LungIQ
Find every patient eligible for screening, even when the chart never says so.
Lung cancer screening eligibility depends on a 20+ pack-year smoking history that is buried in clinical notes or missing altogether. LungIQ reads the notes, diagnoses, and medications already in your records and returns a ranked list of screening candidates, each with the reason they were flagged.
The identification gap
Eligibility hinges on a number most charts never record
The bottleneck in lung cancer screening is not scanner capacity. It is knowing who qualifies. Pack-year history is rarely stored in a structured field, so registry queries miss most eligible patients.
Buried in free text
When smoking history is recorded, it usually lives in a progress note, not a reportable field. A query on structured data cannot see it.
Missing altogether
Often it was never written down. The signal is still there in diagnoses like COPD, cessation prescriptions, and related procedures.
Too slow to find by hand
Reviewing one chart for smoking history takes 15 minutes or more. No team can do that across an entire patient panel. LungIQ is 9x more efficient than manual chart review.
Internal validation study at an Oklahoma FQHC.
About 14.2 million U.S. adults meet USPSTF screening criteria, and nearly 19 million under American Cancer Society guidelines. Fewer than 1 in 5 are screened. Missing pack-year data is also the main reason a lung cancer screening quality measure has not advanced at NCQA.
Sources: American Lung Association, State of Lung Cancer 2022. American Cancer Society lung cancer screening guideline, 2023. Bandi et al., JAMA, November 2025. American College of Radiology public comment, 2025.How LungIQ works
Three steps. Live in days. No IT project.
No EHR integration and no interface build. Send a flat-file export, LungIQ reads the charts, and your team gets a ranked, explained worklist.
Step 1 · Export
Send a flat file. CSV is fine.
Demographics, clinical notes, diagnoses, and medications as delimited text files, under a signed BAA. Any EHR, any column names. We map them at configuration time.
Step 2 · Extract and score
AI recovers what the chart buries
A large language model reads clinical notes for smoking status, pack-years, and nine tobacco-associated risk factors. A machine learning classifier combines them with diagnosis, procedure, and medication signals, and USPSTF rules are applied exactly.
Step 3 · Worklist
Ranked, explained, ready to call
Every patient lands in an action queue, ranked by probability, with a plain-language reason. Your navigators know why a patient was flagged before they pick up the phone.
The output
Every patient lands in a queue with a next step
Not an undifferentiated list. Each queue calls for a different conversation, so they are never pooled.
Documented smoking history that meets USPSTF criteria.
Call. This is the navigator's list.
Known smoker, but packs per day or years smoked were never recorded.
Ask one question. The queue names exactly which fields are missing.
No smoking documentation at all, but a high statistical probability.
Review or reach out, working top-down by probability.
Qualifying pack-years, but quit more than 15 years ago. Outside USPSTF, inside ACS 2023 and NCCN 2025 guidance.
Revisit as guidelines broaden.
Smoking history documented, but excluded, for example over age 80 or prior lung cancer.
No action. The exclusion reason is stated for audit.
Every row carries an explanation column: the specific documented basis for the flag, in plain language. Results come back as a workbook your team can work directly, plus a flat file for loading into your own systems.
Validation
Tested where charts are hardest
LungIQ was externally validated at a rural FQHC network in eastern Oklahoma serving about 35,000 patients across 14 sites. The model had never seen this health system's data and was not retrained for it. A nurse practitioner and a nurse navigator worked the ranked list by phone and confirmed eligibility directly with each patient reached. The entire validation took just 160 hours of staff time.
| Ranked tier | Called | Reached | Confirmed | Precision | Enrolled |
|---|---|---|---|---|---|
| Top 10 | 10 | 4 | 4 | 100% | 1 |
| Top 100 | 100 | 47 | 44 | 94% | 26 |
| Top 1,000 | 1,000 | 420 | 375 | 89% | 182 |
| All called | 3,670 | 986 | 676 | 69% | 273 |
Precision is confirmed candidates divided by patients reached. External discrimination AUROC 0.96.
Yield stays high deep into the list
The probability of being a candidate stayed above 0.87 through roughly the top 2,500 patients. Teams are not limited to skimming the top.
A clear place to stop
Calling stopped when the estimated probability fell below about 5%, where a call cost more than it was likely to return. The ranked list makes that a business decision, not a guess.
Reaching patients was the constraint
Only 27% of patients called were reached. Among those reached, precision held up throughout. Populations that are easier to contact should convert better.
Built from a standing start
Outreach staff were trained during the study, so many early calls were ramp-up. The site also had no reminder system for shared decision-making visits. Programs with trained navigators and patient reminders should see higher contact and enrollment.
Results from the LungIQ FQHC Deployment Report (Oatmeal Health, August 2026). Validated at one external site to date. Because calling followed rank order, low-ranked patients were largely not verified, so the reported AUROC is best read as a conservative floor. Completed LDCT and stage-shift outcomes are not yet measured.
Getting started
Four files, a few business days
LungIQ is a periodic batch process, not a real-time integration. Your EHR analyst can pull everything it needs with a standard export.
What we need
- Demographics. Patient ID, age or date of birth, sex.
- Clinical notes. PCP and pulmonology notes carry about 90% of the smoking signal.
- Diagnoses. ICD-10 codes, with descriptions if available.
- Medications. Free-text drug names are enough.
- Procedures, optional. CPT history adds signal where notes are thin.
What you do not need
- An HL7 or FHIR interface, or a real-time feed
- A specific EHR vendor, schema, or column names
- Structured smoking history fields
- Imaging data or lab results
- Pre-cleaned or de-duplicated data
First export to first worklist typically takes a few business days. Refresh runs every 6 months pick up newly seen patients, patients who age into eligibility, and new documentation.
Security and privacy
Runs on your servers or in our HIPAA-compliant cloud
BAA before any PHI moves
A business associate agreement is signed with every client before data is shared.
On premises or in our cloud
Run LungIQ inside your own environment, or in Oatmeal Health's HIPAA-compliant Google Cloud environment under Google's BAA. Your choice.
Nothing kept after a run
In our HIPAA-compliant cloud, processing is transient. Working data is purged at the end of each batch.
Encrypted end to end
TLS 1.2 or higher in transit and AES-256 at rest.
Your data does not train models
Client data is used only to produce your results.
Least-privilege access
A dedicated, restricted upload location per client, open only to you and the pipeline.
Who uses LungIQ
Built for any organization with a patient panel
LungIQ produces the list. Your own navigators and schedulers own every patient conversation.
Community health centers
Close screening gaps, bill shared decision-making visits, improve UDS metrics, and connect patients to imaging partners.
LungIQ for FQHCsHospitals and health systems
Find the eligible patients already in your own records and fill your screening program, then add LungAI at the point of imaging.
Hospitals and imagingGet started
Find the patients already in your records
Start with a CSV export. Run it on your servers or in our HIPAA-compliant cloud, with no IT project, and get a ranked worklist within days.
LungIQ identifies patients who may be eligible for screening. Eligibility is confirmed with the patient, and screening follows a shared decision-making visit.