More record sharing cut readmissions in community hospitals and raised them inside the VA
A study of 2.4 million veterans a month finds health information exchange volume moving outcomes in opposite directions depending on which side of the exchange you are on.
Health information exchange is one of the few digital health interventions with an unambiguous theory behind it: a clinician who can see what happened to a patient elsewhere makes better decisions than one who cannot. A study published in the Journal of the American Medical Informatics Association on 28 July tested that theory at national scale and got two answers pointing in opposite directions [s1].
The study
The researchers used electronic health record data covering all Veterans Health Administration enrollees from January 2022 to December 2023 — 3,144 medical center-months, representing 2.4 million patients per month [s1].
Their explanatory variable was health information exchange volume, defined as the number of unique patient-days with completed exchanges [s1]. Primary outcomes were 30-day readmissions, avoidable hospitalisations and in-hospital mortality [s1].
The design is the notable part. Exchange volume is not randomly assigned — medical centres that exchange more records differ from those that exchange less in ways that also affect outcomes. To address this, the authors instrumented for exchange volume using each medical centre's count of organisational exchange partners, and estimated models with medical centre and month fixed effects plus controls for direct patient volume [s1]. They then estimated separately for community care and for VHA direct care [s1].
The split result
In community care, a one-standard-deviation increase in exchange volume was associated with 4.07 fewer 30-day readmissions (95% CI -4.62 to -3.51), 1.45 fewer avoidable hospitalisations (95% CI -2.84 to -0.06), and 0.25 fewer inpatient deaths (95% CI -0.44 to -0.06), per medical centre per month [s1].
In VHA direct care, the same increase was associated with 7.07 additional readmissions (95% CI 6.46-7.67) and 0.19 additional inpatient deaths (95% CI 0.03-0.35), with no significant effect on avoidable hospitalisations (95% CI -0.96 to 1.68) [s1].
Same intervention, same health system, same two years — opposite signs on two of three outcomes.
What the authors think is happening
The paper offers three candidate mechanisms and does not claim to have distinguished between them [s1].
The first is asymmetric content: what flows out of the VHA is longitudinal patient history, while what flows in is episodic summaries [s1]. A community clinician receiving a veteran's full record gets something substantively different from what a VHA clinician receives about a community hospital episode.
The second is asymmetric interfaces — modern community electronic health records against the VHA's Joint Legacy Viewer [s1]. Data that arrives in a system that surfaces it poorly is not functionally the same as data that arrives in a system that surfaces it well.
The third is a selection mechanism rather than a clinical one: exchange-enabled reclamation of community patients back into VHA care [s1]. If exchanging records makes it easier for the VHA to identify and bring back veterans who have been treated elsewhere, then higher exchange volume changes who is in the VHA's denominator — plausibly toward patients who are sicker or whose care has been fragmented. Their readmissions would then show up as an increase without anything about their care having got worse.
The authors state that more work is needed to disentangle these mechanisms [s1].
Why the third mechanism matters most
The first two explanations describe an implementation problem — fix the viewer, change what gets sent, and the sign flips. The third describes a measurement artefact, and it would mean the negative VHA finding is not a harm at all.
An instrumental-variables design handles confounding by unmeasured centre characteristics that are uncorrelated with the instrument. It does not handle a situation where the instrument itself changes the composition of the patient population being measured. If exchange partnerships bring different patients into VHA hospitals, the instrument is affecting the outcome through a channel other than better-informed care.
This does not invalidate the study — the authors raise the possibility themselves [s1] — but it is the reason the VHA-side result should be held more loosely than the community-side one.
The wider interoperability problem
A state-of-the-art paper published in Applied Clinical Informatics on 24 June frames why results like this keep being ambiguous [s2]. Its premise is that clinical data exchange remains fragmented and insufficient to support coordinated, safe care, and that decades of technological progress, standards development, and interoperability legislation have produced only incremental improvement [s2].
The authors — drawing on an expert roundtable, further expert discussions, and appraisal of published and online evidence on standards, policies and developing regulations — identify five barrier domains: fragmented terminology standards and value sets; multiple overlapping data exchange standards with uneven FHIR adoption; incomplete or misaligned certification and policy incentives; variable interpretation and implementation of implementation guides; and insufficient funding and support for standards development [s2].
Their proposal is a time-bound US "moonshot" with three pillars: a canonical clinical terminology they call SuperSNOMED, integrating major domains and mappings; FHIR acceleration converging on a minimum set of implementation guides with certified legacy-data transforms during a transition window; and a policy and governance pillar covering certification criteria, incentive alignment, and funding for a paid expert task force to execute a three-year plan [s2].
It is explicitly a call to action rather than an evaluation [s2] — but the second and fourth of its barrier domains, uneven standards adoption and variable implementation, are recognisable in the JAMIA study's finding that identical exchange volume produced opposite effects on either side of a single interface [s1].
What to watch
Whether follow-up work separates the reclamation mechanism from the interface and content mechanisms, since that determines whether the VHA finding is a harm or an artefact. Whether the VHA's viewer modernisation changes the direction of the direct-care result. And whether any interoperability policy begins measuring outcomes rather than exchange volume, which is the metric most current programmes actually track.
Sources
- [s1] Veteran health information exchange volume and 30-day readmissions, avoidable hospitalizations, and in-hospital mortality: evidence from community and Veterans Health Administration direct care. Journal of the American Medical Informatics Association, published online 28 July 2026. https://doi.org/10.1093/jamia/ocag125
- [s2] An Interoperability Moonshot Proposal. Applied Clinical Informatics, published online 24 June 2026. https://doi.org/10.1055/a-2885-7865
Sources
- Veteran health information exchange volume and 30-day readmissions, avoidable hospitalizations, and in-hospital mortality: evidence from community and Veterans Health Administration direct care — Journal of the American Medical Informatics Association , July 28, 2026
- An Interoperability Moonshot Proposal — Applied Clinical Informatics , June 24, 2026
AI scribes are already in the exam room. Nobody has shown they are safe
A February theme issue documents faster notes, happier clinicians and enterprise rollouts to thousands. It also contains an editorial asking the question none of the studies answer.
The first randomised trials of AI scribes arrived. They do not fully agree.
Two November trials found ambient AI cut documentation time and work exhaustion. A parallel emergency department comparison found physicians spent more time in the note, not less.
Two August studies point language models at clinical text, with mixed verdicts
An EHR-embedded model wrote hospital summaries residents edited less — but with more confabulations. A separate pipeline audited 21,041 trial reports against CONSORT at 91.7% agreement with experts.
An Israeli HMO biobank sequenced 1,038 patients to hunt for deafness genes
Linking exome data to electronic medical records solved 15% of unexplained hearing-loss cases and flagged new candidate genes — while showing what the records could not supply.