UpToDate and OpenEvidence in Late 2026: The ACP Partnership, AI Features, Adoption, and EHR Integration

Contents
  1. Origins and content licensing
  2. User adoption and real-world usage
  3. AI architecture and evaluation studies
  4. Position in the 6S pyramid and where errors still occur
  5. EHR integration and Epic's native AI
  6. Where Japan stands
  7. References

On November 9, 2026, the American College of Physicians (ACP) and OpenEvidence announced a licensing collaboration1. ACP represents roughly 163,000 internists across more than 172 countries, including Japan. The agreement brings ACP's clinical guidance and peer-reviewed content from Annals of Internal Medicine into OpenEvidence.

Earlier the same autumn, on October 9, Wolters Kluwer Health announced the domestic launch of UpToDate Expert AI in Japan, rolling out to more than 800 existing UpToDate institutions alongside evaluation at the Japan Institute for Health Security (JIHS)2. On November 15, Apple and OpenEvidence also began a U.S. health AI research study3. Both UpToDate and OpenEvidence now pair generative AI with electronic health record (EHR) workflows.

Key milestones in 2025–2026 (Japan developments and the ACP partnership)
  1. Jun 2025
    Elsevier Japan × OpenEvidence launch ClinicalKey AI in JapanInstitutional subscription combining Elsevier reference books and journals with co-developed AI
  2. Oct 2025
    Veeva Systems (Veeva Japan) × OpenEvidence announce Open VistaJoint life-sciences collaboration targeting clinical trial access and drug development
  3. Feb–Oct 2026
    Rapid expansion of Epic EHR integrations across major U.S. health systemsOE integrates at Sutter, Mount Sinai, MSK, NYU; UpToDate connects into Epic Evidence in Art
  4. Oct 9, 2026
    UpToDate Expert AI domestic launch announced in JapanJapanese queries and answers rolling out to 800+ subscribing sites; evaluated at JIHS
  5. Nov 9, 2026
    ACP × OpenEvidence licensing collaboration announcedIntegrates ACP clinical guidance and peer-reviewed Annals of Internal Medicine content
  6. Nov 15, 2026
    Apple × OpenEvidence launch Apple Health AI Usage StudyJoint study enrolling up to 200,000 U.S. adults aged 21 and older

Sources: References 1–5 and 20–24

OpenEvidence's terms exclude the EU, UK, and EEA but do not list Japan, and Japanese clinicians already query it in Japanese after credential verification6. Official Japan-facing commercial offerings tied to OpenEvidence technology currently include Elsevier Japan's ClinicalKey AI4 and Veeva Japan's Open Vista partnership5.

Two clinicians at a simple desk: stacked reference books with a pink bookmark on the left, and a laptop with a pink-rimmed magnifying glass on the right

Origins and content licensing

UpToDate began in 1992 when nephrologist Dr. Burton "Bud" Rose and colleagues mailed expert-written clinical topics on floppy disks three times a year, before Wolters Kluwer acquired it in 20087,8. It now maintains more than 13,000 peer-reviewed topics across 25 specialties9.

OpenEvidence was founded in 2022 by Daniel Nadler and Zachary Ziegler and grew out of Mayo Clinic Platform Accelerate in early 2023. Offering free access to verified U.S. clinicians drove rapid individual adoption, backed by licensing deals in 2025–2026 with NEJM Group, JAMA Network, NCCN, Wiley/Cochrane, Springer Nature, Sage, and ACP1,10–12.

User adoption and real-world usage

UpToDate grew from over 1 million clinicians in 2014 to 2.5 million in 2020 (with UpToDate Advanced) and over 3 million across 190+ countries by 20257,13. UpToDate Expert AI had reached more than 2,700 U.S. hospitals (>95% of U.S. Enterprise Edition clients) and 2,000+ pilot sites across 70 countries outside the U.S. by November 202614.

For OpenEvidence, a JAMA study published October 15, 2026 (Patel et al.) analyzed U.S. NPI-verified logs from January 2024 through June 202615. Monthly active users jumped from 355 to 703,265 across 952,008 clinicians and 271 million queries. Queries centered on treatment and management (33.1%) rather than diagnosis (13.9%), with up to 88% of emergency medicine queries crossing outside the clinician's specialty.

