatments that providers may not be aware of and support continuous monitoring and timely interventions by independently analyzing real-time data. For example, an LLM can recommend that a patient take their medication when their blood sugar spikes after a meal based on insights into that specific patient’s health and behavior. Predictive analytics powered by the amounts of data an LLM can ingest are also an area of growth for this category. LLMs can help providers take action around which patients are at the highest risk of complications for certain comorbidities or prescriptions, driving specific care recommendations.
For chronic conditions like diabetes, the industry often treats them in a vacuum without analyzing the patient’s entire medical history. A patient’s history of asthma, sleep apnea, anxiety and longitudinal lab metrics will all affect how their body responds to treatment, but these are often overlooked by non-specialists. A.I. enables providers to analyze a patient’s complete medical history, genetic information, lifestyle factors and current health status to create highly personalized treatment plans. This individualized approach ensures that treatments are tailored to each patient’s unique needs, improving outcomes and reducing the risk of adverse effects. These same capabilities can also help providers address social determinants of health (SDOH) when treating patients, realizing and recommending treatment that is more accessible. As a result, providers can treat metabolic health issues alongside comorbidities in a holistic setting without bottlenecking the care they’re seeking.
A.I. is helping care become more effective through personalization and enabling qualified providers to prioritize care and alleviate the burdens that plague them. With less than 10,000 specialists across the country treating 110 million people living with obesity, the more time physicians spend on administrative work like completing prior authorization forms, supply chain issues and research, the less face time they have with patients. The spike in demand for Ozempic-type medicines raised time constraint concerns as the supply chain became a greater issue. In today’s world, A.I. tools can easily solve issues like tracking down prescriptions, which physicians have been spending unnecessary hours on instead of direct patient interaction.
As A.I. continues to be integrated across industries, there’s so much noise for consumers to comb through, which can impact their faith in A.I.-backed care. The reality is metabolic healthcare doesn’t need another nutrition A.I. tool—it needs A.I. that can provide value-based care recommendations to both patient and physician.
But this type of A.I. doesn’t come without challenges. On the integration side, health systems often struggle to fold A.I. tools into their existing electronic health records. Regarding privacy and security, many solutions feed data to OpenAI, Gemini or others without realizing it. The tech has become so “easy” for non-technical people to build solutions that the classic privacy and security guardrails are being forgotten. While the FDA regulates technology designed to treat or diagnose, less clinical solutions often slide by in a gray area of not providing treatment through their behavioral health applications, posing a risk to patients. Despite being an isolated example, Google (GOOGL)’s AI Overview recommendation to eat rocks demonstrates the risks that A.I. technologies could have in clinical settings. Further, education can be a huge undertaking for clinicians, who often don’t understand how this tech should be implemented in their day-to-day work—let alone how to navigate the regulations around it or lack thereof.
These challenges are a work in progress, and we’re only just scratching the surface of how A.I. could impact the quality and efficiency of care. The current landscape of A.I. and its potential in this space is why I joined knownwell. This comprehensive metabolic health company has already begun integrating A.I. into its clinical care decisions for patients managing weight issues through its recent acquisition of Alfie Health, an online precision medical and obesity management clinic that uses A.I. to generate evidence-based recommendations for weight management. By utilizing A.I. to recommend long-term treatment methods and keep track of patient history, healthcare providers can ultimately enhance the patient experience.
A.I. can be daunting, but its potential in industries like healthcare widely outweighs the skepticism. A.I. has the power to expand access to timely, personalized care, so doctors can focus on saving lives instead of drowning in data.
