Kidney-Attributable CV Risk Flag for Cardiology

Concept

A plug-in app for cardiology EHRs built on the SMART on FHIR standard — the common plug-in format that runs inside Epic, Oracle Health (formerly Cerner), and other major electronic health records, distributed through their vendor app programs. It reads creatinine and eGFR results from routine chemistry panels via standard FHIR lab-observation interfaces, with an HL7v2 feed (an older but widely supported hospital data standard) as a fallback for legacy lab systems. No new tests and no new workflow steps are required; it operates entirely on labs already ordered. When routine chemistry shows reduced kidney function, the app flags that the patient belongs to a group with elevated cardiovascular risk and surfaces a guideline-linked action: kidney-protective therapy review, urine ACR confirmation, or nephrology referral.

One honesty boundary is explicit in the design: the 11.5% figure from GBD 2023 [0] is a population-level statistic — the share of cardiovascular deaths worldwide attributable to impaired kidney function. The app cannot compute an individual patient's 'kidney-attributable' risk from that average, and it does not claim to. What it shows is the patient's actual eGFR, a plain statement that reduced kidney function is a modifiable cardiovascular risk factor, and the linked action. Where a validated risk calculator accepts kidney measures, eGFR can be fed into it — but the flag itself is group membership plus the patient's own labs, not a personalized percentage.

Designing with clinicians, not for them

New supporting evidence [1] directly shapes the interface. Research on cardiovascular AI finds that even highly accurate prediction models see limited day-to-day adoption, and argues that tools must 'think with, not just for' clinicians — aligning with how clinicians actually reason instead of outputting opaque scores. This answers one of this iteration's open questions: yes, a population-based flag risks alert fatigue if it behaves like a black box. So the flag is built for cognitive alignment — transparent inputs (the lab value itself), reasoning the cardiologist can verify at a glance, and a specific action rather than a number to trust blindly.

Guardrails against over-treatment: a single reduced eGFR can reflect acute kidney injury rather than chronic disease, so the tool checks prior labs and clinical context before escalating, and prompts urine ACR confirmation before any CKD label or nephrology referral. Acknowledged alerts are suppressed rather than repeated.

First users

Population health managers and cardiology service-line directors at multi-specialty health systems with value-based care contracts — the roles accountable for cardiovascular mortality and chronic kidney disease progression across an entire patient panel. They can deploy the flag across a cardiology cohort using labs already ordered and track completion of the prompted actions (therapy reviews, ACR testing) as a quality metric.

Why now

• GBD 2023 [0] provides the population-level rationale: impaired kidney function accounts for 11.5% (95% uncertainty interval 8.4–14.5%) of cardiovascular deaths, and CKD itself ranks as the 9th leading cause of death worldwide (1.48 million deaths in 2023).
• Multimorbidity research [4] confirms kidney disease rarely travels alone — it clusters with the metabolic and cardiovascular conditions cardiologists already manage — so the flag runs on data already in the chart.
• Cognitive-alignment research [1] explains why accurate-but-opaque clinical AI stalls in practice, and offers the design path this tool follows.

Honest limits and open questions

• The 11.5% is population-level; individual attributable risk cannot be derived from it. The tool's claim is deliberately narrower: this patient has reduced kidney function, a modifiable CV risk factor, and here is the guideline action.
• A single lab value cannot distinguish AKI from CKD; confirmation steps are built in, and over-referral remains a risk if clinicians bypass them.
• Whether existing cardiology workflows already use eGFR for cardiovascular risk stratification (rather than only for drug dosing) is unresolved; if they do, the tool must earn its place through the action-linking, not the flag alone.
• Value-based care contracts vary widely; the first-user hypothesis assumes systems where kidney-protective outcomes are rewarded. Validating that incentive is a next step.
• FHIR lab-data access differs across EHR vendors and versions; the HL7v2 fallback and vendor app-program distribution reduce but do not eliminate integration effort.