Healthcare IT

• KLAS (reported by Fierce Healthcare and HIT Consultant, May 2026): Epic held 43.7% of US acute-care hospitals at the end of 2025, up from 42.3%, and 56.9% of beds.
• Epic added 77 acute-care hospitals in 2025 while Oracle Health lost 56, its third consecutive year of the largest net share loss. Oracle Health fell to 21.9%.
• Only two large health systems made enterprise-wide EHR decisions in 2025, and both chose Epic.
• KLAS (via HIT Consultant, May 2026): 84% of Meditech customers deciding on a move chose to migrate to Expanse, up from 63% in 2024.

• Rock Health (H1 2026 report, via Fierce Healthcare and Healthcare Dive): US digital-health startups raised $7.4B across 244 deals in the first half, versus $6.4B across 245 deals in H1 2025.
• Rock Health Q3 2026 update (via Milled summary): $3.7B across 114 deals in the quarter and $11.1B across 358 deals year to date, with the firm projecting the year to pass $14.8B.
• Rock Health said it is retiring its separate AI-deal tracking because AI has become table stakes. A secondary source (ai2.work) cites its 2025 year-end report as putting AI companies at 54% of 2025 funding, up from 37% in 2024; this was not checked against the original.

The flexibilities lapsed briefly on 31 January 2026 during a partial government shutdown. The Consolidated Appropriations Act, 2026, signed on 3 February 2026, extended them through 31 December 2027 (law-firm summaries from Foley, Greenbaum and AASM, early 2026). The reimbursement ceiling described in chapter 06 therefore moved out by two years but is still set by legislation.

Abridge closed a $316M Series E extension in April 2026 at a $5.3B valuation, according to secondary coverage (valueaddvc and aggregator pages; not confirmed against a company release).

UnitedHealth Group reported Q3 2026 revenue of $112.7B (versus $113.2B a year earlier), a medical care ratio of 88.4%, and raised full-year adjusted EPS guidance to at least $20.50 from $19.50-$20.00 (company 8-K and press coverage, 13 Oct 2026). UnitedHealthcare served 48.3M consumers and Optum over 120M patients and consumers in the release.

UnitedHealth's final count of 192.7M affected individuals (reported August 2025, via HHS site) is still the latest figure found.

Not refreshed: Meditech and Others hospital shares (Others is recomputed as the remainder), HL7/TEFCA status, prior-authorization survey figures, revenue-cycle market size, telehealth claim-line share for 2025-26, Teladoc, FDA AI clearances, ambient-AI health-system counts, UnitedHealth lives covered (52M, differing definition from the 48.3M in the Q3 release), and 2025 full-year Rock Health total.

A mother arrives unconscious. The ER can't see she is allergic to penicillin, because her records sit in another hospital's system. In US healthcare, one patient's history can be scattered across ten or more platforms that cannot talk to each other.

• Safety. Illegible scripts and missing allergy data caused thousands of deaths a year.
• Efficiency. Physicians re-ordered tests they couldn't locate: duplicate work, duplicate cost.
• Coordination. The average patient saw seven doctors; none could see the others' notes.
• Speed. An unconscious patient's history was simply unavailable in the ER.

• Fused layers: OS, language and database welded into one, so you can't upgrade a piece; you replace the whole system.
• Tree 'globals': data stored as branches that exist only when data does. Genius on 4KB, meaningless without the vendor's dictionary.
• Built for concurrency: each department writes to its own branch at once, never colliding. Perfect for the bedside.
• The tradeoff: fast for one patient, brutal across millions. A population query must walk every tree.

1. Chronicles: the live MUMPS database. Fast writes, tree-structured, unsafe to query at scale.
2. Nightly ETL: a batch job copies everything after hours, a ~24-hour cycle.
3. Clarity: a separate SQL warehouse. Relational, analytics-ready, safe to query.
4. The 24-hour gap: reports run on yesterday's data; a deteriorating patient is flagged too late.

• Switching cost. $50-500M and 18-36 months, both systems running at once.
• Clinical safety. A new system must prove it never crashes during care, which needs scale.
• Chicken-and-egg. Reaching scale needs early adopters willing to bet patients' safety.
• Network effects. The value of an EHR rises with how many peers share it.

1. The moat: one patient record no rival can assemble module by module.
2. Analytics: Cogito, Caboodle and SlicerDicer, plus Cosmos across 250M+ patients.
3. Patient and mobile: the MyChart portal; Haiku and Canto clinician apps.
4. Operational modules: scheduling, registration, pharmacy, lab, radiology, OR, billing.
5. Clinical core: EpicCare inpatient and ambulatory, one chart across every setting.

