Clinical research platform GCP · CDASH · SNOMED-CT · LOINC conventions
AnzarSeha
Medical research
Traceable medical research

From raw records to a dataset

Protocol locked, database frozen, every value traced back to its source. AnzarSeha turns clinical records into a research dataset you can defend in front of a reviewer.

Go to your space See how it works
AnzarSeha registry and analysis: a 60-patient cohort, mean IPSS by follow-up milestone, Clavien-Dindo complications and active strategies, all computed live.
Registry & analysis — indicators computed live on the cohort Real screenshot · demonstration account, fictional data
Trusted partner HUIM6 Rabat Pilot site

The problem

a.Two hundred records opened one by one, copied out by hand into a spreadsheet.
b.A column nobody can explain three months later.
c.“How did you measure the primary endpoint?” — and no answer to give.

AnzarSeha closes the gap between the record and the result.

The chain, six steps

From source to export

Every screen below is a real screenshot of the application. Nothing is lost between two steps: each one inherits what the previous one froze.

Open a step to see the matching screen ↓

Patient records live in a structured clinical registry, not a spreadsheet. One source, typed and versioned, that every study in the department builds on.

Structured entryAuthenticated accessRow-level isolation
The patient record registry: 60 records, Active / Lost to follow-up / Alerts filters, patient cards with strategy, IPSS score and change.
Patient records — the department registry, every pathology

You open a study: it inherits a typed variable dictionary — role, scale, UCUM units, plausible bounds, and CDASH, SNOMED-CT, LOINC and ICD-10 mappings. The meaning of every column is written before the first value, and the primary endpoint is declared here, before collection begins.

Role & scaleUCUM unitsCDASH conventions
The study form: each variable carries its technical key, type, unit and registry source; Maximum Clavien-Dindo is marked Primary endpoint.
The study form — “Maximum Clavien-Dindo” declared primary endpoint

Values arrive from the registry, from deterministic rules, or from a model that must quote the exact sentence in the source document. The AI never produces a final value: a human reviews record by record, and every value keeps its origin and its justification.

Quotation requiredHuman validationPseudonymised before sending
The review screen: variable, value, origin and justification columns; each value carries a Registry, Rule, Model or Verified badge.
Record-by-record review — value · origin · justification

The protocol locks before collection — and the lock refuses to close while the primary endpoint is undeclared. The database freezes before analysis: the dataset you analyse is exactly the one you exported. Reopening requires a reason, kept in the log.

Protocol lockedDatabase frozenNo endpoint chosen afterwards
The governance panel: a Protocol card badged Locked, a Database card badged Frozen, each dated and signed.
Governance — lock and freeze dated, signed, reversible with a reason

Every action is written to a log where each event is sealed by the fingerprint of the one before it. Alter a past entry and the chain visibly breaks: tampering shows instead of hiding. ALCOA+ by construction, not by procedure.

Who · what · whenVisible breakALCOA+
The audit log: a Chain intact badge, numbered events — study modified, protocol locked, database frozen — signed and timestamped.
Audit log — chain intact, every event signed

One record per row, one variable per column: the raw material of statistical analysis, which opens in SPSS, R and pandas without a single manual edit — together with the dictionary, the metadata, the missing-data table and a STROBE-format draft.

Four formatsSTROBE draftInstitutional PDF
The cohort matrix: one record per row with pathology, age, active strategy, scale, initial and current scores, and change.
The analysed cohort — what the data sheet will contain
Trust

What holds up in front of a reviewer

Locked

Locking and freezing

The question is frozen before collection, the values before analysis. The goalposts do not move mid-match.

Chain intact

Hash-chained audit log

Any change to a past entry visibly breaks the chain. ALCOA+ traceability by construction.

Quotation required

An AI that cites its source

The model proposes, it does not decide: every proposal carries the sentence from the document it came from, and a human confirms it.

Before any call

Pseudonymisation

A consent gate precedes every AI request; direct identifiers are stripped before sending.

Closed by default

Data isolation

Row-level security, closed by default. The browser holds no key; every access is authenticated.

No claims

Conventions followed

We follow these conventions and reference systems. We claim no certification.

GCP / ICH E6(R3)CDISC CDASHSNOMED-CTLOINCICD-10UCUM
What comes out

One workbook, six sheets

Opens in SPSS, R and pandas without a single manual edit. The methodologist receives a dataset, not a building site.

01 Data one record per row
02 Dictionary the meaning of each column
03 Summary counts and completeness
04 Metadata study, protocol version, freeze
05 Missing what is missing, and why
06 Decisions the choices that made the dataset
The export dialog: content to export, PDF, Word, Excel and CSV formats, 34 selectable columns, timestamped file name.
The export dialog — four formats, columns of your choosing, timestamped file name

The rhythm of a thesis

01

Design

The question, the variables, the protocol. Everything is decided before seeing a single value.

02

Collect

Values arrive with their source, their units and their bounds. The log follows.

03

Close

Freeze the database, export the workbook, draft in STROBE. The study goes to analysis.

SehaInsight

Coming

A question asked in plain language; the answer is computed on the server, from your frozen dataset.

Small counts are suppressed, every query is logged, and the model never receives patient-level rows. Assistive, never deciding.

Frequent questions

What is AnzarSeha?

AnzarSeha is a clinical research platform that turns patient records into an analysable, traceable dataset. The protocol locks before collection, the database freezes before analysis, and every value keeps the source it came from.

Where is the data hosted?

The data is hosted on infrastructure located in the European Union. Every access is authenticated, every study is isolated from the others by row-level security that is closed by default, and the browser holds no key.

Can I export and walk away?

Yes, at any time. The full workbook — data, variable dictionary, summary, metadata, missing-data table and decisions — exports in open formats: Excel, CSV, PDF or Word. It opens in SPSS, R and pandas with no manual editing.

Does the AI see patient names?

No. Direct identifiers — name, record number, contact details — are stripped from the text before any call to an external model, and a consent gate, written to the log, precedes every request. With no recorded decision, only the rules engine runs.

Is it GDPR-compliant?

The application is built for compliant use: data minimisation, pseudonymisation before any external call, access logging and hosting in the European Union. The compliance of a given study remains that of its data controller, not that of the tool.

How long does it take to start a study?

As long as it takes to write your question and choose your variables. The typed dictionary and the export workbook are generated when the study is created, from the pathology catalogue: staging, work-up, scores and biological markers are already there.

Do I need real data to try it?

No. The demonstration runs on sixty fictional records, with a study workbook already assembled and an export ready to download. No real patient data is required to evaluate the tool, and none should ever be entered into it.

Who is AnzarSeha for?

For hospital departments running studies on their own records: department head, practitioner, resident writing a thesis, methodologist. The study form is generated from the catalogue of the pathology under study; the demonstration covers a urology department.

How is this different from a spreadsheet?

A spreadsheet has no types, no units, no sources and no log: nothing stops anyone renaming a column or choosing an endpoint after seeing the results. AnzarSeha freezes the protocol before collection and writes every action to a chained log.

Every screenshot on this page comes from the demonstration account — fictional patients only, no figure here is a result.

Page last updated

Your space

Open your research space

The demonstration is already running: sixty fictional records, a study workbook, an export ready to download. No real data is required to try it.

Go to your space