Real-World Data Needs Real-World Doctors: When Automation Advances Faster Than Clinical Workflow.

Real-World Data Needs Real-World Doctors: When Automation Advances Faster Than Clinical Workflow - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice editorial J Particip Med . 2026 May 29;18:e97972. doi: 10.2196/97972 Real-World Data Needs Real-World Doctors: When Automation Advances Faster Than Clinical Workflow Amy Price 1, 2, 3, ✉ , Christine Von Raesfeld 4 Editors: Stephanie Law , Tiffany Leung Author information Article notes Copyright and License information 1 Department of Community and Family Medicine (CFMED), Dartmouth-Hitchcock Clinics, Lebanon, NH, United States 2 Journal of Participatory Medicine, JMIR Publications, Toronto, ON, Canada 3 Center for Global Health, Colorado School of Public Health, University of Colorado Anschutz Medical Campus, 13001 E 17th Pl # B119, Aurora, CO, 80045, United States, 1 5618437372 4 The Light Collective, Eugene, OR, United States ✉ Amy Price, MA, MSc, DPhil, Center for Global Health, Colorado School of Public Health, University of Colorado Anschutz Medical Campus, 13001 E 17th Pl # B119, Aurora, CO, 80045, United States, 1 5618437372; dr.amyprice@gmail.com Received 2026 Apr 11; Revised 2026 May 2; Accepted 2026 May 13; Collection date 2026. Copyright © Amy Price, Christine Von Raesfeld. Originally published in Journal of Participatory Medicine (https://jopm.jmir.org) This is an open-access article distributed under the terms of the Creative Commons Attribution License ( https://creativecommons.org/licenses/by/4.0/ ), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in Journal of Participatory Medicine, is properly cited. The complete bibliographic information, a link to the original publication on https://jopm.jmir.org , as well as this copyright and license information must be included. PMC Copyright notice PMCID: PMC13221120  PMID: 42214035 Abstract Health care is entering an era of unprecedented detection. Artificial intelligence (AI)–driven monitors and real-world data streams now identify clinical risks in minutes, promising a future of proactive, earlier intervention. While AI automation is often marketed as a tool to reduce administrative burden and allow health care providers to focus more on direct patient care, this unrealized potential currently stands in contrast to our reality. Building high-speed data “freeways” without “off-ramps” such as clinical staffing, workflow synergy, and the patient education required for meaningful response is like building a superhighway without well-engineered off-ramps to provide a safe way to get home, and this creates a dangerous paradox. Earlier detection without earlier care does not improve outcomes; it simply redistributes anxiety and extends the patient’s period of uncertainty. We argue that the “public as a sensor” is already signaling a systemic infrastructure gap. True safety in clinical AI isn’t found in more algorithmic guardrails, but in participatory co-design that ensures every digital alert has a viable human pathway to care and resolution. We must stop building high-speed roads that lead to a cliff edge of clinical unavailability and consider that while the technology is a feat of engineering, it’s our human architecture that makes it medicine. Introduction Health care systems can detect clinical problems earlier than at any point in history. Continuous cardiac monitors record rhythms for weeks, wearable devices capture subtle physiological changes throughout the day, and artificial intelligence (AI) algorithms analyze real-world data to identify patterns clinicians might otherwise miss [ 1-3 ]. Alerts generated by these systems move rapidly through electronic health records, often within minutes. The promise underlying this innovation is clear: earlier detection should lead to earlier intervention and better outcomes [ 4 ]. However, detection alone does not improve care [ 5 ]. Without a timely clinical response, earlier identification of risk can simply extend the period of uncertainty experienced by patients. This uncertainty is not neutral. It can be distressing, particularly when patients receive alerts suggesting potential clinical deterioration but cannot access timely evaluation. At the same time, discussions about clinical AI frequently emphasize guardrails designed to protect patients [ 6 ], while access to timely care remains constrained [ 7 ]. Patients experience a paradox: increased monitoring and earlier detection paired with delayed clinical response. In this context, safety narratives focused solely on algorithmic governance risk overlooking the lived experience of patients navigating uncertainty without timely clinical support [ 8 ]. This is detailed in Figure 1 , which provides a narrative account of a patient encountering this systemic delay. Figure 1. the When detection outpaces care: the patient's voice. Open in a new tab Building Participatory Off-Ramps Despite growing adoption of clinical AI, patients and the public are still rarely invited into meaningful partnerships to shape implementation, co-design workflows, or develop AI literacy initiatives. This absence is notable given that patients are often the first to experience the downstream consequences of a delayed response, increased alerting, and fragmented workflows [ 9 ]. Emerging research suggests that AI-enabled monitoring increases clinical signals and alerts, yet workflows and workforce capacity have not evolved at the same pace [ 10 , 11 ]. Without corresponding investment in response pathways, automation risks amplifying uncertainty rather than improving care. This experience referenced in is increasingly common, even within large academic medical centers in the United States designed to provide rapid access to subspecialty expertise. In the US health care context, policy initiatives have encouraged the generation and use of real-world data. The issue is not technological capability alone, but whether health care systems have sufficient clinical capacity and workflow infrastructure to respond when new information emerges [ 12 , 13 ]. The Paradox of Progress Over the past decade, US policy initiatives have encouraged the generation and use of real-world data, including remote physiologic monitoring, patient-generated health data, and AI-supported triage systems. These technologies have expanded rapidly and demonstrated improved detection of deterioration across cardiovascular, metabolic, and chronic disease populations [ 4 ]. The engineering feats we are witnessing are nothing short of miraculous. We now have AI models identifying pancreatic cancer 3 years before a traditional diagnosis and wearable technology detecting silent heart failure from a wristband [ 14-16 ]. These are the high-speed lanes on the freeway of a new era in medicine. To truly honor these breakthroughs, we must match this digital brilliance with an equally sophisticated human infrastructure. Our task is not to slow down the “freeway,” but to build the robust, graded “off-ramps,” the clinical time, the team-based staffing, and the participatory co-design that ensure every signal reaches its destination: a healthy patient. When we invite the public to coproduce technology as “sensors” and co-designers, we aren’t just fixing a workflow; we are building a health care system worthy of the technology we’ve created [ 17 ]. Table 1 shows the promise, reality, and proposed solutions. Table 1. Mapping the paradox of progress in clinical artificial intelligence (AI). The promise The reality gap The participatory solution Earlier detection Alerts outpace workforce capacity. Detection happens in minutes; clinical response can take weeks. Workflow co-design: involving patients and clinicians in prioritizing alerts and defining response expectations for high-quality health care. Continuous monitoring Increased patient anxiety. Earlier awareness without earlier care simply extends the period of uncertainty. Human-centered guardrails: shifting focus from purely algorithmic governance to systemic safety, ethical agreement, and patient-supported partnership. AI efficiency Fragmented, “siloed” responses. Automation shifts effort toward interpretation and coordination, increasing cognitive demands. Integrated care pathways: co-constructing response infrastructure, such as customer support and team-based care, as a primary part of deployment. Open in a new tab Productive AI requires the meaningful involvement of patients and clinicians throughout the entire process, from co-design and ethics to defining alert thresholds, testing prototypes, assessing application quality, and shaping both feedforward concerns and feedback solutions, all while evaluating workflow integration. To ensure AI implementation truly serves those it is meant to help, we recommend fully compensated [ 18 ] “patient” and “clinician-in-the-loop” team roles, with genuine voting power over co-created outcomes. Real-world data hold enormous promise for improving outcomes and personalizing care. They allow clinicians to observe disease trajectories in everyday settings and identify changes earlier than traditional episodic visits permit. Yet data alone do not improve health. Without clinicians available to interpret signals and systems capable of responding quickly, detection may simply shift the burden of waiting from before diagnosis to after notification and before timely intervention. Participatory Medicine Declaration This work was co-produced, co-authored and pre–peer reviewed with patients. The Structure-Infrastructure Gap As emphasized above, we seek not to slow down the freeway, but to build off-ramps to ensure access to the destination: high-quality health care. Figure 2 illustrates the missing off-ramp for AI in health care. Figure 2. The missing off-ramp in modern health care artificial intelligence (AI) implementation. This conceptual illustration depicts the current disconnect between high-speed AI detection (the freeway) and the necessary systemic infrastructure required for clinical response (the off-ramp). Safe passage to high-quality health care requires bridging the gap between automated alerts and human-centered care through workforce capacity, protected clinical time, and participatory co-design. Image generated with the assistance of Google Gemini 3 Flash. Open in a new tab AI automation is marketed as an efficient, cost-cutting solution to workforce shortages. Yet emerging evidence suggests a more complex reality. AI-enabled alerting systems, ambient documentation tools, and automated triage systems frequently introduce new cognitive demands, increase alert fatigue, and create additional coordination work for clinicians when workflows are not redesigned accordingly [ 19 , 20 ]. Rather than reducing workload, automation can shift effort from documentation to interpretation, triage, and coordination, tasks that still require complex and expert human clinical judgment [ 21 ]. Workforce and Workflow Challenges These workflow challenges occur in the context of worsening workforce shortages. Projections suggest significant physician shortages across primary care and subspecialties in the coming decade, further limiting the system’s ability to respond to newly detected risks. At the same time, structural barriers limit workforce expansion [ 22 , 23 ]. Medical school costs have risen substantially, creating financial barriers for many prospective physicians and influencing specialty choice toward higher-compensation fields, exacerbating shortages in primary care and community-based settings [ 24 ]. Training bottlenecks further constrain workforce growth. Residency and fellowship placements remain limited, meaning that even when medical school enrollment increases, postgraduate training capacity may not expand in proportion to the need [ 25 ]. This creates a structural mismatch between educational output and workforce needs, particularly as digital health technologies generate more clinical signals requiring interpretation [ 11 ]. Advanced practice clinicians, including nurse practitioners and physician associates, are increasingly expected to help address these shortages [ 26 ]. However, they often face a paradoxical situation: in some contexts, they are criticized for perceived gaps in expertise, while in others they are expected to manage increasingly complex patient populations and respond to AI-generated alerts beyond traditional scope-of-practice boundaries [ 27 ]. Evidence suggests that team-based care models improve outcomes when roles are clearly defined and supported, rather than when advanced practice clinicians are used as substitutes for unavailable physicians [ 27 ]. For patients, this gap shifts responsibility in subtle but important ways. When notified of abnormal findings, individuals often monitor symptoms more closely, search for information, and attempt to interpret data while waiting for professional guidance [ 13 ]. For those living with chronic or complex conditions, this period can be particularly challenging. Instead of reducing uncertainty, earlier detection may extend it. In this sense, automation without response capacity risks creating a new form of harm: early awareness without early care [ 13 , 28 ]. Conclusion: Toward a Sociotechnical Solution Emerging informatics and implementation science emphasizes that successful digital transformation requires co-design with clinicians and patients, workflow redesign, and evaluation of downstream response capacity, not simply detection accuracy [ 7 , 29 ]. These findings reinforce the view that AI implementation is fundamentally a sociotechnical challenge rather than a purely technological one. The next phase of digital health policy must move beyond generating data and toward ensuring that care systems can act on the information they produce. Praising automation while neglecting clinical workflow and workforce capacity risks creating unintended harm. Real-world data will only transform health care if real-world doctors, and the teams who support them, are available to respond. We have invested billions in data infrastructure, that is, the high-speed lanes, while neglecting the off-ramps that reduce mortality and increase safety: the clinicians, time to task, patient education, and cost required to transition from a digital alert to a clinical resolution. As highlighted in Figure 2 , safety is not just a guardrail; it means ensuring there is a safe way to get home. Supplementary material Multimedia Appendix 1. Prompt used in image generation. jopm-v18-e97972-s001.docx (21.8KB, docx) DOI: 10.2196/97972 Acknowledgments This work was our idea, concept, and research. We found our own citations. We used artificial intelligence for rephrasing awkward sentences, spell-checking, checking for duplication, and creating the figure ( Multimedia Appendix 1 shows the prompt we used). Abbreviations AI artificial intelligence Footnotes Funding: This work was unfunded. Authors’ Contributions: CVR and AP contributed to all aspects of the publication. Conflicts of Interest: AP is the editor-in-chief of the Journal of Participatory Medicine , and CVR is on the patient advisory board. References 1. Meng X, Yan X, Zhang K, et al. The application of large language models in medicine: a scoping review. iScience. 2024 May 17;27(5):109713. doi: 10.1016/j.isci.2024.109713. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Placido D, Yuan B, Hjaltelin JX, et al. A deep learning algorithm to predict risk of pancreatic cancer from disease trajectories. Nat Med. 2023 May;29(5):1113–1122. doi: 10.1038/s41591-023-02332-5. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Lång K, Josefsson V, Larsson AM, et al. Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial (MASAI): a clinical safety analysis of a randomised, controlled, non-inferiority, single-blinded, screening accuracy study. Lancet Oncol. 2023 Aug;24(8):936–944. doi: 10.1016/S1470-2045(23)00298-X. doi. Medline. [ DOI ] [ PubMed ] [ Google Scholar ] 4. Katonai G, Arvai N, Mesko B. AI and primary care: scoping review. J Med Internet Res. 2025 Aug 15;27:e65950. doi: 10.2196/65950. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Labkoff S, Oladimeji B, Kannry J, et al. Toward a responsible future: recommendations for AI-enabled clinical decision support. J Am Med Inform Assoc. 2024 Nov 1;31(11):2730–2739. doi: 10.1093/jamia/ocae209. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Rozenblit L, Price A, Solomonides A, et al. Toward responsible AI governance: balancing multi-stakeholder perspectives on AI in healthcare. Int J Med Inform. 2025 Nov;203:106015. doi: 10.1016/j.ijmedinf.2025.106015. doi. Medline. [ DOI ] [ PubMed ] [ Google Scholar ] 7. Rozenblit L, Price A, Solomonides A, et al. Towards a multi-stakeholder process for developing responsible AI governance in consumer health. Int J Med Inform. 2025 Mar;195:105713. doi: 10.1016/j.ijmedinf.2024.105713. doi. Medline. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Gandhi TK, Classen D, Sinsky CA, et al. How can artificial intelligence decrease cognitive and work burden for front line practitioners? JAMIA Open. 2023 Oct;6(3):ooad079. doi: 10.1093/jamiaopen/ooad079. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Cordovano G. AI, ML, and NLP snub patients’ right of access. Journal of AHIMA. [19-04-2026]. https://journal.ahima.org/page/ai-ml-and-nlp-snub-patients-right-of-access URL. Accessed. 10. Newton N, Bamgboje-Ayodele A, Forsyth R, et al. Experiences of alert fatigue and its contributing factors in hospitals: qualitative study. J Med Internet Res. 2026 Feb 19;28(1):e78676. doi: 10.2196/78676. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Bedard J, Kropp M, Hsu M, Karaman OT, Hawes J, Kellerman GR. When Using AI Leads to “Brain Fry. Harv Bus Rev. Mar 5, 2026. [09-04-2026]. https://hbr.org/2026/03/when-using-ai-leads-to-brain-fry URL. Accessed. 12. Arogyaswamy S, Vukovic N, Keniston A, et al. The impact of hospital capacity strain: a qualitative analysis of experience and solutions at 13 academic medical centers. J Gen Intern Med. 2022 May;37(6):1463–1474. doi: 10.1007/s11606-021-07106-8. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. del Rey Puech P, Saund J, Panteli D, McKee M. Demystifying Artificial Intelligence in Health: What Health Policy-Makers Need to Know. European Observatory on Health Systems and Policies; 2026. [20-05-2026]. https://eurohealthobservatory.who.int/publications/i/demystifying-artificial-intelligence-in-health-what-health-policy-makers-need-to-know URL. Accessed. [ PubMed ] [ Google Scholar ] 14. Attia ZI, Harmon DM, Dugan J, et al. Prospective evaluation of smartwatch-enabled detection of left ventricular dysfunction. Nat Med. 2022 Dec;28(12):2497–2503. doi: 10.1038/s41591-022-02053-1. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Meng F, Zhang Z, Hou X, et al. Machine learning for prediction of sudden cardiac death in heart failure patients with low left ventricular ejection fraction: study protocol for a retroprospective multicentre registry in China. BMJ Open. 2019 May 16;9(5):e023724. doi: 10.1136/bmjopen-2018-023724. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Kwon JM, Jo YY, Lee SY, et al. Artificial intelligence-enhanced smartwatch ECG for heart failure-reduced ejection fraction detection by generating 12-lead ECG. Diagnostics (Basel) 2022 Mar 8;12(3):654. doi: 10.3390/diagnostics12030654. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Price A, Damaraju A, Kushalnagar P, et al. Coproduction, coeducation, and patient involvement: everyone included framework for medical education across age groups and cultures. JMIR Med Educ. 2021 Nov 3;7(4):e31846. doi: 10.2196/31846. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Price A, Richards DP. Patient partner research has a compensation crisis. BMJ. 2025 Feb 27;388:r381. doi: 10.1136/bmj.r381. doi. Medline. [ DOI ] [ PubMed ] [ Google Scholar ] 19. Mache S, Bernburg M, Würtenberger A, Groneberg DA. Artificial intelligence in primary care: support or additional burden on physicians’ healthcare work?-A qualitative study. Clin Pract. 2025 Jul 25;15(8):138. doi: 10.3390/clinpract15080138. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Goodson DA, Garcia B, Hogarth M, Tu SP. Artificial intelligence and physician burnout: a productivity paradox. Learn Health Syst. 2025 Oct;9(4):e70013. doi: 10.1002/lrh2.70013. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Bienefeld N, Keller E, Grote G. AI interventions to alleviate healthcare shortages and enhance work conditions in critical care: qualitative analysis. J Med Internet Res. 2025 Jan 13;27:e50852. doi: 10.2196/50852. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Walensky RP, McCann NC. Challenges to the future of a robust physician workforce in the United States. N Engl J Med. 2025 Jan 16;392(3):286–295. doi: 10.1056/NEJMsr2412784. doi. Medline. [ DOI ] [ PubMed ] [ Google Scholar ] 23. Petterson SM, Liaw WR, Phillips RL, Rabin DL, Meyers DS, Bazemore AW. Projecting US primary care physician workforce needs: 2010-2025. Ann Fam Med. 2012;10(6):503–509. doi: 10.1370/afm.1431. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Ramesh T, Kadakia KT, Liu M, Yu H. Federal loans among US medical students. JAMA. 2026 Jan 13;335(2):179–181. doi: 10.1001/jama.2025.20905. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Lakhan SE. The wrong fix: why America doesn’t need more medical schools to solve the physician shortage. Cureus. 2025 Jun;17(6):e85935. doi: 10.7759/cureus.85935. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Auerbach DI, Buerhaus PI, Staiger DO. Implications of the rapid growth of the nurse practitioner workforce in the US. Health Aff (Millwood) 2020 Feb;39(2):273–279. doi: 10.1377/hlthaff.2019.00686. doi. Medline. [ DOI ] [ PubMed ] [ Google Scholar ] 27. Price A, Price K. A call to recognize the integral role of physician associates and nurse practitioners in modern health care: editorial. J Particip Med. 2026 Jan 23;18(1):e89262. doi: 10.2196/89262. doi. Medline. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Osnat B. Patient perspectives on artificial intelligence in healthcare: a global scoping review of benefits, ethical concerns, and implementation strategies. Int J Med Inform. 2025 Nov;203:106007. doi: 10.1016/j.ijmedinf.2025.106007. doi. Medline. [ DOI ] [ PubMed ] [ Google Scholar ] 29. deBronkart D. The patient’s voice in the emerging era of participatory medicine. Int J Psychiatry Med. 2018 Nov;53(5-6):350–360. doi: 10.1177/0091217418791461. doi. Medline. [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Multimedia Appendix 1. Prompt used in image generation. jopm-v18-e97972-s001.docx (21.8KB, docx) DOI: 10.2196/97972 Articles from Journal of Participatory Medicine are provided here courtesy of JMIR Publications Inc. ACTIONS View on publisher site PDF (692.6 KB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Exploring Technological Solutions for Interoperability Between Patient Electronic Medical Records and Clinical Registries: Scoping Review.#227746
• Development of a digital tool to support overview in complex patient cases: Which information elements support the clinical workflow?#220980
• FedEHR-Agents: Federated Agentic Optimization for Automated EHR Modeling#781141
• FedEHR-Agents: Federated Agentic Optimization for Automated EHR Modeling#779766
• EveryQuery: A Promptable Foundation Model for Clinical Prediction Tasks over Electronic Health Records#1116225