Generalist medical AI reimbursement challenges and opportunities

Generalist medical AI reimbursement challenges and opportunities

Generalist AI: from narrow to multifunctional tools

The emergence of generalist AI marks a fundamental shift in medical technology. Unlike traditional narrow AI systems designed for specific tasks, generalist AI is characterized by the ability to handle a variety of tasks with minimal or no reliance on specialized training or specially labeled data designed for each specific task2. These systems can seamlessly handle diverse clinical responsibilities while providing transparent reasoning for their recommendations.

Traditional narrow AI systems perform well at isolated tasks – detecting intracranial hemorrhage in head CTs, flagging lung nodules in chest X-rays, or identifying diabetic retinopathy in fundus photographs3. In contrast, early research on generalist AI systems explores their potential to analyze an entire chest CT holistically: simultaneously detecting pulmonary emboli, characterizing lung nodules, identifying coronary calcification, assessing bone density, flagging incidental findings, and generating a comprehensive, actionable clinician note4.

This shift has broad implications for healthcare delivery. Generalist AI systems are bridging critical gaps in care delivery by reducing cognitive load on healthcare providers, minimizing handoffs between specialized systems, and streamlining complex clinical workflows2.

However, just as the Health Information Technology for Economic and Clinical Health Act (HITECH Act) in the United States provided necessary financial incentives and penalties for widespread electronic medical record system adoption5, similar comprehensive policy initiatives may accelerate the integration of these transformative generalist AI systems into standard clinical practice.

Challenges in reimbursement for generalist AI

Healthcare payment systems have typically been structured around discrete, clearly definable medical services and episodes of care6. While the following discussion primarily focuses on the United States healthcare system, many of the underlying challenges and considerations may apply to other health systems (Fig. 1).

Fig. 1: Overview of generalist AI in medicine.
Generalist medical AI reimbursement challenges and opportunities

Example core functions and applications of generalist medical AI systems across clinical care, research, administration, and patient engagement domains.

Full size image

A physician performs a specific procedure, orders a particular test, or provides a defined consultation – each with its own billing code or associated payment structure6. But generalist AI fundamentally challenges this paradigm. While narrow AI tools like automated diabetic retinopathy screening have clear, single-purpose functions that may fit neatly into existing billing codes, generalist AI may defy more simplistic categorization.

For example, a narrow AI tool might excel at measuring tumor size in breast cancer imaging – a discrete task with a clear billing pathway. In contrast, a generalist AI system may simultaneously analyze the mammogram, correlate it with pathology slides, integrate genomic test results, and synthesize clinical trial data to recommend personalized treatment strategies. This has potential to create a fundamental reimbursement challenge: current frameworks might struggle to capture the value of systems that not only perform multiple discrete tasks but generate novel insights through their interaction. The current framework might attempt to value each task individually, but this misses both the efficiency gains and benefits that come from multi-step integrated analysis1. Generalist AI systems can address tasks beyond their initial training, making it challenging to establish payment structures that reflect their expanding and dynamic capabilities. This is further complicated by difficulties in determining appropriate compensation for physician oversight and interaction as these systems span assistive to autonomous roles, and by fundamental challenges in characterizing software costs under current practice expense methodologies7.

An additional issue in the reimbursement of AI systems is that, as these technologies become increasingly agentic (autonomous), they perform tasks traditionally undertaken or overseen by providers. At present, this issue is mitigated by the necessity of human oversight for most AI applications, allowing reimbursement to remain tied to physician effort and workload and supervision or reasoning. However, as AI systems approach greater autonomy, the development of novel payment models may be required that consider not only the value of the AI’s contribution but also the evolving role of physicians7.

Rethinking reimbursement models: toward a future framework

There is demand to explore new payment frameworks that can capture both the comprehensive capabilities and dynamic nature of generalist AI systems.

Value-based reimbursement models might offer one path forward, moving beyond the limitations of traditional fee-for-service approaches. A generalist AI system managing chronic conditions like diabetes could coordinate comprehensive care by monitoring continuous glucose data, adjusting medication recommendations, analyzing routine labs, and scheduling follow-up care. Rather than billing for each interaction, reimbursement would align with meaningful outcomes such as reduced hospitalizations or improved A1C levels. This approach is particularly well-suited for generalist AI systems whose functions can’t be appropriately divided into discrete services and whose value derives from their comprehensive, integrated capabilities8.

Existing care coordination reimbursement codes in part exemplify the healthcare system’s progression toward value-based payment models, potentially offering a framework for integrating multi-capable AI systems9. In oncology care, generalist AI might simultaneously perform triage, detection, diagnosis, and report generation for a single case. Similar to how current care coordination codes bundle multiple services under a single payment for chronic disease management, comprehensive AI coordination codes that capture the full scope of AI’s multi-modal analysis and care recommendations. For example, an AI-enabled cancer care coordination billing code, potentially scaled by levels of service or technology provided, could cover the AI system’s integrated analysis across imaging, pathology, and genomics, rather than requiring separate codes for each component. This approach may better reflect the value of AI’s comprehensive analysis while simplifying billing processes8.

Drawing from the AMA’s AI taxonomy, another generalist AI reimbursement framework could scale reimbursement based on the level of AI involvement in clinical care10. Payment structures could reflect the progression from assistive functions (like flagging abnormalities for radiologist review), to augmentative capabilities (providing prognostic insights), to autonomous operations (generating complete reports with minimal oversight). This tiered approach would align payment with the increasing sophistication and autonomy of AI systems while providing clear pathways for valuing new capabilities as they emerge.

Broader implications and ethical considerations

As new reimbursement models are designed, ensuring equitable access and robust oversight to advance the use of quality AI tools becomes paramount.

Equitable access and patient-centered implementation

The emergence of generalist AI systems presents both opportunities and challenges for achieving equitable healthcare access. While larger health systems with substantial resources can invest in multiple specialized and narrow AI tools from several vendors, financial constraints may prevent smaller or resourced-limited healthcare organizations from acquiring such diverse AI portfolios, potentially creating disparities in their clinical capabilities11. Generalist AI systems, through their capacity to handle multiple tasks via a single platform, thus present an opportunity to redistribute technological capabilities more evenly across healthcare settings. However, achieving this equitable distribution may be further facilitated by payment frameworks that address both overconcentration of AI capabilities in well-resourced institutions and underutilization in resource-limited settings. Strategic policy interventions, such as targeted government subsidies similar to electronic health record adoption programs, may facilitate more equitable distribution of these technologies, and on an individual level, reduce healthcare disparities12.

Quality standards and oversight

Reimbursement frameworks will require mechanisms to ensure both ongoing oversight and predictable pricing stability. Unlike traditional medical devices, generalist AI systems require continuous evaluation across diverse populations and clinical scenarios, necessitating regular assessment of performance, outcomes, and cost-effectiveness. Reimbursement frameworks should incentivize systems that demonstrably improve care quality while minimizing excessive costs. Similarly, oversight mechanisms must balance innovation with accountability, ensuring that reimbursement adapts to ever-changing capabilities of the underlying technologies while preventing excessive costs and a misallocation of resources.

Conclusion

The advent of generalist AI in medicine presents both an unprecedented opportunity and a complex challenge for some healthcare reimbursement systems. Success will require collaborative effort among policymakers, clinicians, and AI developers to create frameworks that promote innovation while ensuring equitable access. As health systems navigate this transition, goals must be calibrated to develop reimbursement models that recognize the transformative potential of generalist AI while maintaining the highest standards of patient care and system sustainability.

Related Articles

A hybrid multi model artificial intelligence approach for glaucoma screening using fundus images

Glaucoma, a leading cause of blindness, requires accurate early detection. We present an AI-based Glaucoma Screening (AI-GS) network comprising six lightweight deep learning models (total size: 110 MB) that analyze fundus images to identify early structural signs such as optic disc cupping, hemorrhages, and nerve fiber layer defects. The segmentation of the optic cup and disc closely matches that of expert ophthalmologists. AI-GS achieved a sensitivity of 0.9352 (95% CI 0.9277–0.9435) at 95% specificity. In real-world testing, sensitivity dropped to 0.5652 (95% CI 0.5218–0.6058) at ~0.9376 specificity (95% CI 0.9174–0.9562) for the standalone binary glaucoma classification model, whereas the full AI-GS network maintained higher sensitivity (0.8053, 95% CI 0.7704–0.8382) with good specificity (0.9112, 95% CI 0.8887–0.9356). The sub-models in AI-GS, with enhanced capabilities in detecting early glaucoma-related structural changes, drive these improvements. With low computational demands and tunable detection parameters, AI-GS promises widespread glaucoma screening, portable device integration, and improved understanding of disease progression.

From video to vital signs: a new method for contactless multichannel seismocardiography

