The rise of AI in health care: transforming the future

Abstract

As with every technological revolution in health care, the advent of artificial intelligence (AI) has been met with a mix of enthusiasm and fear. Central to these concerns is the extent to which medical decisions are being transferred to a technology operating on incompletely understood logic and the perceived loss of connection between health care practitioners and patients. However, when viewed from the perspective of potential rather than hazard, AI can catalyze a much-needed transformation in medical care. Particularly in the current landscape, where health care systems are facing critical shortages, compromising system sustainability and equity, the urgent need for a technological revolution is paramount. In the United States, 52% of a clinician’s time is dedicated to documentation tasks, increasing disillusionment and burnout.1 In addition, delays in accessing care stem from escalating demands for health care services amidst worsening provider shortages, leading to waiting lists for appointments, treatments, or procedures, as well as screening, diagnosis, and preventive care. With the advent of AI, a field of computer science focused on developing systems capable of performing tasks that normally require human intelligence, the paradigm is likely to change. Many types of AI, such as generative AI, machine learning, and other specialized forms, such as natural language processing and computer vision, have been successfully applied in health care. AI systems can efficiently allocate resources and schedule appointments by analyzing patient data and predicting care demand, including the likelihood of “no-shows.” In surgical care, AI technologies are starting to revolutionize orthopedic surgical planning and navigational accuracy toward more precise, safe, and efficient settings.2 In postsurgical, AI can be used to predict compilations (eg, lumbar spine fusion).3 These are just a few examples where AI applications can have profound rippling effects across health systems. Taking AI more deeply into clinical care, though, can prove more challenging. Here, we firmly believe that clinicians and engineers need to work hand-in-hand to identify points of stress and/or failure and deploy AI to solve them. We argue that many improvements can be done at the operational or administrative level, with profound second-order and third-order effects that result from downstream choices about how to use the efficiencies gained. AI can also significantly promote health equity by, first and foremost, enhancing health care accessibility, especially in underserved areas and populations. In a direct testimonial to this potential are AI-powered care systems that remotely allow for musculoskeletal rehabilitation4 as well as addressing other health domains such as the prevention of vision loss and blindness (Lions Outback Vision), dental assessment (Pearlii), and high-risk pregnancy detection.5 AI can ease consultation, diagnosis, and definition of treatment plans, without geographical or scheduling constraints, making high-quality care accessible to all. Although biased AI models may exacerbate existing disparities, when their design follows “human-in-the-loop” approaches and recommended development best practices, fair and ethical AI can be attained.6 These best practices should include the use of large, representative real-world samples for data acquisition, integration of bias-mitigation techniques during model development (such as constraints, regularization terms, and fair representation learning), provision of explainability techniques to clarify decision-making processes, and model surveillance after algorithm deployment.6 In addition, AI analytical power outpaces traditional analyses,7 being able to uncover patterns or underlying mechanisms that contribute to health care access disparities or low-value care interventions, which are more prevalent among marginalized populations.8 Leveraging health records and claims, AI systems can predict individuals at risk of chronic conditions or unnecessary surgeries.9 Technology has the power to enable early interventions, preventing care escalation and improving clinical outcomes while reducing costs for patients and health care systems. Including social determinants of health data in AI systems has proven to remarkably increase model performance.10 All of the above demonstrates that the power of AI can be harnessed toward health equity. Perhaps more controversial is the potential of AI to humanize health care via stronger therapeutic alliances between patients and their providers and patient-centered approaches. AI should be designed to complement, rather than replace, the clinician’s critical reasoning and emotional intelligence, ensuring that automation preserves the human connection and empathy central to the clinician-patient relationship—elements vital for effective diagnosis and care. While it’s common to argue that technology dehumanizes health care, AI actually offers a true opportunity to do the opposite, especially since the massification of modern computing. Past experiences have shown that the effect of a given technology is dependent on the right use case. For instance, while smartphones have been accused of creating social isolation, they actually allow individuals to better communicate with their loved ones. Similarly, AI humanizing potential can be explored through 2 concurring aspects, as illustrated below. AI algorithms can help streamline administrative tasks such as documentation, scheduling, and billing in musculoskeletal management (eg, MetaDoc.AI), allowing clinicians to free up time to focus on patients.8 AI systems can also rapidly interpret vast amounts of complex medical data for clinical decision support, enhancing triage and diagnostic accuracy (eg, RevelAI Health). Eliminating routine clerical tasks and enabling faster, more accurate decision-making can enhance care delivery efficiency and quality. It also encourages human-centered care, allowing clinicians to build trust, communicate effectively, and show empathy—key to achieving treatment compliance and patient satisfaction. This is where another aspect comes into play. With the inception of generative AI, communication in the clinical setting has also started to undergo a transformation. Humanization of interactions through AI can be embedded within clinical workflows to enable remote care delivery and reduce clinicians’ documentation burden. In stressful or emotional situations, AI can guide clinicians to use empathetic language, fostering compassionate interactions. In postoperative or postpartum situations, for example, remote human-like conversations have proven effective in promptly responding to a patient’s queries and providing reassurance from the comfort of their home.2 We believe it is only a matter of time until more evidence on this subject amounts. A third aspect in which AI may improve care and reduce costs involves patient selection for expensive or risky therapies. For example, whereas 50% of a large patient population may improve with a pain-relieving procedure like decompression surgery or arthroplasty, this success rate is actually composed of patients extremely likely to benefit (eg, those with an acute, progressive MRI-confirmed radiculopathy with no psychosocial issues, opioid use, or secondary gain) and those unlikely to benefit (eg, patients with central sensitization, degenerative noncompressive etiology, psychosocial issues, work-related disability, opioid use), alongside those who “could” benefit. There are dozens of clinical and demographic variables that may affect outcomes, and in the future, AI may be able to synthesize them to precisely estimate the benefits (eg, 69% + 9% or 18% + 4%) instead of a generic 50% “coin flip” estimate. Although the above examples may seem obvious, there are several known challenges and unknown obstacles along the way. However, we strongly believe that this potential needs to be embraced, as the stakes were never higher. Executing it will, however, require a nondogmatic discussion about the role of AI within health care. If anything, the rise of AI should lead us to question the evidence and doctrine behind current practices, their true need, and the efficiency of their delivery. During this process, we must not fall into the trap of examining only the care that is being provided, but also—and especially—the care that is not. By adopting an overly cautious position that hinders AI exploration, we risk missing out on AI’s transformative benefits. When built with an emphasis on ethical considerations and within safety boundaries, AI risks can be effectively mitigated. Health care needs change, and AI offers just the venue to do this. As health care providers and administrators dedicated to improving the lives of patients, we must grasp this opportunity as it arises, as it can be done right. Conflicts of interest disclosures Dr V.Y. is reported to be an employee of Sword Health, and Dr V.B. is the CEO of Sword Health. In addition, both authors reported holding equity in Sword Health. Dr S.P.C. is an independent scientific and clinical consultant who received adviser honorarium from Sword Health. Declaration of generative AI and AI-assisted technologies in the writing process No AI tools/services were used during the preparation of this work.

Ähnliche Arbeiten