ClinicalMiner — Follow the thread

The thread

A clinical archive is a labyrinth. This is the way through.

Scanned letters, op notes, discharge summaries, echo reports — years of them, in no particular order. Somewhere inside is the thing you actually need. ClinicalMiner unwinds through your documents and leads you to it, station by station.

Extract the thread goes in

Drop in PDFs — discharge summaries, op notes, diagnostic reports — or enter patients directly with the structured rapid-entry form. Claude, Anthropic's AI, reads the documents; you stay in charge.

De-identify it never lets go of a name

Before anything can leave your machine, a local multi-layer engine (pattern rules, curated gazetteers, local language models — Greek and English) removes names, IDs, dates and places. An independent guard then re-scans every outgoing document and holds anything suspicious for your review; every send is recorded so you can inspect the exact text that left, any time. Prefer nothing leaves at all? A hard offline mode blocks outbound connections entirely.

Structure the labyrinth becomes a database

Findings land in a validated database: ICD-10 and procedure coding (ΕΤΙΠ / OPCS-4), completeness dashboards, duplicate detection, cell-level change history, follow-up worklists — encrypted at rest, per-account, with Touch ID unlock.

Analyse assumption-checked, reference-validated

Kaplan-Meier with proper censoring, log-rank and Cox, Table 1, regression, ROC, power calculations, meta-analysis — guided flows recommend the right test and say why. Every statistic is cross-checked against SciPy, statsmodels and lifelines.

Publish and back out into the light

Journal-style figures (colour or grayscale-for-print), one-click Word reports with ready-made methods text, labelled exports for statisticians, and a per-project archive of every analysis you ran.

The find.

The cohort, the complication rate, the survival curve, the paper waiting to be written — reached with nothing lost, and no identity given away along the way.

“The thread never lets go of a name.” De-identification happens before anything leaves the room — not in someone else's cloud.

No ClinicalMiner cloud. Your database lives in encrypted project files on your computer. There is no server-side copy.
A second, independent check on every egress. Each outgoing document is re-scanned by a separate privacy guard; flagged documents are held for your review, and per-document sent-records let you audit exactly what left — and that's only if you use the AI features at all.
Evidence, not promises. De-identification tested on public benchmarks; re-identification risk analysis and a plain-English transparency view built in.
Designed for GDPR-conscious environments. Audit log, retention controls, encrypted passphrase-sealed backups, idle lock, per-account separation.

Featuring

MorpheusDe-identification engine

The first purely local de-identification engine for Greek clinical documents.

Greek and English, in one pass. Names, dates, addresses, hospital and phone numbers are replaced with realistic stand-ins — while diagnoses, drugs, eponyms and ward names stay exactly as written.
Runs entirely on your computer. An ordinary laptop, no GPU, no server, no internet needed. Nothing is read by anyone but the engine.
Same document in, same result out. Byte-identical every time, with a self-test on every start — so what you released yesterday can be audited today.
Measured, not promised. Tested on public benchmarks; a second, independent guard re-checks anything that leaves, and flagged documents wait for your eyes.

More about the De-ID engine

Morpheus is built into ClinicalMiner — every document is passed through it before any AI feature can see it. No de-identifier is perfect; Morpheus is designed so that what it misses is caught, logged and reviewable.

AI extraction

Years of PDFs. One afternoon.

Drop in discharge summaries, operation notes, clinic letters and diagnostic reports. Everything is de-identified on your machine first; then Claude reads what is left and fills your database, field by field, with you checking as it goes.

You don't start from a blank sheet. Nineteen specialty templates come ready-built — the fields your specialty actually records, already chosen, already coded, already validated.

Adult Cardiac SurgeryPaediatric Cardiothoracic Thoracic SurgeryVascular SurgeryNeurosurgery General SurgeryUrologyENT / Head & Neck OrthopaedicsCardiologyOncology NeurologyHaematologyRespiratory Gastroenterology & HepatologyNephrology Endocrinology & DiabetesEmergency Medicine General Internal Medicine 10 medical · 9 surgical

DemographicsAge, sex, identifiers, referral source

DiagnosesICD-10 coded, primary and secondary

ProceduresΕΤΙΠ or OPCS-4, with dates and operators

OutcomesMortality, survival, length of stay, follow-up

ComplicationsSpecialty-specific, structured

Your own fieldsAdd, rename or remove anything

The Data Extraction tab: documents queued on the left, the fields to extract chosen on the right, and two extracted patients in the preview grid below — **Documents in, structured patients out.** Choose the fields — or let it choose them — then check every extracted value before it goes near your database. Right-click any of them to see the exact sentence in the document it came from.

