LegendEHR Predictive Decision Support Intervention #

Transparency Disclosure Table #

Use Case: Psoriasis Flare Prediction #

(Prepared for ONC §170.315(b)(11) Certification)

Predictive DSI Transparency Attributes

Purpose / Intended Use #

Predicts likelihood of a psoriasis flare within 30 days to support proactive treatment planning. Not a diagnostic tool; intended as a clinical aid.

Target Population #

Adult patients (18+) with a confirmed psoriasis diagnosis.

Input Data Sources #

EHR data: demographics, diagnosis codes, medication adherence, lab values (e.g., CRP), lifestyle factors (smoking, BMI), prior flare history.

Training Data
Historical de‑identified patient records from dermatology clinics (2023–2025), ~500 patient encounters

Model Type / Methodology #

Gradient boosting machine learning model trained on structured EHR data.

Validation / Performance Metrics #

AUROC: 0.82; Sensitivity: 78%; Specificity: 75%. Validated on independent dataset of 10,000 patients.

Fairness / Bias Assessment #

Tested across subgroups (age, gender, Fitzpatrick skin types I–VI). Performance consistent across groups; slight sensitivity drop in skin type V–VI addressed by rebalancing training data.

Limitations #

Predictions may be less accurate for patients with rare comorbidities or incomplete medication adherence data. Not validated in pediatric populations.

Update / Maintenance Frequency #

Model retrained annually with most recent 2 years of data; knowledge base reviewed quarterly.

Human Oversight #

Final treatment decisions remain with the dermatologist. Alerts can be overridden or dismissed.

Transparency of Logic #

Risk score generated from weighted factors (recent flare history, biologic adherence, inflammatory markers). Clinician can view top contributing factors per prediction.

Risk Management #

Alerts flagged as “advisory” only; no automatic treatment changes. Providers must confirm before acting.

Governance / Accountability #

Maintained by LegendEHR Clinical AI Team; oversight by Medical Advisory Board.

Contact for Feedback / Issues #

Support portal: support@legendehr.com — clinicians can report anomalies, request clarifications, or suggest improvements.