Mammograms catch heart disease 💓, Info blocking hits court 🔨, Forus, OpenMed
Right to Read, but Wrong to Write: The Info-Blocking Case That Sets a Precedent
A federal court just issued a dual preliminary injunction in Vyne v. Henry Schein — the second information-blocking injunction in the Fourth Circuit. The ruling draws a bright line: Vyne can read from Dentrix databases but is barred from writing back.
The precedent matters more than the parties. The court is establishing that the 21st Century Cures Act protects read access to health data while allowing EHR vendors to restrict write access on security grounds. For every clinician-builder whose product reads EHR data via any method — API, RPA, print driver — this ruling defines the legal guardrails.
😤 “This is a dental software fight. How does it apply to me?” The legal mechanism — information blocking under the Cures Act — applies to all certified health IT. The precedent travels. If your product touches EHR data, this ruling affects your legal posture.
Forus Raises $160M to Automate the Gap Between Prescription and Patient
Forus (formerly Tandem) raised $160M from Thrive Capital, General Catalyst, Accel, Bain Capital Ventures, Redpoint, BoxGroup, and Pear VC. The product automates everything between a clinical decision and a patient starting treatment: insurance authorization, financial assistance, fulfillment routing.
Five of the top 10 global biopharma companies already work with Forus. Providers across all 50 states use the platform, supporting millions of patients annually.
This is the prior-auth-and-access stack that clinician-builders keep trying to build in-house. Forus raised $160M because the workflow between “prescribed” and “patient gets the drug” is still absurdly broken — and nobody owns it end-to-end.
😤 “Another prior auth startup. We’ve seen this before.” Forus isn’t just prior auth. It’s the full treatment-access pipeline: PA, financial assistance, specialty pharmacy routing, copay support. The investor roster (Thrive, GC, Accel) at this stage suggests real traction, not vaporware.
The Signal You’re Not Capturing: AI Finds Heart Disease in Mammograms
AI can now detect breast arterial calcifications (BAC) in routine mammograms — providing a cardiovascular risk signal from an exam that’s already being performed. A JACC study found 22.7% prevalence of BAC in the mammogram population and a 17% relative increase in major adverse cardiac events per 10-percentile BAC increase.
This is opportunistic screening at its most elegant. No new test. No new visit. No new cost. The data is already sitting in the mammogram. An AI algorithm extracts a cardiovascular risk signal from it.
😤 “Opportunistic findings create downstream workups and anxiety.” True. But 22.7% prevalence and 17% MACE increase per 10-percentile means this isn’t incidentaloma noise — it’s a clinically significant signal in a population (women, often pre-traditional-screening age) that’s systematically under-screened for cardiovascular disease.
💡 80/20: Watch GE HealthCare’s AI Advocacy work here. This is the template for “secondary signal extraction from existing imaging” — a product category that applies to chest X-rays (osteoporosis from vertebral height), chest CTs (coronary calcifications), and any high-volume modality where AI can find signals humans weren’t looking for.
Makary Out at FDA — Kyle Diamantas Takes Over as Acting Commissioner
FDA Commissioner Marty Makary resigned Monday under pressure from the White House, ending a 13-month tenure marked by clashes over e-cigarette approvals and mifepristone access. Kyle Diamantas, Deputy Commissioner for Food, is acting head.
The regulatory weather just got harder to predict — again. I flagged Makary’s situation last week. The official departure means any builder in the SaMD pipeline needs to watch who gets named permanently and whether the RAPID pathway survives the transition.
💡 80/20: If you have an FDA submission in flight or planned, add a 30-day buffer to your timeline assumptions. Leadership transitions create review-queue uncertainty even when the career staff are stable.
Your Epic System Has No Index — and Agent Factory Might Fix That
John Lee at Hitdoc makes the case that most health systems running Epic have no systematic metadata governance. He calls it “organizational fibrillation” — build teams diverging without shared standards, order sets multiplying without deduplication, clinical content accumulating without an index.
Edward Elmhurst Health used metadata governance to reduce order sets by one-third. Epic Agent Factory is the AI-powered layer that could automate this kind of build auditing at scale.
💡 80/20: If you’re building inside a health system’s Epic environment, ask your CMIO about metadata records. The build governance infrastructure determines whether your tool integrates cleanly or creates another silo. Lee’s piece is the most practical framework I’ve seen for understanding Epic organizational architecture.
🛠️ From the Workbench
OpenMed: 1,000+ Healthcare LLMs That Never Leave Your Device
OpenMed 1.4.0 ships with 1,000+ healthcare-specific LLM models that run entirely on-device. The Nemotron Privacy Filter handles on-device PHI detection. HIPAA Safe Harbor ready. Apache 2.0 license. Apple MLX and Swift OpenMedKit integration for macOS and iOS.
OpenMed Agent is now in preview — an AI-powered medical assistant for the terminal, private and sandboxed, built for clinical workflows.
For clinician-builders running Ollama or LM Studio, this is the largest curated library of medical models available under a permissive license. The on-device PHI detection layer is the key differentiator — it means you can experiment with clinical NLP without sending patient data anywhere.
💡 80/20: If you have an M-series Mac, install OpenMed this weekend. Run a clinical NLP task on synthetic data (Synthea patient records work well). The experience of running a medical LLM locally — fast, private, free — changes your mental model of what’s possible.
What are you building this week? Reply and tell me — I read every one.
— Kevin