Create Jira Ticket
This idea cannot be sent to Jira.
CAIRN by GREENBOX Labs | Community Awareness for Invisible-Illness Reference and Navigation | An Early-Lyme Read
CAIRN by GREENBOX Labs is a free, on-device early-Lyme read designed to avoid telling patients they are fine when they might not be. GREENBOX Labs is an engineering team building the Trust Layer for AI: software that turns public data into clear answers and remains transparent about its uncertainty.
CAIRN started at our kitchen table. Founder and CEO Lisa Yang watched her mother, brother, and sister lose quality of life to Lyme that early tests missed, caught in the confusion of symptoms no one connected. CAIRN is the tool we wish our family had during those first weeks.
It runs on any phone or computer with no account, login, or personally identifiable information. CAIRN joins open federal data with a machine-learning spatial risk model that states its own uncertainty, alongside a transparent, non-diagnostic read built directly from published Centers for Disease Control and Prevention (CDC) and Infectious Diseases Society of America (IDSA) guidelines. It returns a plain-language result, a one-page doctor summary, and a standard medical-records file (HL7 FHIR R4 mapped to SNOMED CT) to support patient-carried data portability.
By avoiding a false all-clear, CAIRN advances health equity for the patients this sprint targets. A 2026 Johns Hopkins study (Rebman et al.) found women, particularly premenopausal women, are least likely to test positive in the first weeks and more likely to have symptoms misattributed. CAIRN takes those early signs seriously for every user.
CAIRN is powered by our GREENBOX Clone methodology, which transforms data and guidelines into a transparent decision tool. Our proprietary architecture centers on how we structure the model, quantify uncertainty, and bound critical decisions. While the underlying framework generalizes to any condition where early testing routinely fails patients, expanding it requires a dedicated Clone engineered specifically for that disease's epidemiology and clinical reality.
Solution Overview (MVP Demonstration):
User-facing Application
Live demo, prototype, or working product
Access instructions (if needed)
CAIRN is live, barrier-free, and requires no app store install, account creation, or login.
Zero-Friction Access. Open the link in any modern browser on a phone or computer. Enter a state and ZIP code or county to establish a localized environmental baseline, with no personally identifiable information collected.
Two-Minute Workflow. Complete a short, plain-language checklist covering the last three weeks, including potential tick exposure, rashes, and acute symptoms.
Actionable Clinical Handoff. Receive an immediate early-Lyme read with a a calibrated local Lyme-activity tier, paired with a concise one-page summary structured to support a discussion with a provider.
Full Model Transparency. Open the Model Card page to review every published CDC and IDSA guideline applied to the evaluation.
Fastest Review Path. Home > Get Your Read > Model Card.
Results for the American People:
CAIRN turns open federal data into a clear, privacy-first read for patients and providers.
The Care Gap. Each year, an estimated 476,000 Americans are diagnosed and treated for Lyme, yet only about 89,000 appear in the official CDC 2023 count. The roughly 387,000 missing are not a data error. They are patients dismissed after an early blood test the CDC notes frequently misses Lyme in the first weeks, when a short antibiotic course works best. Many face prolonged health and work impacts from an infection treatable early.
The Equity Impact. These testing limits hit specific groups hardest. A 2026 Johns Hopkins study found women, particularly premenopausal women, are least likely to test positive early.
The Solution. Sara pulls a tick off her leg after a hike. Days later she is feverish and aching, but her rapid test is negative. Unaware early tests often miss Lyme, she opens CAIRN on her phone. In two minutes she sees Lyme activity in her state, aligns her symptoms with CDC and IDSA checklists, and generates a one-page doctor summary. Rather than a false all-clear, it shows the guidelines used and flags that her symptoms warrant a closer clinical look despite the negative test.
Open Access and Community. CAIRN is free, runs on any device, and needs no login. If Sara chooses, her on-device read adds one de-identified point to a public map of tickborne activity, with no personally identifiable information. Helping herself helps the community.
Data, Tech & Evidence Driven Approach
CAIRN runs on U.S. open data and uses AI only where it earns trust.
The Data. Combines CDC-reported county Lyme cases, normalized to a rate using U.S. Census Bureau population, with six federal environmental signals: tick establishment and pathogens (CDC ArboNET), temperature (NOAA), warm-season moisture (gridMET), forest cover (USGS Land Cover), and homes at the forest edge (U.S. Forest Service Wildland-Urban Interface). Missing county data falls back to the state rate, and areas lacking both are omitted entirely. Every data point is public, dated, and linked directly to its source.
The AI. Employs a ridge regression machine-learning model to estimate local Lyme expectation from environmental signals. Validated via whole-state holdouts to ensure generalization rather than memorization, achieving a held-out Spearman correlation of 0.62 (lifted from 0.46 by local tick surveillance). Outputs a calibrated tier (Elevated, Moderate, or Minimal) rather than a false-precise number, with 90% accuracy on held-out counties.
