Artificial Intelligence in Value-Based Health Care

Abstract

Artificial intelligence (AI) presents new opportunities to advance value-based healthcare in orthopedic surgery through 3 potential mechanisms: agency, automation, and augmentation. AI may enhance patient agency through improved health literacy and remote monitoring while reducing costs through triage and reduction in specialist visits. In automation, AI optimizes operating room scheduling and streamlines administrative tasks, with documented cost savings and improved efficiency. For augmentation, AI has been shown to be accurate in diagnostic imaging interpretation and surgical planning, while enabling more precise outcome predictions and personalized treatment approaches. However, implementation faces substantial challenges, including resistance from healthcare professionals, technical barriers to data quality and privacy, and significant financial investments required for infrastructure. Success in healthcare AI integration requires careful attention to regulatory frameworks, data privacy, and clinical validation.

Introduction

Healthcare delivery is moving from traditional fee-for-service models to value-based healthcare (VBHC) models, which emphasize the delivery of high-quality care that maximizes patient benefit relative to cost [17,33]. This approach carries particular significance in orthopedic surgery, given the high costs and variable outcomes associated with specialty care and surgical interventions [3]. Yet to be effective, solutions to improve value in orthopedic surgery also need to contend with inequities in care access and the resultant healthcare disparities [13].

Strategies to improve value have involved policy, clinical pathways, and interventions developed amid limitations to data access and processing capacity. Artificial intelligence (AI) is enhancing the ability to take on these challenges, working in tandem with a rapid rise in computational firepower, access to open-source data, and advancements in computational models and techniques. In musculoskeletal health, the pace of development is being fueled by an expanding evidence base and greater acceptance by stakeholders in the field [31]. Traditional AI included machine learning (ML) and natural language processing, which enabled computers to learn from data patterns and understand human language. More recently, generative AI technologies, particularly large language models (LLMs), have emerged to create new insights from vast amounts of data.

These AI technologies can advance VBHC through 2 mechanisms. First, AI can be used to improve health outcomes by enhancing patient health literacy through improved patient education, delivering more personalized care through predictive analytics, and augmenting surgical care through data-driven planning and decision support [3,17]. Second, AI has the potential to decrease costs by optimizing operating room efficiency, reducing unnecessary specialty referrals through automated screening, decreasing administrative burden in documentation and prior authorizations, and preventing costly readmissions and emergency visits through improved patient monitoring and communication [37]. Orthopedic surgery presents a particularly compelling opportunity for AI implementation due to its heavy reliance on accurate diagnostic imaging interpretation, high procedural costs, and documented variability in care delivery and outcomes across different regions and providers [17].

The integration of AI into value-based orthopedic care represents a significant opportunity to improve both the quality and efficiency of surgical and nonsurgical musculoskeletal care. To elucidate the impact of AI’s contribution to VBHC in orthopedic surgery, this article examines implementation challenges and future directions while considering practical and ethical implications.

AI Applications in VBHC

AI is changing healthcare delivery through 3 domains: agency, automation, and augmentation.

Agency

AI applications can simultaneously enhance outcomes and reduce costs by improving patient agency. First, AI can enhance the patient experience by improving health education and literacy. Freely available AI dialogue platforms can simplify complex medical education materials; studies have shown the conversion of orthopedic materials from 12th- to 6th-grade reading levels while maintaining accuracy and clinical detail [22]. LLMs can also outperform medical experts in creating patient-centered summaries across multiple tasks, including radiology reports, patient questions, and clinical documentation, with summaries being superior or equivalent to those of a physician for completeness and correctness 81% of the time [39]. However, in up to 6% of cases, this model generated incorrect information that was not present in the input text, a phenomenon known as “hallucination.” Nevertheless, automated enhancement of patient education can significantly boost healthcare value by improving patient understanding and satisfaction without placing additional demands on limited provider time.

