News & Comment

AI holds the potential to transform healthcare, promising improvements in patient care. Yet, realizing this potential is hampered by over-reliance on limited datasets and a lack of transparency in validation processes. To overcome these obstacles, we advocate the creation of a detailed registry for AI algorithms. This registry would document the development, training, and validation of AI models, ensuring scientific integrity and transparency. Additionally, it would serve as a platform for peer review and ethical oversight. By bridging the gap between scientific validation and regulatory approval, such as by the FDA, we aim to enhance the integrity and trustworthiness of AI applications in healthcare.

Digital health technologies (DHTs) can transform neurological assessments, improving quality and continuity of care. In the United States, the Food & Drug Administration (FDA) oversees the safety and efficacy of these technologies, employing a detailed regulatory process that classifies devices based on risk and requires rigorous review and post-market surveillance. Following FDA approval, DHTs enter the Current Procedural Terminology, Relative Value Scale Update Committee, and Centers for Medicare & Medicaid Services coding and valuation processes leading to coverage and payment decisions. DHT adoption is challenged by rapid technologic advancements, an inconsistent evidence base, marketing discrepancies, ambiguous coding guidance, and variable health insurance coverage. Regulators, policymakers, and payers will need to develop better methods to evaluate these promising technologies and guide their deployment. This includes striking a balance between patient safety and clinical effectiveness versus promotion of innovation, especially as DHTs increasingly incorporate artificial intelligence. Data validity, cybersecurity, risk management, societal, and ethical responsibilities should be addressed. Regulatory advances can support adoption of these promising tools by ensuring DHTs are safe, effective, accessible, and equitable.

In recent years the intersection of wearable technologies and machine learning (ML) based deep learning (DL) approaches have highlighted their potential in sleep research. Yet, a recent study published in NPJ Digital Medicine highlights the generalization limitations of DL models in sleep-wake classification using actigraphy data. Here, this article discusses some of the challenges and opportunities presented by domain adaptation and self-supervised learning (SSL), innovative methodologies that use large-scale unlabeled data to bolster the generalizability of DL models in sleep assessment. These approaches not only improve sleep-wake classification but also hold promise for extending to more comprehensive sleep stage classification, potentially advancing the field of automated sleep assessment through efficient and user-friendly wearable monitoring systems.

Recent developments in large language models (LLMs) have unlocked opportunities for healthcare, from information synthesis to clinical decision support. These LLMs are not just capable of modeling language, but can also act as intelligent “agents” that interact with stakeholders in open-ended conversations and even influence clinical decision-making. Rather than relying on benchmarks that measure a model’s ability to process clinical data or answer standardized test questions, LLM agents can be modeled in high-fidelity simulations of clinical settings and should be assessed for their impact on clinical workflows. These evaluation frameworks, which we refer to as “Artificial Intelligence Structured Clinical Examinations” (“AI-SCE”), can draw from comparable technologies where machines operate with varying degrees of self-governance, such as self-driving cars, in dynamic environments with multiple stakeholders. Developing these robust, real-world clinical evaluations will be crucial towards deploying LLM agents in medical settings.

Generative AI is designed to create new content from trained parameters. Learning from large amounts of data, many of these models aim to simulate human conversation. Generative AI is being applied to many different sectors. Within healthcare there has been innovation specifically towards generative AI models trained on electronic medical record data. A recent review characterizes these models, their strengths, and weaknesses. Inspired by that work, we present our evaluation checklist for generative AI models applied to electronic medical records.

In the medical literature, promising results regarding accuracy of medical AI are presented as claims for its potential to increase efficiency. This elision of concepts is misleading and incorrect. First, the promise that AI will reduce human workload rests on a too narrow assessment of what constitutes workload in the first place. Human operators need new skills and deal with new responsibilities, these systems need an elaborate infrastructure and support system that all contribute to an increased amount of human work and short-term efficiency wins may become sources of long-term inefficiency. Second, for the realization of increased efficiency, the human-side of technology implementation is determinate. Human knowledge, competencies and trust can foster or undermine efficiency. We conclude that is important to remain conscious and critical about how we talk about expected benefits of AI, especially when referring to systemic changes based on single studies.

Boussina et al. recently evaluated a deep learning sepsis prediction model (COMPOSER) in a prospective before-and-after quasi-experimental study within two emergency departments at UC San Diego Health, tracking outcomes before and after deployment. Over the five-month implementation period, they reported a 17% relative reduction in in-hospital sepsis mortality and a 10% relative increase in sepsis bundle compliance. This editorial discusses the importance of shifting the focus towards evaluating clinically relevant outcomes, such as mortality reduction or quality-of-life improvements, when adopting artificial intelligence (AI) tools. We also explore the ecosystem vital for AI algorithms to succeed in the clinical setting, from interoperability standards and infrastructure to dashboards and action plans. Finally, we suggest that algorithms may eventually fail due to the human nature of healthcare, advocating for the need for continuous monitoring systems to ensure the adaptability of these tools in the ever-evolving healthcare landscape.

The utilization of artificial intelligence (AI) in diabetes care has focused on early intervention and treatment management. Notably, usage has expanded to predict an individual’s risk for developing type 2 diabetes. A scoping review of 40 studies by Mohsen et al. shows that while most studies used unimodal AI models, multimodal approaches were superior because they integrate multiple types of data. However, creating multimodal models and determining model performance are challenging tasks given the multi-factored nature of diabetes. For both unimodal and multimodal models, there are also concerns of bias with the lack of external validations and representation of race, age, and gender in training data. The barriers in data quality and evaluation standardization are ripe areas for developing new technologies, especially for entrepreneurs and innovators. Collaboration amongst providers, entrepreneurs, and researchers must be prioritized to ensure that AI in diabetes care is providing quality and equitable patient care.

We explore the evolving landscape of diagnostic artificial intelligence (AI) in dermatology, particularly focusing on deep learning models for a wide array of skin diseases beyond skin cancer. We critically analyze the current state of AI in dermatology, its potential in enhancing diagnostic accuracy, and the challenges it faces in terms of bias, applicability, and therapeutic recommendations.

Historically, the Centers for Medicare and Medicaid Services (CMS) has formed partnerships with select private sector entities, including large traditional hospital and health system networks, nursing homes, and payer groups. However, innovations from technology-enabled services companies and digital technology companies are uniquely poised to aid CMS in addressing key barriers toward advancing its mission of improving healthcare access and equity. There are four pivotal opportunity areas where partnerships with technology businesses and tools would enhance the work of CMS: (1) improving consumer awareness about CMS programs, (2) mitigating access gaps through virtual care programs, (3) streamlining the complexity of different payer plan models, and (4) using technology-enabled services to address social risk factors without imposing additional burdens on providers. We offer examples of digital and technology-enabled solutions that improve patient access to care and close equity gaps, as well as propose specific recommendations for CMS to advance and expand the reach and impact of these solutions. Namely, these recommendations include partnerships with private sector companies that can educate and support consumers about their benefits, the extension of telehealth reimbursement parity for virtual care solutions, allowing for cross-state licensure across plans and reimbursement for care coordination services that alleviate provider burden to screen and address patients’ social determinants of health needs. We argue that CMS has an imperative role in leveraging the innovations of technology-enabled services and digital health technologies to lower healthcare access barriers, mitigate provider burden, stimulate innovation, and close equity gaps at the patient, provider, and innovator levels.