Screenshots show synthetic patients. No real record appears anywhere on this site.

Database

Everything a working clinical database needs, in one place.

Specialty-agnostic: start from a template, shape the fields to your practice, and import what you already have.

Fast manual entry

A structured patient form with rapid-entry mode, live validation (hard rules block, soft rules warn) and clinical calculators such as EuroSCORE II built in.

AI extraction from documents

Structured fields extracted from PDFs after local de-identification, with crash-safe recovery on long runs — and linked diagnostic reports (echo, CT, Holter) per patient.

Extraction uses Claude by Anthropic.

Coding systems included

ICD-10 diagnoses (Greek ΚΕΤΕΚΝΥ or WHO English) and procedures in ΕΤΙΠ or OPCS-4 — switchable per project, with search and snap-to-code.

Data quality, continuously

Completeness dashboard, per-project quality rules, duplicate review-and-merge, and a cell-level change history of who changed what, when.

Follow-up that runs itself

Overdue / due-soon worklists from your own dates, bulk tools that respect deceased patients, and serial measurements charted over time since operation.

Your data, portable

CSV/Excel import with column mapping, labelled exports for statisticians, and encrypted backups that move whole projects between machines.

The Data Table showing a cohort of patients with diagnoses, procedures and outcomes — **The Data Table.** Every patient, every field, in one editable grid — search, filter, frozen columns, cell-level history of who changed what, and a right-click back to the source document.

Analysis

The find, not the search.

A guided statistics workspace that recommends the right test, checks its assumptions out loud, and writes the methods sentence for you — with just a few clicks.

Survival analysis done properly — Kaplan-Meier with censoring, number-at-risk tables, freedom-from-reintervention, log-rank, Cox regression.
The full working toolkit — Table 1, t-tests to Kruskal-Wallis, correlation and regression (logistic with odds ratios), ROC curves, power calculations, meta-analysis.
Cross-checked, not hand-rolled — every statistic validated against reference implementations (SciPy, statsmodels, lifelines) by an included open validation kit.
Ready for the manuscript — journal-style charts, one-click Word reports with methods text, and saved analyses per project.

Kaplan-Meier survival curves for two procedures, with censoring marks and a number-at-risk table — **Kaplan-Meier, done properly.** Censoring marks, a number-at-risk table and a log-rank test — not a line chart that happens to go downwards.

Cohort overview table showing counts, percentages, medians and interquartile ranges — **Table 1, written for you.** The cohort overview every paper opens with — counts and percentages, medians and interquartile ranges — built from your database the moment you open it.

Figures produced by the app from a synthetic cohort.

FAQ

Questions clinicians ask first.

Where does my patient data live?

On your computer, in encrypted project files, separated per user account. There is no ClinicalMiner cloud and no server-side copy of your database.

Does patient-identifiable data ever reach the AI?

The AI features use Claude, Anthropic's AI model, and operate only on text that has been de-identified locally first — names, identifiers, dates and places replaced on your machine. A separate egress guard re-scans every outgoing document and holds anything it flags for your review before sending. Every send is recorded — in a tamper-evident audit log and in per-document sent-records you can inspect. If you never use the AI features, nothing leaves at all, and a hard offline mode can block outbound connections entirely.

Is it only for one specialty?

No — the database is template-based and specialty-agnostic. Field sets, procedure lists and report types are content you can shape to your practice, from cardiac surgery to any other discipline.

Can I trust the statistics?

The statistics engine is built on the same open scientific libraries used in published research (SciPy, statsmodels, lifelines), and ships with a validation suite that cross-checks every result the app produces against those reference implementations.

What if I change computers?

Your account and your patients live on the machine that made them — there is no cloud to sign into from somewhere else, which is exactly what keeps the data yours. To move, you take an encrypted backup on the old computer, carry the file across, and restore it on the new one. Your projects come with you; you create a fresh account on the new machine, which takes a moment. The app explains all of this under Help, and the passphrase you choose is what opens the backup — so keep it. If your department has an administrator, they can hold a recovery key that opens a backup when someone forgets theirs.

What does it run on?

macOS today; a Windows build is in progress. It is a real desktop app — your data and the de-identification models run locally, so no internet connection is required for day-to-day database work.

Follow the thread.