Within established clinical workflows, with mechanisms for reviewing uncertain, discordant

The gap

Evidence profile

Sourced from the recommendations and conclusions and limitations of the source papers, classified as general, spanning 6 journals.

Supporting evidence — 6 representative gaps

• Comparison of direct anterior and posterior approaches in total hip arthroplasty within a mature surgical practice: A single surgeon analysis of complications and perioperative metrics (2026) · Journal of Musculoskeletal Surgery and Research · doi

  In the long-term, prospective studies are warranted to determine which surgical approach yields superior outcomes across these key metrics. Authors’ contributions: SWF: Conceptualized, performed formal analysis, conducted research, and wrote the final and initial draft; CJD: Conceptualized, conducted research, provided resources, oversight, and wrote the final and initial draft. All authors have critically reviewed and approved the final draft and are responsible for the manuscript’s content and similarity index. Ethical approval: The research/study was approved by the Institutional Review Board at the University of Rochester Department of Human Subjects Protection, number STUDY00010189, dated February 26, 2025. Declaration of patient consent: The authors certify that they have obtained all appropriate patient consent forms. In the form, the patient has given consent for clinical information to be reported in the journal. The patient understands that the patient’s names and initials will not be published and due efforts will be made to conceal their identity, but anonymity cannot be guaranteed.

  generalrecommendations
  Keywords: patient authors final draft consent conceptualized conducted wrote initial approved long term prospective warranted determine
• Risk and liability in the deployment of AI systems for surgery: a SAGES white paper (2026) · Surgical Endoscopy · doi

  To address the aforementioned risks associated with the implementation and utilization of AI systems, we recom- mend creating an AI Governance Blueprint to more clearly communicate and mitigate the potential harm to patients, clinicians, and institutions. how inpatient clinical pharmacists monitor the appropri- ate use of drugs in a hospital, ensuring that the technol- ogy is deployed correctly and its outputs are appropriately integrated into surgical decision-making [38]. While these measures may not solve the issue of liability when there is harm, they are important for decreasing the risk of clinical AI use in the first place.

  generalrecommendations
  Keywords: harm clinical address aforementioned risks associated implementation utilization systems recom mend creating governance blueprint clearly
• From Evaluation to Implementation: A Delphi Consensus Framework with Domain-Weighted Decision Models for Robotic Surgery Adoption (2026) · Healthcare · doi

  Future research should focus on multicenter validation, longitudinal evaluation, and refinement of the proposed weighting models and scoring thresholds across diverse health- care contexts. Consequently, its broader applicability remains to be established. By integrating per- spectives from clinical, operational, safety, engineering, procurement, financial, and admin- istrative stakeholders within a single decision-support framework, this study addresses an important gap in the literature, where existing approaches often focus on isolated aspects of robotic surgery evaluation rather than the broader institutional context.

  generalconclusions
  Keywords: focus evaluation broader future multicenter validation longitudinal refinement proposed weighting models scoring thresholds across diverse
• Artificial intelligence-driven therapeutics for disease modification in type 1 diabetes: a digital public health and clinical translation framework (2026) · Frontiers in Public Health · doi

  within established clinical workflows, with mechanisms for reviewing uncertain, discordant, or clinically implausible outputs limitations to clinicians and and for communicating model patients. Integration of privacy protection, fairness assessment, interpretability, reproducibility, and patient-centered governance throughout the AI lifecycle will be critical for ensuring that AI-enabled approaches improve T1D care while minimizing the risk of amplifying existing disparities.

  generalrecommendations
  Keywords: within established clinical mechanisms reviewing uncertain discordant clinically implausible outputs limitations clinicians communicating model patients
• Artificial intelligence in surgical decision-making across the perioperative continuum: a scoping review (2026) · Frontiers in Digital Health · doi