Second, AI enables personalized care through remote monitoring and wearable technologies. ML algorithms linked to wearable devices can track specific metrics such as home exercise compliance, daily step counts, knee range of motion, and opioid use after total knee arthroplasty, providing objective data to guide individualized postoperative care [30]. These AI-powered monitoring platforms enable customized treatment approaches at scale without incurring additional costs per patient, making personalized care both clinically effective and economically sustainable. AI-enabled activity tracking and visualization of these data can trigger healthy behaviors and improve mobility and joint health.

AI also has the potential to reduce costs while expanding access to value-based care models. AI algorithms can triage patients and reduce unnecessary specialty referrals, enabling more efficient resource allocation. Studies show AI systems achieve similar accuracy to specialists in multiple domains; for example, an AI respiratory triage system identified low-risk patients who did not require specialist evaluation, while an AI dermatology platform matched specialist-level accuracy in determining whether or not referral was necessary [10,18]. These successes in other medical domains suggest that similar AI applications could reduce costly orthopedic and musculoskeletal specialist visits. For musculoskeletal conditions, AI has been shown to enable substantial expansion of clinician capacity, with 1 study showing a 2.3-fold increase in the patient-to-clinician ratio without compromising outcomes [5].

Beyond improving individual patient care, AI chatbots—automated conversational systems that simulate human dialogue—optimize resource utilization by preventing complications and emergency visits. One study found that following arthroplasty, multilingual chatbots reduced readmissions from 8.3% to 0% and emergency department visits from 8% to 0.9% [34]. In a less complex use case, chatbot-based follow-up systems achieved comparable effectiveness to manual follow-up while reducing provider time by over 90% [7]. This evidence suggests AI tools can reduce costs by triaging care and preventing avoidable specialty visits and complications, thus supporting VBHC goals.

Finally, a growing number of AI chatbots are trained to engage people with mental health concerns [16]. Increasing access to support for stress, distress, and unhelpful thinking—factors that can affect comfort and capability in musculoskeletal health—can have a substantial impact on longer-term health-related outcomes.

Automation

While improving patient agency is a critical first step, efficient delivery of care is equally important. Operating room efficiency and resource utilization represent critical challenges in healthcare delivery. AI can optimize these complex systems through advanced predictive analytics. For example, traditional scheduling methods result in ~50% of cases running over time, leading to cascading delays and inefficient resource use [35]. AI-driven scheduling systems analyze multiple variables simultaneously—including case duration patterns, surgeon-specific factors, and procedure complexity—to generate more precise predictions. At 1 institution, implementation of ML scheduling reduced overtime rates by 21% while maintaining surgical volume, resulting in estimated cost savings of $469,000 over 3 years [2].

Beyond surgical scheduling, AI demonstrates significant potential to streamline administrative healthcare tasks through automation. Montage Health’s implementation of AI-powered prior authorization and claims adjudication automation achieved a 22% reduction in authorization work queue volume and processed over 5600 authorization status checks in 1 year [4]. The system saved their healthcare organization 300 staff hours/month while decreasing accounts receivable days by 13%. Studies show that AI-based systems can achieve authorization decision accuracy rates above 96% [25], enabling staff to focus on complex cases while routine checks are handled automatically. While AI shows promise in automating prior authorizations, it is important to note that these models still require human review for complex cases and must be carefully monitored to prevent algorithmic bias.

Building on these operational efficiencies, ambient AI scribes represent a solution for reducing physician documentation burden and burnout. A recent pilot study at Stanford University demonstrated that ambient AI scribes led to significant reductions in physician workload, with providers spending 0.57 fewer min/note on documentation and saving nearly 20 min/day in total electronic health record time [28]. The technology showed results in reducing burnout, with physicians reporting significant decreases in task load and work exhaustion on standardized scales [36]. By automating documentation tasks, physicians could potentially increase their patient panel size, improving access to care in a system facing increasing provider shortages. Beyond time savings, ambient AI scribes may enhance value by allowing for more time for physician-patient engagement during visits.

