Clinical Decision Support · FHIR
CareSignal
The information may already be in the chart. The challenge is knowing what needs attention.
A FHIR-based clinical decision-support prototype that applies deterministic rules to structured patient information and surfaces potential chronic-disease care gaps.
01 — The problem
The data can already be there and still be hard to see.
Chronic-disease management involves many recurring measurements, tests and follow-up requirements.
The necessary data can already exist in the patient record while an overdue test or potential care gap remains difficult to notice inside an information-dense chart.
Can structured EMR data be evaluated systematically so clinicians can see which patients may need attention first?
02 — Scope
Four conditions. Nine deterministic clinical rules.
Type 2 diabetes
Recurring measurement and monitoring requirements
Hypertension
Blood pressure follow-up and review intervals
Chronic kidney disease
Function monitoring and staging review
Dyslipidemia
Lipid monitoring and medication review
Each rule is explicit and reproducible. Given the same patient data, a rule produces the same result every time — which is what makes an alert traceable back to the logic that raised it.
03 — The workflow
From FHIR data to a clinician decision.
FHIR patient data
Deterministic rules
Care gap
Priority
Clinician review
Follow-up action
04 — Three levels
The product answers three questions in order.
Population
See the overall pattern across the panel.
Prioritize
Identify which patients require attention first.
Understand
Review the underlying data and care gaps for one patient.
01 · Population — what is happening across the panel?
05 — Prioritization
Who should be reviewed first?
Identified care gaps are assigned a priority level so a clinician can start with the patients whose gaps matter most, rather than reading every chart in sequence.
CareSignal does not only generate individual alerts — it organizes the whole panel by urgency.
Priority ordering is a review aid. It does not replace clinical judgment about which patient needs attention.
02 · Prioritize — which patients need attention?
06 — Where AI sits
The boundary is the design decision.
AI summarizes.
Rules decide.
Azure OpenAI GPT-4o generates concise clinical summaries to make the chart easier to review.
It does not determine whether a care gap exists.
The alerts and priority levels come from deterministic clinical logic based on defined rules. The summary is labelled in the interface so a clinician always knows which part of the screen a model wrote.
07 — FHIR
Built around standards-based resources.
The application is designed around FHIR R4 resources and standards-based workflows. Working against defined resource types is what allows the same clinical logic to be expressed independently of any one vendor’s internal data model.
Designing against FHIR R4 is not the same as being integrated with a given EMR. Connecting to a real system would require its own authorization, data-mapping and validation work.
08 — What I learned
The interesting question was not where AI could be added. It was where it should not be responsible for the decision.
CareSignal made the distinction between deterministic logic and generative AI much clearer to me.
A language model can help summarize complex information.
But when an alert needs to be reproducible and traceable, the underlying logic should remain explicit.
Limitations
- Independent prototype
- Not a medical device
- Not clinically validated
- Not connected to live patient records
- Demonstration data, not real patients
- Not a substitute for clinical judgment