An approved engagement gives the organization a Governance Decision Package that states what the available evidence supports, what remains uncertain or blocked, and whether the named clinical AI use case should Proceed, Modify, Pause, or Stop.
The engagement is bounded and time-limited. Scope and duration are set in the written Statement of Work and depend on site approvals, data readiness, and the named use case.
This page records what the retrospective, component-specific evidence does and does not support, so a governance reviewer can judge the record directly rather than take a claim on trust.
An evidence review of the documentation pathway supporting one named clinical AI use case.
A written account of limitations, including where the local data cannot answer the question.
Findings as they are, including null, adverse, conflicting, and blocked results.
A governance recommendation: Proceed, Modify, Pause, or Stop.
A Governance Decision Package written to be reviewed and challenged.
Decision rights and limits
VitaSignal recommends; the health system decides. The evaluation is read-only and retrospective first. Silent-mode or observational work is conditional, separately authorized in writing, and only considered after prespecified evidence gates are met.
What the evidence supports
Individual analytical components, examined retrospectively on historical records.
What it does not support
Prospective performance, clinical claims, or approval of any deployment. Prospective performance has not been established.
These findings support individual analytical components, not the complete Clinical AI Documentation Readiness Evaluation, whose scope and duration are set in the written Statement of Work and depend on site approvals and data readiness. Each evaluation is scoped to one named AI use case, and only evidence-matched components supported by the required fields, workflow definitions, ground truth, and written scope are included.
Evidence
Evidence matched to the question being asked
Cards are ordered by evidence strength, beginning with the strongest current retrospective component result. Each card states the evidence class and the controlling limitation for one component. Supportive, null, adverse, conflicting, blocked, rollback, abstention, insufficient-evidence, planned, and absent findings are reported in their own classes rather than blended. No finding establishes complete-platform validation, prospective clinical performance, patient or workflow benefit, fairness benefit, production integration, or deployment readiness.
Evidence class: Strongest current retrospective component result; adverse calibration and utility findings reported with it
A frozen retrospective confirmation showed added discrimination over a respiratory comparator in an institutionally external cohort. The same evaluation produced adverse findings that are reported with it: the external Brier score was worse, the calibration slope was weak, positive predictive value was low, and performance degraded under missingness. The cohort is institutionally external and historical, not historically untouched, and unavailable demographic subgroup, hospital, and vendor fields plus prior historical use of the source data further limit interpretation. This is retrospective component evidence for one component. It is not a validated product, and it does not establish prospective performance, clinical utility, treatment benefit, workflow benefit, generalizability, deployment readiness, or beneficial patient outcomes.
In a retrospective ALOTT analysis, conservative admission-day flowsheet observations added discrimination for a later first ICU start beyond laboratory and order features. Medication activity did not add a supported increment beyond flowsheets, and the planned earlier-cutoff analysis was blocked, so the finding holds only at the end-of-day cutoff, which may be too late for some operational decisions. These three results are reported together. This is an internal component finding, not external validation, prospective performance, treatment effect, workflow benefit, or complete-platform validation.
Overall next-day extubation forecasting produced a positive component result. The extubation-failure subgroup was null or adverse, and next-day SOFA prediction was worse than persistence. These results do not establish intervention benefit, adaptive-threshold benefit, closed-loop effectiveness, or patient benefit.
Forecasting endpoint feasibility in a second external dataset
Evidence class: Endpoint reconstruction blocked; no model result
In a separate feasibility assessment on a different external multi-center dataset, direct predictor fields were available, but the required extubation and 48-hour failure endpoint could not be reconstructed defensibly, so no model result was created and no proxy endpoint was substituted. The blocked endpoint is a completed fail-closed data-pathway finding. It is not a model success and not a model failure, and it is reported separately from the institutionally external confirmation above.
Transfer-readiness audit stopped before scoring
Evidence class: FAIL-CLOSED READINESS RESULT; NO MODEL SCORE OR PERFORMANCE METRIC
The transfer-readiness audit stopped before scoring because exact feature-parity and hospital-aware evaluation gates did not close. No model was scored, target labels were not used for recalibration, and no performance metric was produced. The correct recommendation is Stop. This is a completed fail-closed readiness result, not evidence of successful transfer, external validation, clinical utility, or deployment readiness.
A published alarm-adjudication component reproduced with a positive result within its defined component scope. The broader all-alarm analysis was null, and planned robustness checks were blocked and did not run. The positive reproduction is reported only together with its null broader result and blocked checks, and it does not establish alarm-reduction benefit, clinical utility, or patient benefit.
Documentation activity proxy
Evidence class: Mixed proxy; direct burden construct blocked
Documentation-event volume supported a positive high-volume classification result, while continuous-count prediction was adverse. Event activity is not measured documentation time, workload, staffing need, productivity, burnout, or overtime.