Deterministic Logic (No AI). The clinical read uses transparent, deterministic rules based on CDC and IDSA guidelines, displaying each applied rule directly. AI estimates environmental conditions, but published guidelines dictate the read.
Trust and Privacy. Demographics (sex and age) are intentionally excluded from the read, as state-level CDC data contradicts a national confirmation gap. All calculations compute on-device in under a second without server round-trips. De-identified summaries are stored strictly on an opt-in basis without PII, and the app generates a coded medical-records file (HL7 FHIR R4 mapped to SNOMED CT via a fixed, deterministic lookup table) on-device to support patient-carried data portability.
Execution, Efficiency & Delivery
CAIRN is a functional solution today, operating with the following core capabilities.
Live Guardrails. Displays Lyme activity by state and season directly from public CDC data, with explicit dates on metrics so an old number is not passed as current.
Reproducible Rules. Evaluates published CDC and IDSA early-Lyme guidelines on-device offline, displaying each applied rule to ensure consistent, reproducible outputs for identical inputs.
Bounded AI. Integrates a machine-learning spatial estimate of local Lyme activity based on six federal environmental signals, presenting a calibrated tier estimate without altering the underlying clinical rules.
Transparent Guidance. Generates plain-language assessments (such as flagging that care is warranted now, suggesting close monitoring, or indicating non-elevated risk) drawn directly from CDC and IDSA guidelines and explicitly linked to the triggered rules.
Clinical Handoff. Produces a concise clinical summary and a standardized, de-identified medical-records file (HL7 FHIR R4 mapped to SNOMED CT) on-device, enabling patient-carried data portability across providers.
On-Device Efficiency. Executes its compute directly on the user's device without server dependencies, eliminating per-query inference costs and reducing breach exposure by avoiding centralized medical databases. Operates at minimal infrastructure overhead while scaling without the recurring computational expenses associated with server-side LLMs.
Scalability, Reuse & Societal Impact:
CAIRN is built from three reusable parts: open federal data, an AI model that estimates risk while stating its uncertainty, and transparent published guidelines for the clinical read. This combination is a replicable pattern, not a one-off.
Rapid Condition Swapping. Adapting to new conditions is highly efficient. By swapping in appropriate federal surveillance data, retraining the environmental model, and loading published symptom checklists, the framework pivots quickly. Long COVID, Fibromyalgia, Ehlers-Danlos, and endometriosis could follow this path, sharing the systemic challenge of patients dismissed when early or routine testing comes back normal.
Semantic Interoperability. To enhance portability, CAIRN generates a SNOMED CT coded records file, designed to help a patient's evidence follow them to subsequent providers and reduce the need for specialists to start from zero. This aims to mitigate the fragmented care that prolongs complex invisible illnesses.
Federal Reuse. Integrating open public data with published guidelines demonstrates a transparent approach to trustworthy AI. Avoiding opaque modeling, it provides clear visibility into risk estimates and deterministic rules, making CAIRN a strong candidate for reuse as an adaptable care-navigation template.
Lisa, CAIRN may be the most trust-worthy design I’ve seen in this sprint: no account, everything computed on the device, clinical rules taken straight from CDC and IDSA guidance with each applied rule displayed, and a machine-learning model that is used only where it earns that trust. I especially admire three choices: validating with whole-state holdouts so the model can’t just memorize, reporting a calibrated tier (Elevated, Moderate, Minimal) in place of false precision, and building a tool whose whole purpose is to avoid a false all-clear. Starting from your own family’s experience makes the mission unmistakable.
You and I start from the same finding. Kugeler et al. (Emerg Infect Dis 2021) estimated from insurance claims that about 476,000 people were diagnosed and treated for Lyme each year from 2010 to 2018, against about 35,000 reported through CDC surveillance in the same period (chart attached). The authors say the gap mixes under-reporting with possible overdiagnosis and overtreatment, and the data can’t separate the two, which is why a tool that never claims a diagnosis, like CAIRN, is the right shape. Because CDC changed its surveillance case definition in 2022, I find the same-period comparison the cleanest way to state the gap, and it makes your care-gap argument even harder to dispute.
My entry turns papers like this one into simulators you can change and check against federal data: https://invisibleillness.crowdicity.com/post/896844
Chaitra’s LymeSentinel also predicts county-level tick status from environmental signals and reports held-out performance, and Victoria’s One Health Atlas keeps missing county data visible, much as CAIRN handles gaps with its state-rate fallback. Chaitra, Victoria, I’d be glad to hear your thoughts on how your approaches compare with Lisa’s.
Congratulations to you and the GREENBOX team on a polished, responsible, family-driven tool. I’m cheering you on.
Users tagged:
Reply 0