Seismocardiography (SCG) is a technique that non-invasively measures the chest wall’s local vibrations caused by the heart’s mechanical activity. Traditionally, SCG signals have been recorded using accelerometers placed at a single location on the chest wall. This study presents an innovative, cost-effective SCG method that utilizes standard smartphone videos to capture data from multiple chest locations. The analysis of vibrations from multiple points can offer a more thorough understanding of the heart’s mechanical activity compared to signals obtained solely from a single chest location. Our approach employs computer vision and deep learning techniques to extract and improve the resolution of multichannel SCG maps obtained by video capture of chest movement. We attached a grid of patterned stickers to the chest surface and recorded videos of chest movements during different respiratory phases. Using a deep learning-based object detector and a template tracking method, we tracked the stickers across video frames and extracted the corresponding SCG signals from sticker displacements. We also developed a robust algorithm to estimate heart rate (HR) from these chest videos and identify the optimal chest location for HR estimation. The method was tested on 28 chest videos captured from 14 healthy participants. The results demonstrated that our method effectively extracted multichannel SCG maps and enhanced their resolution with a mean squared error of 0.1078 and 0.0418 for right-to-left and head-to-foot SCG signals, respectively. We observed intersubject chest vibration patterns corresponding to cardiac events including opening and closure of the heart valves. Moreover, our algorithm accurately estimated HR from 1968 SCG signals extracted from the videos compared to the gold-standard HR measured from each subject’s electrocardiogram (bias ± 1.96 SD = 0.04 ± 2.14 bpm; r = 0.99, p < 0.001). The findings from this study underscore the potential of our approach in developing a cardiac monitoring tool using a smartphone that would be widely accessible to the general public and might provide more timely detection of diseases.

Segment Anything for Microscopy

Accurate segmentation of objects in microscopy images remains a bottleneck for many researchers despite the number of tools developed for this purpose. Here, we present Segment Anything for Microscopy (μSAM), a tool for segmentation and tracking in multidimensional microscopy data. It is based on Segment Anything, a vision foundation model for image segmentation. We extend it by fine-tuning generalist models for light and electron microscopy that clearly improve segmentation quality for a wide range of imaging conditions. We also implement interactive and automatic segmentation in a napari plugin that can speed up diverse segmentation tasks and provides a unified solution for microscopy annotation across different microscopy modalities. Our work constitutes the application of vision foundation models in microscopy, laying the groundwork for solving image analysis tasks in this domain with a small set of powerful deep learning models.

Cellpose3: one-click image restoration for improved cellular segmentation

Generalist methods for cellular segmentation have good out-of-the-box performance on a variety of image types; however, existing methods struggle for images that are degraded by noise, blurring or undersampling, all of which are common in microscopy. We focused the development of Cellpose3 on addressing these cases and here we demonstrate substantial out-of-the-box gains in segmentation and image quality for noisy, blurry and undersampled images. Unlike previous approaches that train models to restore pixel values, we trained Cellpose3 to output images that are well segmented by a generalist segmentation model, while maintaining perceptual similarity to the target images. Furthermore, we trained the restoration models on a large, varied collection of datasets, thus ensuring good generalization to user images. We provide these tools as ‘one-click’ buttons inside the graphical interface of Cellpose as well as in the Cellpose API.

Inhibition of sympathetic tone via hypothalamic descending pathway propagates glucocorticoid-induced endothelial impairment and osteonecrosis of the femoral head

Osteonecrosis of the femoral head (ONFH) is a common complication of glucocorticoid (GC) therapy. Recent advances demonstrate that sympathetic nerves regulate bone homeostasis, and GCs lower the sympathetic tone. Here, we show that the dramatically decreased sympathetic tone is closely associated with the pathogenesis of GC-induced ONFH. GCs activate the glucocorticoid receptor (GR) but hinder the activation of the mineralocorticoid receptor (MR) on neurons in the hypothalamic paraventricular nucleus (PVN). This disrupts the balance of corticosteroid receptors (GR/MR) and subsequently reduces the sympathetic outflow in the PVN. Vascular endothelial cells rapidly react to inhibition of sympathetic tone by provoking endothelial apoptosis in adult male mice treated with methylprednisolone (MPS) daily for 3 days, and we find substantially reduced H-type vessels in the femoral heads of MPS-treated ONFH mice. Importantly, treatment with a GR inhibitor (RU486) in the PVN promotes the activation of MR and rebalances the ratio of GR and MR, thus effectively boosting sympathetic outflow, as shown by an increase in tyrosine hydroxylase expression in both the PVN and the sympathetic postganglionic neurons and an increase in norepinephrine levels in both the serum and bone marrow of the femoral head of MPS-treated mice. Rebalancing the corticosteroid receptors mitigates GC-induced endothelial impairment and ONFH and promotes angiogenesis coupled with osteogenesis in the femoral head, while these effects are abolished by chemical sympathectomy with 6-OHDA or adrenergic receptor-β2 (Adrb2) knockout. Furthermore, activating Adrb2 signaling in vivo is sufficient to rescue the GC-induced ONFH phenotype. Mechanistically, norepinephrine increases the expression of the key glycolytic gene 6-phosphofructo-2-kinase/fructose-2,6-bisphosphatase 3 (PFKFB3) via Adrb2-cyclic AMP response element-binding protein (CREB) signaling. Endothelial-specific overexpression of PFKFB3 attenuates endothelial impairment and prevents severe osteonecrosis in MPS-treated Adrb2 knockout mice. Thus, GC inhibits sympathetic tone via the hypothalamic descending pathway, which, in turn, acts as a mediator of GC-induced ONFH.

Responses

Your email address will not be published. Required fields are marked *