Published global scale of UpToDate and U.S. real-world usage of OpenEvidence
UpToDate (Global · IR & Official Releases)
Reported licensed clinician scale
201420202025–
  • *2020 9M IR combines UpToDate & UpToDate Advanced
  • Expert AI: 2,700+ U.S. hospitals (>95% of U.S. Enterprise)
  • 2,000+ pilot sites in 70 countries outside the U.S. (Nov 2026 IR)
OpenEvidence (U.S. NPI-Verified · JAMA 2026)
Monthly active users (MAU) and physician query intent
Monthly active users (Jan 2024 → Jun 2026)
Jan 2024355
Jun 2026703k
2.5-yr total: 952,008 clinicians (556,200 MDs) / 271M queries
Query intent & out-of-specialty rate (JAMA 2026)
Clinical decisions51.9%
Treatment/mgmt33.1%
Diagnosis13.9%
Out-of-specialty35–88%
  • Queries/active user-mo: 18.0→39.5 (NP/PA 44.5 · MD 37.1)
  • Early-career MDs (<10 yrs) submit 61.8% of MD queries
  • 1M clinical consultations in 24h (March 10, 2026)

Sources: References 7 and 13–16

OpenEvidence's valuation climbed from $1 billion in February 2025 to $15 billion in September 202616.

AI architecture and evaluation studies

The core distinction discussed in an earlier post (Secondary clinical resources vs. AI search)—an editorially curated secondary reference versus an AI search engine over primary literature—remains central even after both added generative AI.

UpToDate Expert AI vs. OpenEvidence at a glance
UpToDate Expert AI
Sources: 13,000+ peer-reviewed clinical topics across 25 specialties + Lexidrug
Output: Shows citations, assumptions, and rationale on screen; abstains when evidence is lacking
Revenue: Paid subscription (ad-free)
Ecosystem: Suki, Doximity, ChatGPT for Healthcare, AgileMD acquisition agreement
OpenEvidence
Sources: Licensed journals and society guidelines (NEJM, JAMA, Cochrane, NCCN, Springer Nature, ACP)
Output: Four-model family + GRADE-inspired EvidenceGrade™ indicators
Revenue: Free for verified clinicians + pharma ads (with JAMA commentary calling for scrutiny19)
Ecosystem: Visits, Coding Intelligence, Learn, Access Intelligence, Anthropic global rollout

Sources: References 1, 9–12, and 17–19

OpenEvidence launched EvidenceGrade™ in July 2026 to visualize cited study strength—addressing the same systematic-review and GRADE workflow previously built into a custom Chrome extension—and followed in September 2026 with a four-model family17,18.

OpenEvidence Model Family (announced September 2026)
Osler (~5 sec)
Fast point-of-careQuick lookups during clinic or ward rounds▶ Default model
Sackett (~30 sec)
Interactive clarifyingAsks follow-up context to narrow guidance▶ Named for EBM pioneer
Snow (~5 min)
Deep complex researchParallel literature runs for multi-morbid cases▶ Successor to DeepConsult
Darwin (Preview)
Frontier reasoningApplication-only preview for biosafety governance▶ Not public

Company-stated timing estimates. Source: Reference 18

In a June 2026 Nature Medicine head-to-head comparison (Vishwanath et al.), OpenEvidence and UpToDate Expert AI scored closely on MedQA, HealthBench, and real-world clinical questions (RCQs)25. The sharpest gap was abstention on RCQs: 1% for OpenEvidence versus 19% for UpToDate Expert AI, reflecting Expert AI's conservative threshold when evidence is limited.

Head-to-head comparison (Nature Medicine, June 2026)
OpenEvidence UpToDate Expert AI
MedQA accuracy500 questions
89.6%
88.4%
HealthBench500 items (0–100)
62.6
61.3
Real Clinical Qs100 items (1–4, NS)
3.24 / 4.0
3.17 / 4.0
RCQ abstentionRate across 100 items
1% (1/100)
19% (19/100) conservative abstention

Evaluated late 2025–Feb 2026 (prior to September OE model family). Source: Reference 25

In citation audits, Lotan et al. (npj Health Systems, September 2026) found zero fabricated citations across 4,979 OpenEvidence references alongside three abstract attribution errors26. Systematic reviews and specialty audits—including 125 MASLD eConsults (12 minor errors, 1 potentially major error) and 20 pharmacy questions (40% strict accuracy, 82.5% overall quality)—have also been published27–29.

Position in the 6S pyramid and where errors still occur

Evidence-based medicine often organizes clinical sources by degree of aggregation into the six-tier "6S pyramid"30. In an earlier post (Secondary clinical resources vs. AI search), I treated AI search answers as roughly comparable to a narrative review and built a Chrome extension that steered queries toward systematic reviews, meta-analyses (SR/MA), and GRADE recommendations.

Today, alongside UpToDate Expert AI, OpenEvidence's deals with Cochrane, NCCN, and ACP plus EvidenceGrade™ mean that within the upper aggregated tiers—secondary summaries, clinical guidelines, and systematic reviews—both tools can surface reliable, grounded guidance. Outside existing guidelines and SR/MA, however, AI answers remain closer to a narrative review assembled from individual studies, making primary-paper appraisal essential.

UpToDate and OpenEvidence through the 6S aggregation pyramid
1. SystemsEHR-linked CDSS
EHR-integrated decision support
2. SummariesSecondary refs
Secondary summariesUpToDate, DynaMed, etc.
3. Synopses of SynthesesSR abstracts
Synopses of systematic reviewsCochrane Clinical Answers, etc.
4. SynthesesSR/MA & guidelines
Systematic reviews & guidelinesCochrane Review, NCCN, ACP, etc.
Aggregated Evidence Tiers
AI retrieves & synthesizes existing guidelines, SRs, and summaries
UpToDate Expert AI OpenEvidence Partnered with Cochrane, NCCN, ACP
5. Synopses of StudiesStudy summaries
Appraised study synopsesShort critical summaries of RCTs
6. StudiesOriginal research
Original studiesRCTs, cohorts, case reports
Traditional AI Coverage
Outside guidelines & SRs, treat as narrative-review level & check originals

Adapted from Haynes RB (Evid Based Med 2006) and DiCenso A, Bayley L, Haynes RB (Evid Based Nurs 2009). Sources: Reference 30 and prior posts (/en/articles/124911743/ and /en/articles/2890/)

Real citations do not guarantee a safe answer either. OpenEvidence rarely fabricates references26, yet depending on how a question combines comorbidities or local prescribing rules, an answer backed by real papers can still be clinically wrong.

A concrete example also discussed in What to Delegate to AI in Hospitals Toward 2040 involves quetiapine. Consider a 78-year-old man with Parkinson's disease and type 2 diabetes hospitalized for aspiration pneumonia who develops hyperactive delirium. Asking an AI for pharmacologic options may cite overseas guidelines—such as the German Society of Neurology guideline—and suggest avoiding haloperidol while starting low-dose quetiapine31. Although quetiapine is sometimes used in Japan for Parkinson's disease delirium alone, whereas the U.S. FDA label carries a hyperglycemia warning rather than a contraindication, Japan's PMDA package insert lists patients with diabetes or a history of diabetes as a contraindication because severe hyperglycemia and ketoacidosis can be fatal32.

Real citations, wrong local recommendation (delirium prescribing example)
Case: 78M · Parkinson's disease + type 2 diabetes + hyperactive delirium (aspiration pneumonia)
Typical AI Answer (Real Overseas Citations)
  • Avoid haloperidol in patients with Parkinson's disease
  • Consider low-dose quetiapine (DGN guideline, etc.)
The cited papers are real and the prose reads smoothly, yet local contraindications and diabetes may be missed.
Check Against Japan's PMDA Label
  • Contraindicated in patients with current or past diabetes
  • Marked hyperglycemia and ketoacidosis are cited as potentially fatal adverse outcomes
Even if used for isolated Parkinson's delirium, type 2 diabetes rules it out under Japanese labeling (whereas the U.S. FDA label lists a warning rather than a contraindication).

Sources: References 31–32 and article /en/articles/3020/

In practice, checking systematic reviews, guidelines, and UpToDate first—and using AI answers as a starting point to verify primary papers and domestic prescribing labels whenever comorbidities or local rules apply—keeps clinical searches grounded.

EHR integration and Epic's native AI

In 2026, U.S. health systems moved rapidly to embed clinical AI directly inside Epic. OpenEvidence rolled out Epic integrations at Sutter Health, Mount Sinai Health System, Cedars-Sinai, Memorial Sloan Kettering Cancer Center (adding OncoKB), Northwestern Medicine, and NYU Langone Health20–22.

Wolters Kluwer likewise announced connecting UpToDate knowledge into Epic's clinician assistant via Evidence in Art and into Epic Agent Factory (broad availability slated for 2027)23,24. Meanwhile, Epic pairs its three native agents—Art (clinicians), Emmie (patients), and Penny (operations)—with Cosmos, its de-identified EHR network33.

A clinician at a desktop monitor beside three stacked thin horizontal layers, with the middle layer highlighted in pink, representing layered clinical information inside the EHR
Three knowledge layers converging inside U.S. EHRs
1. UpToDate (Curated secondary reference) Peer-reviewed recommendations & drugs → Evidence in Art / Agent Factory
2. OpenEvidence (Primary literature & guidelines) NEJM, JAMA, ACP search → Enterprise Epic integrations
3. Epic Native AI (Art / Emmie / Penny + Cosmos) Charting & ops + real-world patterns from comparable EHR patients
Single EHR Workspace
  • Active patient chart & labs
  • Expert consensus (1)
  • Recent trials & guidelines (2)
  • Real-world cohort trajectories (3)
Three layers meet at the point of care

Sources: References 20–24 and 33

Where Japan stands

Japanese hospitals operate a fragmented mix of EHR vendors, so in-chart clinical AI integrations like those on Epic remain uncommon. Even so, the October 2026 Japan rollout of UpToDate Expert AI enables subscribing hospitals to begin navigating UpToDate recommendations in Japanese, while OpenEvidence offers fast access to new English trials and ACP or NCCN guidelines—and domestic tools such as MedGen Japan help cover Japanese guidelines and reimbursement context.

A clinician holding a single sheet marked with a tiny pink tab while speaking calmly across a table with a seated patient

How far these three layers—curated secondary guidance (UpToDate), primary-literature AI search (OpenEvidence), and real-world EHR data (Epic Cosmos)—reach into Japanese clinics and hospital charts will be a defining trend over the next few years.

References

  1. American College of Physicians. ACP and OpenEvidence collaborate to enhance physician access to trusted clinical guidance. November 2026 (distributed November 9). https://www.acponline.org/acp-newsroom/acp-and-openevidence-collaborate-to-enhance-physician-access-to-trusted-clinical-guidance
  2. Wolters Kluwer Health (Japan). Launch of UpToDate Expert AI in Japan. October 9, 2026. https://www.wolterskluwer.com/ja-jp/news/uptodate-expert-ai-japan-launch
  3. Apple Newsroom. Apple launches research study to explore the use of AI for health. November 15, 2026. https://www.apple.com/newsroom/2026/11/apple-launches-research-study-to-explore-the-use-of-ai-for-health/
  4. Elsevier Japan. ClinicalKey AI launch in Japan. June 27, 2025. https://www.elsevier.com/ja-jp/about/press-releases/clinicalkey-ai-launch
  5. Veeva Japan. OpenEvidence and Veeva Announce Open Vista Partnership. October 23, 2025. https://www.veeva.com/jp/resources/openevidence-and-veeva-announce-open-vista-partnership/
  6. OpenEvidence Inc. Network Terms of Use. https://www.openevidence.foundation/policies/terms
  7. Wolters Kluwer. Wolters Kluwer celebrates 30 years of UpToDate. December 13, 2022. https://www.wolterskluwer.com/en/news/wolters-kluwer-celebrates-30-years-of-uptodate
  8. Wolters Kluwer. Acquisition of UpToDate (SEC filing, 2008). https://www.sec.gov/Archives/edgar/vprr/0800/08005522.pdf
  9. Wolters Kluwer. What is UpToDate Expert AI?. https://ai.uptodate.com/docs/user-guide/introduction-to-expert-ai/what-is-expert-ai/
  10. OpenEvidence. OpenEvidence and NEJM Group Sign Content Agreement. February 19, 2025. https://www.openevidence.com/announcements/openevidence-and-nejm
  11. OpenEvidence / JAMA Network. OpenEvidence and the JAMA Network sign strategic content agreement. June 5, 2025. https://www.openevidence.com/announcements/openevidence-and-the-jama-network-sign-strategic-content-agreement
  12. Cochrane. Cochrane evidence to inform OpenEvidence users. March 3, 2026. https://www.cochrane.org/about-us/news/cochrane-evidence-inform-openevidence-users
  13. Wolters Kluwer. 2014 Full-Year Results / 2020 Nine-Month Trading Update / 2025 Microsoft Integration Release. https://www.wolterskluwer.com/en/news/wolters-kluwer-to-integrate-uptodate-into-healthcare-agent-service-in-microsoft-copilot-studio
  14. Wolters Kluwer. Nine-Month 2026 Trading Update. November 4, 2026. https://assets.contenthub.wolterskluwer.com/api/public/content/3891876-2026-11-04-wolters-kluwer-nine-month-2026-trading-update-575ee0543d
  15. Patel VR, Finlayson SG, Liu-Hass M, Ray M, Burk-Rafel J, Zack TI, Jena AB. Clinician Use of an Artificial Intelligence–Enabled Clinical Information Platform. JAMA. Published online October 15, 2026. https://doi.org/10.1001/jama.2026.21868 (PMID: 42879271)
  16. Becker's Hospital Review. OpenEvidence raises $250M at $15B valuation. September 28, 2026. https://www.beckershospitalreview.com/healthcare-information-technology/ai/openevidence-raises-250m-at-15b-valuation/
  17. PR Newswire / OpenEvidence. OpenEvidence launches EvidenceGrade™. July 10, 2026. https://www.prnewswire.com/news-releases/openevidence-launches-evidencegrade-empowering-physicians-to-see-the-strength-of-cited-evidence-beneath-each-ai-answer-302822750.html
  18. Business Wire / OpenEvidence. Introducing the OpenEvidence Model Family. September 3, 2026. https://www.businesswire.com/news/home/20260903878517/en/
  19. Singh N, Mello MM. Advertising in AI-Powered Clinical Decision Support Tools. JAMA. 2026;336(1):15-16. https://doi.org/10.1001/jama.2026.8078
  20. Sutter Health / Mount Sinai / Cedars-Sinai / Northwestern Medicine. OpenEvidence EHR Integration Releases. February–October 2026. https://news.nm.org/openevidence-and-northwestern-medicine-launch-collaboration-to-integrate-clinical-evidence-into-the-ehr/
  21. Memorial Sloan Kettering Cancer Center. MSK and OpenEvidence partner to advance precision oncology at point of care. September 16, 2026. https://www.mskcc.org/news-releases/mskcc-msk-and-openevidence-partner-to-advance-precision-oncology-at-point-of-care
  22. NYU Langone Health. OpenEvidence and NYU Langone Health partner to bring patient-aware intelligence to 12 million visits in New York City. October 29, 2026. https://nyulangone.org/news/openevidence-and-nyu-langone-health-partner-bring-patient-aware-intelligence-12-million-visits-new-york-city
  23. Wolters Kluwer. UpToDate and Epic collaborating to streamline workflows and help transform patient care. September 11, 2026. https://www.wolterskluwer.com/en/expert-insights/uptodate-and-epic-collaborating-to-streamline-workflows-help-transform-patient-care
  24. Epic. At Agent Factory Build-a-Thon, Health Systems Create Their Own AI Agents. September 15, 2026. https://www.epic.com/epic/post/at-agent-factory-build-a-thon-health-systems-create-their-own-ai-agents/
  25. Vishwanath K, Alyakin A, Ghosh M, et al. General-purpose large language models outperform specialized clinical AI tools on medical benchmarks. Nat Med. 2026;32:2405-2409. https://doi.org/10.1038/s41591-026-04431-5
  26. Lotan T, et al. Reference quality of OpenEvidence across five medical specialties. npj Health Systems. 2026;3:87. https://doi.org/10.1038/s44401-026-00142-8
  27. Artsi Y, Sorin V, Glicksberg BS, et al. OpenEvidence clinical question-answering platform: systematic review of early evidence. npj Digit Med. 2026. https://doi.org/10.1038/s41746-026-03077-4
  28. Stephenson CR, et al. Large Language Model–Generated Electronic Consultations: A Blinded Comparative Study in MASLD. Mayo Clin Proc Digit Health. 2026. https://doi.org/10.1016/j.mcpdig.2026.100404
  29. Issa N, Colmenares E, Douglas A, et al. Accuracy and Quality Assessment of Artificial Intelligence Platforms and a Drug Information Database. Am J Health Syst Pharm. 2026. https://doi.org/10.1093/ajhp/zxag379
  30. Haynes RB. Of studies, syntheses, synopses, summaries, and systems: the "5S" evolution of information services for evidence-based healthcare decisions. Evid Based Med. 2006;11(6):162-164. https://doi.org/10.1136/ebm.11.6.162-a / DiCenso A, Bayley L, Haynes RB. Accessing pre-appraised evidence: fine-tuning the 5S model into a 6S model. Evid Based Nurs. 2009;12(4):99-101. https://doi.org/10.1136/ebn.12.4.99-b
  31. Witt K, et al. Diagnostics and treatment of impulse control disorders, psychosis and delirium: systemic review-based recommendations - guideline “Parkinson’s disease” of the German Society of Neurology. J Neurol. 2024;271:7402-7421. https://doi.org/10.1007/s00415-024-12576-x
  32. Quetiapine fumarate package insert (Japan). Pharmaceuticals and Medical Devices Agency (PMDA).
  33. Epic. AI at Epic (Art, Emmie, Penny) / Cosmos. https://www.epic.com/software/ai/ / https://cosmos.epic.com/
English translation of the Japanese article.