Doximity is building two complementary trust layers for clinical AI. PeerCheck provides physician review — 10,000+ medical experts validating AI-generated clinical answers for accuracy, evidence strength, and potential bias. DocInsight provides strategic intelligence — turning physician engagement and behavioural data into insight products for life-sciences partners. Together, they represent a trust model that combines human expert review with behavioural intelligence within a single platform serving 85% of US physicians.
The combination is strategically significant because it addresses trust at two levels: individual output trust (PeerCheck verifies that specific answers are accurate) and systemic trust (DocInsight identifies where the system, the guidelines, or the users are consistently struggling — enabling continuous improvement).
Why Review Alone Is Not Enough
Physician review checks individual outputs — verifying that a specific clinical answer is accurate, well-evidenced, and clinically appropriate. This is valuable and credible: a named cardiologist reviewing an AF management answer adds trust that anonymous AI generation cannot. PeerCheck-certified answers carry attribution, creating an accountability chain from AI generation through physician verification to the clinician who reads the answer.
But review is necessarily selective. With tens of thousands of possible clinical queries across dozens of specialties, not every answer can be reviewed by every relevant specialist in every clinical context. The reviewed subset creates a verified core. The unreviewed majority relies on the AI's baseline performance plus the surrounding trust architecture — source quality, retrieval precision, fidelity controls, fail-safe behaviour, and feedback mechanisms.
Review must also account for several dimensions that simple verification does not automatically address. Jurisdiction specificity — a US cardiologist's review may not apply to UK clinical practice where NICE guidelines, UK formulary preferences, and NHS referral pathways differ from US practice. A review that certifies an answer as accurate for US physicians may be inaccurate for UK clinicians. Versioning — the review date should be visible so clinicians know whether it reflects current guidance. Guidelines change. Drug approvals change. Evidence evolves. A review from 2024 may not apply in 2026. Audit trail — the review process and criteria should be transparent. Who reviewed this answer? Against what standard? When? What methodology? Evidence hierarchy — the review should reference specific evidence sources, not just expert opinion. Expert opinion is the lowest level of the evidence hierarchy. A reviewed answer that cites "expert consensus" rather than a specific guideline or trial has a different trust weight.
Where Insight Fits in the Trust Architecture
Review checks outputs one at a time. Insight identifies patterns across many outputs — where the clinical AI system, the underlying guidelines, the user workflows, or the educational infrastructure repeatedly create difficulty. These are different functions that serve different purposes.
If clinicians consistently ask the same prescribing question hundreds of times — a question the AI can answer but that clinicians repeatedly feel they need to verify — that pattern is a signal. The guideline may be hard to apply. The AI answer may not be presented in a way that builds sufficient confidence. The educational system may not be teaching the topic effectively. The clinical workflow may not give clinicians time to absorb the answer. Individual output review cannot detect these systemic patterns. Aggregate insight analysis across many users, many queries, and many sessions can.
DocInsight operationalises this at US scale — using Doximity's physician engagement data to identify patterns in clinical AI usage, drug-information access, documentation behaviour, and workflow engagement. iatroX Insights operationalises the UK equivalent — with different inputs (UK clinical queries grounded in NICE/CKS/SmPC, regulatory context, NHS adoption dynamics) and different outputs (guideline-friction reports, clinician validation studies, market-entry readiness assessments, clinical safety advisory).
iatroX's UK Trust Architecture
iatroX Insights combines four trust layers for UK clinical AI, each addressing a different dimension of the trust problem.
Guideline-first evidence. Ask iatroX provides source-grounded answers from UK authoritative sources — with visible citations that clinicians can check, algorithmic fidelity controls that keep outputs aligned with retrieved evidence, fail-safe behaviour that narrows or abstains when evidence is insufficient, and clinician feedback mechanisms that enable continuous quality improvement. The clinical answer layer is independent of any commercial partnership — non-negotiably.
Clinician validation. Structured evaluation of clinical AI products by UK healthcare professionals — assessing accuracy, usability, trust, workflow fit, and clinical appropriateness. Not a satisfaction survey. A validation sprint with structured clinical scenarios, independent evaluation criteria, named clinician evaluators, and actionable findings. The output is a validation report with clinical credibility — useful for marketing, procurement bids, investor due diligence, and product development prioritisation.
Regulatory advisory. MHRA classification analysis, DTAC readiness assessment, DCB 0129/0160 clinical safety case development, CSO services, and claims review. Understanding regulatory requirements early prevents the costly process of retrofitting governance after launch — and ensures the product can access NHS procurement channels from the earliest possible stage.
Aggregate trend analysis. Anonymised, aggregated analysis of UK clinical query patterns — what clinicians ask, where they struggle, what they trust, where the knowledge infrastructure fails. Feeding into guideline-friction reports, educational-need assessments, and partner intelligence. All within explicit ethical boundaries: aggregate only, no individual profiling, no answer manipulation, no patient-identifiable data.
A Productised Offer: UK Clinical AI Trust Review
A combined assessment covering: source quality (where do answers come from?), citation accuracy (do citations match source content?), UX and usability (can clinicians verify quickly and confidently?), clinical claims (are claims safe under UK regulations?), safety architecture (does the system fail safely when evidence is insufficient?), regulatory readiness (is MHRA/DTAC/DCB compliance in place or achievable?), and clinician validation feedback (what do UK healthcare professionals think when they actually use the product?). Delivered as a structured report with actionable recommendations.