  Interpretability and transparency are also concerning in surgical decision-making. Surgeons must be able to explain and justify AI-supported and multidisciplinary teams. Black-box models may be unsuitable for high-stakes surgical decisions because the rationale underlying interpret. Several their predictions methodological reviews emphasize that interpretable models and clinically meaningful essential prerequisites for trust and adoption in surgical practice (1, 8). explanation methods is often difficult to patients are to Regulatory and governance considerations also influence the adoption of AI in surgical practice. Although regulatory approval has begun to emerge for selected applications, most AI systems reported literature remain at the preclinical or developmental stage (9). The need to distinguish between such tools that merely inform the clinicians and those that actively influence the choices that are made by clinicians is an initial regulatory deliberations (4, 9). that was highlighted in the surgical issue the in Prospective evaluation is another recurring theme in the AI– surgery is an literature. Although retrospective validation important first step, it alone is insufficient to establish clinical utility. Ultimately, AI-supported decisions must demonstrate improvements in patient outcomes, reduction of complications, workflow efficiency, or other meaningful clinical endpoints. Consequently, prospective real-world impact evaluations remain essential for clinical translation (6, 10). studies and Lastly, the ethical aspects such as bias, equity, and fairness are also becoming known as part of AI-supported decision-making. Patient equity is greatly affected by the ramifications of access to care, the time of intervention, and the resources to utilize during the postoperative period. When AI models are trained on lopsided or incomplete information, they may be able to support or further increase differences. Background reviews emphasize the importance of performing an audit of AI systems on demographic and clinical subgroups and applying ethical control and to implementation (8, 11). the process of model development In summary, the existing evidence highlights both the potential and the limitations of artificial intelligence (AI) in technical surgical decision-making. Although supporting capabilities have advanced rapidly, translation into clinically reliable and tools remains implementable decision-support uneven across perioperative settings. Much of the literature has focused on predictive performance, while less attention has been given to how AI influences real-world clinical decisions and implementation. Therefore, a scoping review is warranted to comprehensively map the current evidence on AI-supported surgical decision-making across the perioperative continuum, identify areas of maturity, and highlight barriers and gaps affecting clinical translation. To address these gaps, this scoping review was guided by four key research questions: (1) What types of AI applications are used across the perioperative phases of surgical care? (2) How does AI contribute to decision-making in preoperative, intraoperative, and postoperative settings? (3) How does the performance of AI-based traditional clinical or statistical approaches compare with methods? and (4) What are the main barriers to the clinical implementation of AI in surgical practice? To this end, this review aimed to map the existing literature on AI in surgical decision-making across the entire perioperative continuum, describe its current uses in various phases, the most important decisions aided by AI, and gaps/barriers to clinical translation.

  generalrecommendations
  Keywords: surgical clinical decision making supported decisions literature translation across perioperative models practice regulatory implementation review
• Artificial intelligence and robotic technologies redefining precision and personalization in orthopedic surgery: a narrative review (2026) · Frontiers in Bioengineering and Biotechnology · doi

  AI algorithms form a robust technical foundation for intelligent assistance in orthopedic surgery, yet their translation into clinical practice remains constrained by key technical limitations. Foremost among these is the “black-box” nature of many AI models, which compromises interpretability and, consequently, adoption and clinical trust (Blackman and Veerapen, 2025; Zheng et al., 2025). In orthopedic surgical settings, doctors need to understand the reasoning behind AI-derived predictions to assess the validity of proposed operative pathways or risk assessments. Yet the opacity of algorithmic decision-making, coupled with the inherent complexity of underlying computational processes, often limits a surgeon’s ability to evaluate the logic of outputs, thereby reducing confidence and hindering real-world application (Kumar et al., 2022; Ramkumar et al., 2022; Pai et al., 2024). In recent years, advances in explainable AI have begun to address this challenge. For example, heatmap-based visualization tools can illustrate the specific anatomical regions or features influencing model outputs, enabling surgeons to better interpret AI recommendations and fostering greater trust in their application (Akula et al., 2022; Mukhtorov et al., 2023). A central objective of digital orthopedics is to optimize resource utilization; however, current algorithmic frameworks often demand substantial human input for design, validation, and refinement. For instance, in one study aimed at developing an AI system capable of automatically detecting metacarpal fracture lines, the initial model failed to achieve satisfactory accuracy. Consequently, 10 specialist physicians engaged in a 3-month training process to iteratively teach the system, enabling it to acquire basic automated recognition capabilities (Kim et al., 2024). Beyond algorithms, the hardware platforms that enable AI applications, such as intelligent navigation platforms, surgical robots, and sensor-based devices, are critical in translating computational outputs into precise clinical actions. Their performance and reliability directly determine clinical value, yet several barriers remain. A major challenge lies in the absence of standardized interfaces for heterogeneous data sources, which constrains the efficiency of multimodal data integration. In formats between orthopedic procedures, variations electronic health systems, and biomechanical sensors compel developers to create bespoke dataconversion modules for each device, increasing system complexity, prolonging development cycles, and introducing risks of data transmission errors (Su and Pei, 2024). in data imaging (EHRs), records for soft their algorithms compensation Technical limitations in intraoperative adaptability further impede precision. While certain orthopedic robotic systems employ generative AI to construct three-dimensional skeletal models, tissue deformation still rely primarily on static preoperative data and lack the capacity to fully adapt to dynamic intraoperative load variations, leading to suboptimal feedback during procedures (Ryu et al., 2024; Zhang Z. et al., 2024). Furthermore, device incompatibility and limited system-level interoperability impede technological communication protocols and operating systems among devices hinder real-time data synchronization, complicate intraoperative AI deployment, and disrupt the seamless incorporation of AI into orthopedic workflows (Sherrod et al., 2023). In addition, the predominant dependence on cloud-based computing introduces network latency, which can delay robotic arm responses; in high-precision settings, such delays may heighten the risk of neurovascular injury and undermine the clinical feasibility of AI-assisted interventions evolution.

  generallimitationsevidence 5/5
  Keywords: orthopedic clinical system algorithms technical outputs based systems intraoperative intelligent limitations among models consequently trust

Questions about this gap