Augmentation

AI is increasingly augmenting clinical decision-making in orthopedic surgery through improved diagnostic accuracy and surgical planning. In musculoskeletal imaging, deep learning algorithms have demonstrated high accuracy in detecting fractures and abnormalities. A systematic review found that AI achieved near-perfect prediction (area under the curve [AUC] 0.95-1.0) for fracture detection across multiple anatomic sites, often outperforming both radiologists and orthopedic surgeons [24]. The clinical benefits of AI-augmented surgical planning were recently demonstrated in a randomized trial of total hip arthroplasty, in which AI-assisted planning led to significantly better Harris hip scores, shorter operative times (104.3 vs 134.3 minutes, P < .05), and reduced frequency of intraoperative adjustments [40].

AI can improve shared decision making between surgeons and patients [20]. Jayakumar et al conducted a randomized trial evaluating an AI-enabled patient decision aid that incorporated digital patient education, preference assessment, and person-specific predictions of relevant health outcomes. The AI system significantly improved decision quality scores, level of shared decision making, improved patient-reported outcome measurements, and patient satisfaction compared to digital educational materials alone [19].

ML algorithms have also demonstrated significant potential in predicting clinical outcomes and identifying high-risk patients. In total shoulder arthroplasty, ML techniques achieved 93% to 99% accuracy in predicting which patients would achieve minimally clinically important differences in outcomes (AUC 0.85-0.97) [23]. Similarly, in total joint arthroplasty, AI algorithms effectively predicted meaningful clinical improvements (AUC 0.78-0.89) [14]. These predictive capabilities enable more informed surgical decision-making by identifying patients at higher risk of complications before surgery. For example, ML demonstrated good discriminatory ability (AUC 0.80-0.84) in predicting survival outcomes for patients with extremity metastases [38]. This actionable intelligence allows surgeons to optimize perioperative protocols and engage in more informed shared decision-making discussions.

A recent landmark randomized controlled trial demonstrated that AI-enabled electrocardiography risk prediction tools could identify high-risk hospitalized patients and led to a significant reduction in 90-day mortality rates (3.6% vs 4.3%, hazard ratio [HR] = 0.83) compared to usual care. This mortality benefit was particularly pronounced in the high-risk group identified by AI (HR = 0.69), as it enabled more timely monitoring and interventions [27]. The trial provides compelling evidence that AI risk prediction when properly integrated into clinical workflows with clear actionable alerts, can help physicians identify vulnerable patients early and modify care pathways to improve outcomes.

The comprehensive impact of AI on VBHC is particularly evident in osteoporosis management, where AI enhances risk prediction, detection, and treatment optimization. Recent studies demonstrate that ML algorithms can accurately identify at-risk patients and predict fracture risk using routine clinical data, achieving high accuracy (AUC >0.90) in classifying bone health status [12]. AI-enhanced opportunistic screening using routine computed tomographic imaging has shown excellent accuracy in bone mineral density prediction (86%-96%) and osteoporosis classification (AUC 0.927-0.984), enabling identification of at-risk patients without requiring dedicated dual-energy X-ray absorptiometry scans [32]. In addition, AI models have demonstrated high accuracy (AUROC 0.96-0.99) in recommending personalized exercise protocols based on patient characteristics [12].

Challenges and Limitations

Despite promising applications, several significant barriers must be addressed for AI to be successfully implemented in orthopedic settings.

Human and Clinical Barriers

Healthcare organizations face significant resistance to implementing AI technologies in orthopedic clinical settings. Studies have shown that healthcare professionals often lack a comprehensive understanding of AI principles and express concerns about the potential consequences of widespread clinical AI use [8]. Resistance to technological change is particularly evident when providers do not fully understand the systems or are concerned about liability and responsibility for AI-assisted decisions [29]. Patient acceptance also remains a challenge, with studies indicating that only 50% of patients view AI in healthcare as an important opportunity, while 11% consider it a potential danger to their care [8].

There are also substantial financial barriers to implementing AI. Organizations must invest heavily in infrastructure modifications, software maintenance, hardware updates, and comprehensive staff training programs [11]. These combined human and financial barriers can significantly slow the adoption of AI technologies in orthopedic practices.

Technical barriers also present challenges to implementing AI in orthopedic settings. Data quality is particularly crucial for AI applications, as ML models require extensive, properly labeled datasets to achieve reliable performance. The “black box” nature of many AI algorithms creates additional complexity, as clinicians struggle to understand and trust systems that cannot clearly explain their decision-making processes [9]. Healthcare data often exists in disconnected silos with varying formats and standards, creating substantial interoperability barriers that affect AI model training and validation.

AI systems also face unique privacy challenges, as ML models can potentially expose sensitive patient information through inference attacks or model memorization [26]. Healthcare organizations must carefully balance model accuracy with privacy protection while ensuring AI systems maintain sufficient access to training data.

Future Directions and Guidelines

An analysis by Gartner provides a framework for evaluating AI applications based on both technical feasibility and the potential of value generation [15]. The most promising applications combine high feasibility with substantial clinical value (Fig. 1). Clinical documentation automation shows particularly strong potential, with AI systems effectively converting medical conversations into structured records while requiring minimal workflow changes. Predictive analysis, augmentation, and improving patient care navigation also emerge as high-value, implementable solutions that can overcome existing technical and regulatory barriers.

Fig. 1.

Rates the different use cases of generative AI in healthcare by their potential clinical value and technical feasibility.

Source: Gartner, Use-Case Prism: Generative AI for Healthcare Providers, 28 July 2023. GARTNER is a trademark of Gartner, Inc. and/or its affiliates.

The evolution of AI technology is accelerating these applications’ viability through several key advances. Deep learning models are becoming more sophisticated in processing medical data, with new architectures better suited for integration into clinical workflows. The ability to generate novel medical data complements limited real-world datasets, enabling more robust model training [6]. Recent improvements in step-by-step reasoning capabilities (demonstrated by new medical AI models achieving over 90% accuracy in knowledge tests) show how domain-specific training can enhance model performance [6]. Such advances may make AI tools more reliable and easier to implement in healthcare systems.

Nevertheless, the implementation of AI must follow legal and ethical guidelines. For example, the American Academy of Orthopedic Surgeons (AAOS) has issued guidelines that include promoting health equity throughout the AI lifecycle, ensuring algorithm transparency and explainability, and engaging patients in AI development; the AAOS statement also emphasizes defining clear algorithmic fairness standards and establishing accountability frameworks for equity in outcomes [1].

When AI augments rather than replaces clinical judgment, it must operate within existing clinician-patient relationship frameworks, maintaining professional standards of competency, trust, and autonomy [41]. Currently, most consumer-facing AI applications lack Health Insurance Portability and Accountability Act compliance and adequate professional oversight. Digital watermarks and clear disclosures about AI involvement are essential, while clinicians retain ultimate responsibility and liability, necessitating a thorough understanding of AI capabilities and limitations [21]. Implementation requires compliance with regulatory frameworks to ensure AI enhances rather than replaces clinical expertise in musculoskeletal care, to establish data privacy protections and informed consent protocols, and to pay special attention to protecting vulnerable populations [20].

Conclusion

AI is advancing VBHC in orthopedic surgery through 3 mechanisms: improving patient agency, automating routine tasks, and augmenting clinical decision-making. Mistakes by autonomous vehicles built on AI have led to fatalities. This example of flawed AI decision-making leading to critical injury serves as a stark reminder that rapid technology deployment without adequate safety protocols can have devastating consequences. As with any innovative technology, AI is likely following the typical hype cycle; after initial enthusiasm, a more nuanced understanding of capabilities and limitations emerges. Success in healthcare AI integration will require careful attention to regulatory frameworks, data privacy, and clinical validation while learning from implementation challenges in other industries. With appropriate guidelines, stakeholder engagement, and measured deployment, AI can serve as a powerful tool in achieving the quadruple aim of better outcomes, lower costs, improved patient experience, and enhanced clinician satisfaction.