Life sciences and real-world evidence · Tidegate
Life sciences and real-world evidence

Know what a dataset holds before you rely on it.

The Tidegate engine shows real counts of what a dataset contains, including what is buried in notes, measured in place, with condition status on every finding. Once you choose a dataset, the engine links records within your data and documents how well it worked, for your submission team and reviewers. Participants with no record in a source are reported as coverage gaps. Synthetic data for development and testing is released only after memorization tests, with a synthetic data report.

Cohort: statin discontinuation · in your data platform RWE-014 · EXCLUSION: PRIOR MI · HISTORICAL
Clinic note Discharge summary Pharmacy claims PT-A-K72M41

Stopped atorvastatin after four weeks due to myalgia. Father had MI at 52. No personal history of MI.

MEDICATION · CONDITIONFAMILYNEGATEDSUGGESTED CODES · ROADMAP
Atorvastatin, stopped at 4 weeksMEDICATION · CITEDSUGGESTED CODES · ROADMAP
MyalgiaCONDITION · CITED
Heart attack (MI), fatherCONDITION · FAMILY
Personal MI historyCONDITION · NEGATED · EXCLUSION NOT MET
Illustrative. Fictional study and patient, