Inertia Linter
Inertia Linter detects where chronic disease treatment has stalled, classifies why, and resolves barriers — without sending patient data to an LLM.
Core trick
A deterministic TypeScript rule engine scores guideline gaps and sends only abstract status codes to the LLM, which drafts prior-auth forms.
Tags
Description
Inertia Linter detects where chronic disease treatment has stalled, classifies why, and resolves barriers — without sending patient data to an LLM.
I'm a 20-year cardiologist. I built this for a problem I see weekly — patients whose meds aren't optimized despite clear guidelines.
One click audits treatment across diabetes, heart failure, and hypertension, producing an Optimization Score (0–100) with 15 blocker codes. Zero blame, maximum clarity.
Detection alone doesn't save patients. The Resolution Layer auto-generates PA forms with ICD-10 and guideline citations, suggests $4/month bridge prescriptions, and switches to generics. 30-minute workflow becomes 30 seconds.
A deterministic TypeScript rule engine handles all clinical logic. Opus 4.6 receives only abstract status codes — never patient data. PHI protection is architectural, not policy.
Only 1% of patients are on optimal therapy. I built this to close the gap between what guidelines say and what patients receive.
README excerpt
# Inertia Linter **A clinical inertia auditor that detects where chronic disease treatment has stalled, resolves barriers in 30 seconds instead of 30 minutes -- without ever sending patient data to an LLM.** Built by [Yusuke Fukushima, MD (@shimayuz)](https://x.com/shimayuz) | Cerebral Valley AI Hackathon 2026 --- ## The Problem Clinical inertia -- the failure to intensify therapy when treatment goals aren't met -- affects every chronic disease. Only **1%** of eligible heart failure patients receive all four guideline-directed drug classes at target dose ([CHAMP-HF, JACC 2018](https://doi.org/10.1016/j.jacc.2018.04.070)). **6.7 million** HF patients in the US alone ([HFSA 2024](https://doi.org/10.1016/j.cardfail.2024.01.001)). Over **70% mortality reduction** is achievable with full optimization ([Vaduganathan et al., LANCET 2020](https://doi.org/10.1016/S0140-6736(20)30748-0)). The guidelines exist. The drugs exist. The patient is sitting right there. And nothing changes. ## What It Does ### Detection One click runs a full audit across multiple disease domains: - **Heart Failure (HF-GDMT)**: 4-pillar analysis (ARNI/ACEi/ARB, Beta-blocker, MRA, SGLT2i) with GDMT Optimization Score (0-100) - **Diabetes Management (DM)**: HbA1c target tracking, SGLT2i/GLP-1 RA gap detection - **Hypertension Control (HTN)**: BP target compliance, medication class coverage Each pillar is classified as `ON_TARGET`, `UNDERDOSED`, `MISSING`, or `CONTRAINDICATED`, with 15 blocker codes exp