Engineering conformance
Evidence class: Bounded internal checks only
Several evidence-governance and provenance components have passed bounded internal synthetic or static checks. The fixtures, implementations, and expected results are not independently authored, and the work does not establish production interoperability, independent validation, security, clinical effectiveness, or governance effectiveness.
Public evidence summaries identify the dataset type, endpoint, timing window, evidence class, and controlling limitation. Generated records shown in the Interactive Product Tour are illustrative and are never presented as part of this empirical record. Additional controlled technical materials may be reviewed under an appropriate agreement when relevant to a scoped institutional decision.
Synthetic workflow and conformance demonstrations.
A Governance Decision Package format.
A proposed bounded 90-day evaluation method.
What is not established
Completed partner-site validation.
Production EHR connectivity or deployment.
Customer demand or willingness to pay.
Clinical or operational outcomes.
Savings, return on investment, staffing, workload, or burnout effects.
Repeatable enterprise delivery capacity or product-market fit.
Data handling at each stage
Public website and intake
VitaSignal does not request or accept PHI.
Qualification call
No patient examples, screenshots, exports, identifiers, or PHI are requested.
Scoping
Synthetic, aggregate, or properly de-identified information only, unless a separately approved route exists.
Evaluation
Retrospective or approved read-only access only, with institutional authorization and a documented minimum-data contract.
If protected data may be needed, the institution and its authorized counsel determine the required BAA, DUA, privacy, security, and contracting terms before any data transfer. VitaSignal does not state that it is HIPAA compliant and does not automatically sign BAAs. No security certification, penetration test, SOC 2, HITRUST, FedRAMP, vendor certification, or universal contracting position is claimed.
How the framework maps to current clinical AI governance concerns
VitaSignal has compared the current Governance Decision Package structure with lifecycle concerns described in the American Medical Association's Ethical AI Use in Medicine materials. A crosswalk is a traceability tool. It shows coverage and VitaSignal's disposition for each concern, not blanket adoption of any external framework and not endorsement by any organization. It does not indicate that the AMA, Duke Institute for Health Innovation, or Health AI Partnership has reviewed, approved, endorsed, or validated VitaSignal.
This is a document-level crosswalk. Dispositions may be Adopted, Mapped, Conditional, Not adopted, or Unresolved, and not every external item is treated as mandatory. VitaSignal's own governance model remains controlling. Nothing here establishes endorsement, equivalence, validation, adoption, customer demand, or commercial value.
Adopted
Mapped
Conditional
Not adopted
Unresolved
Tool appropriateness, intended purpose, and limitations
Strong conceptual alignment
Preserve one bounded use case, permitted and prohibited uses, local evidence questions, uncertainty, and a pause when appropriateness is unclear. Do not claim that the process improves institutional decisions.
Health outcomes, equity, and individualized care
Partial alignment
Preserve subgroup, adverse, null, and uncertain evidence. Do not claim clinical outcomes, fairness, individualized-care utility, or patient benefit. Patient-level decisions remain outside the current read-only evaluation scope.
Disclosure to patients
Gap or use-case-dependent requirement
Add a decision field for whether patient disclosure is required, not required, or unresolved, with rationale and an accountable owner.
Permission and informed consent
Partial gap
Record whether consent is required, not required, or unresolved, the governing authority, rationale, evidence, and accountable owner. VitaSignal does not independently make the legal or ethical determination.
Accountability, human oversight, overrides, and escalation
Strong conceptual alignment
Preserve named owners, human review, stopping rules, dissent, unresolved questions, and Proceed, Modify, Pause, or Stop recommendations. Do not imply that VitaSignal replaces institutional authority.
Continuous monitoring, model drift, incidents, and subgroup degradation
Partial alignment with an implementation gap
Record the future monitoring plan, review interval, drift and subgroup checks, incident triggers, and stopping thresholds. Do not imply that continuous production monitoring has been implemented or validated.
Automation bias, overreliance, and loss of skill
Gap
Add a human-factors field covering appropriate reliance, verification burden, override behavior, escalation, and possible loss of skill before clinical workflow evaluation.
Error, near-miss, and safety reporting
Partial gap
Add the institution's reporting route, escalation threshold, response owner, and evidence-preservation rule before any live or prospective evaluation.
Document-level crosswalk of governance concerns, current coverage outcome, and VitaSignal's disposition or next control.
Governance concern
Current outcome
VitaSignal disposition or next control
Tool appropriateness, intended purpose, and limitations
Strong conceptual alignment
Preserve one bounded use case, permitted and prohibited uses, local evidence questions, uncertainty, and a pause when appropriateness is unclear. Do not claim that the process improves institutional decisions.