1. More hospitals adopt Epic.
2. Care Everywhere grows more valuable.
3. ~90% of residents train on Epic.
4. Switching costs reach $50-500M.
5. Churn falls toward zero.
6. The loop repeats: data and network effects.

Six standards in forty years. Each solved a layer. None solved all four. Machines can now exchange the envelope of clinical data. They still can't read the letter inside.

1. 1987HL7 v2: a shared message format, but ~20% stayed proprietary.
2. 2005HL7 v3: rigorous, unreadable, barely adopted.
3. 2009HITECH: $27B digitized the silos, not the connections.
4. 2014FHIR: web-style APIs cut interface cost from $500K to ~$50K.
5. 2016Cures Act: made data-blocking illegal, with fines up to $1M.
6. 2023TEFCA: one national hub instead of thousands of contracts.

1. Foundational: the pipes connect (systems connect).
2. Structural: the format is consistent.
3. Organizational: governance and consent align.
4. Semantic: meaning is shared. This is the gap (machines understand).

A large share of American healthcare exists to translate care into a bill, and then to argue over it. Roughly $1 trillion a year, about a quarter of all health spending, pays for administration, not care. Per capita, the US spends ~3.5x what Canada does.

1. Encounter: the clinician documents the visit in the EHR as free text.
2. Coding: 185,000 coders translate notes into ICD-10 and CPT codes.
3. Clearinghouse: Change Healthcare routes ~40% of all US claims.
4. Adjudication: the payer approves, denies or downcodes, 14 to 30 days.
5. Paid or denied: 10%+ are denied; 80%+ of appeals are later overturned.

• Behavioral health. The durable winner: most telehealth visits, no physical exam needed.
• Chronic-care check-ins. Medication management and follow-ups moved online for good.
• Primary and urgent care. Reverted largely to in-person once offices reopened.
• The reimbursement cliff. Pandemic payment parity was temporary and has been extended in two-year steps; policy, not tech, sets the ceiling.

Telehealth's limit is no longer the technology. It is the reimbursement rule. The durable question is not whether a visit can happen over video, but whether it will be paid the same when it does.

For sixty years, software could exchange clinical data but never understand it. Language models changed that. Every constraint in this deck collapses into one bottleneck, machine comprehension of clinical language, that is finally being lifted.

• Ambient documentation: AI listens and drafts the note live, 100+ systems (Abridge, DAX, Ambience).
• Coding and prior auth: models read the record and generate payer-specific, code-accurate requests.
• Trial matching: free-text records screened against trial criteria in minutes, not weeks.
• The unlock: LLMs finally read clinical language. 'MI' = 'heart attack' = 'acute MI'.
• The Olive lesson: an $850M startup died automating the UI, not the meaning. Semantics is the moat.
• The pincer: enterprise compression inside the system, patient aggregation outside, at once.

A two-by-two map plots AI disruption potential (vertical) against incumbent lock-in (horizontal). Top-left is the open field for AI and bottom-right is the entrenched moats. Ambient documentation, prior authorization and patient data aggregation sit toward the open field. Revenue-cycle and coding sits top-right, as contested high ground. Clearinghouses and claims, medical imaging AI, and the EHR system of record sit toward the entrenched, high-lock-in side, with the EHR system of record lowest on disruption potential.

• The EHR fortresses: Epic (private), Oracle Health, Meditech, the systems of record.
• The verticals: UnitedHealth/Optum, CVS/Aetna, insurance-to-pharmacy under one roof.
• The admin layer: athenahealth, Waystar, Veradigm, RCM and claims plumbing.
• Ambient AI: Abridge, Ambience, Nuance/Microsoft, Suki, the scribe wars.
• Consumer rails: Apple Health, Hims, Doximity, GoodRx, the patient front door.
• The platforms: OpenAI, Anthropic, Google, Amazon, betting the bottleneck is gone.

Healthcare didn't refuse to modernize. It modernized perfectly, for a question the world stopped asking. The bill for that mismatch is a trillion dollars a year, and the receipt is a record no one can fully read.

• CMS: National Health Expenditure data (2024).
• KLAS Research: US acute-care EHR market share (2024-25).
• ONC / HealthIT.gov: hospital EHR adoption and interoperability metrics.
• McKinsey & Company; Annals of Internal Medicine: healthcare administrative cost.
• American Medical Association: prior authorization physician survey (2023-24).
• Rock Health: US digital health funding reports (2021-2024).
• FAIR Health: Monthly Telehealth Regional Tracker.
• Institute of Medicine: 'To Err Is Human' (1999); Pew Charitable Trusts / GAO: patient matching.
• Figures marked 'estimate' are compilations of the above.