Health outcomes, equity, and individualized care
Partial alignment
Preserve subgroup, adverse, null, and uncertain evidence. Do not claim clinical outcomes, fairness, individualized-care utility, or patient benefit. Patient-level decisions remain outside the current read-only evaluation scope.
Disclosure to patients
Gap or use-case-dependent requirement
Add a decision field for whether patient disclosure is required, not required, or unresolved, with rationale and an accountable owner.
Permission and informed consent
Partial gap
Record whether consent is required, not required, or unresolved, the governing authority, rationale, evidence, and accountable owner. VitaSignal does not independently make the legal or ethical determination.
Accountability, human oversight, overrides, and escalation
Strong conceptual alignment
Preserve named owners, human review, stopping rules, dissent, unresolved questions, and Proceed, Modify, Pause, or Stop recommendations. Do not imply that VitaSignal replaces institutional authority.
Continuous monitoring, model drift, incidents, and subgroup degradation
Partial alignment with an implementation gap
Record the future monitoring plan, review interval, drift and subgroup checks, incident triggers, and stopping thresholds. Do not imply that continuous production monitoring has been implemented or validated.
Automation bias, overreliance, and loss of skill
Gap
Add a human-factors field covering appropriate reliance, verification burden, override behavior, escalation, and possible loss of skill before clinical workflow evaluation.
Error, near-miss, and safety reporting
Partial gap
Add the institution's reporting route, escalation threshold, response owner, and evidence-preservation rule before any live or prospective evaluation.
Standards posture
Standards-aware evidence, not certification
VitaSignal maintains a version-aware standards crosswalk that distinguishes regulatory baselines, optional Standards Version Advancement Process versions, local test evidence, and unresolved implementation requirements. The 2026 SVAP versions became available for voluntary certification use by applicable Certified Health IT developers on August 29, 2026. VitaSignal does not claim ONC certification or conformance to these versions.
Current optional standards targets
These are current optional standards targets. They are not implemented capabilities, not certified capabilities, and not production integrations.
USCDI Version 6.
HL7 FHIR US Core STU 9.0.0.
HL7 C-CDA Release 5.0.0.
Da Vinci CRD 2.2.1, DTR 2.2.0, and PAS 2.2.1 for separately approved prior-authorization use cases.
Regulatory authorization and health-system readiness answer different questions. A health system may still need to determine whether available evidence, its local population and data conditions, workflow assumptions, monitoring controls, and unresolved uncertainty support advancing a specific clinical AI use case.
A 2026 PLOS Digital Health evidence-mapping study included 1,357 FDA-cleared or approved AI or ML-enabled devices through December 5, 2025. Under the authors' linkage and outcome definitions, 34 were linked to registered prospective trials and three evaluated the specified patient-centered outcomes. The study may not capture every form of device evidence. These findings do not validate VitaSignal or establish that any individual device lacks adequate evidence.
This section covers what remains blocked and the historical publication record. Only evidence-matched components supported by the available fields, workflow definitions, ground truth, and written scope are included in any engagement.
Blocked or readiness-only methods
Not available for interpretation until the required local constructs, timing, authorization, or adjudicated ground truth exist. None is available for clinical use.
Shift-end and handoff interpretation
Blocked until verified local shift boundaries, assignments, handoff events, and appropriate outcomes exist. Timing and concentration analyses are reported only where verified local shift boundaries exist.
Documentation reliability
Readiness-only data-quality review of candidate documentation-availability and metadata-quality indicators, pending adjudicated reliability ground truth. Not a validated reliability score and not a readiness certification.
Subgroup governance
Planning framework only. Requires prespecified groups, sufficient cell sizes, uncertainty reporting, and explicit insufficient-evidence states. It does not certify fairness or establish equitable benefit.
Bedside, cognitive-load, and burnout constructs
Blocked by missing direct constructs, timing, authorization, or adjudicated ground truth. Readiness audits and generic clinical data do not replace those missing criteria.
Qualitative external problem context
A regional Ontario hospital initiative has described duplicated privacy, cybersecurity, vendor, clinical-safety, ethical, policy, and procurement review and is developing reusable governance assets while preserving each hospital's local authority. This is qualitative external problem context, not evidence of United States demand, savings, endorsement, or VitaSignal validation.
Documentation-native clinical intelligence and equity audits, July 2026.
Forthcoming book, Routledge
Nursing and the Algorithm: Ethics, Safety, and the Future of Clinical Judgment. Being published December 2026.
Recognition of research and writing is not evidence that any AI system is safe or effective for your setting, and it does not imply a health-system relationship, endorsement, or completed engagement.
Study design, data pathways, limitations, blocked methods, and the historical publication record are consolidated on this page under Methods, blocked work, and historical scholarship. Protected method families are not described publicly and are discussed only under a written agreement. Terminology used across these pages is defined in the Glossary.
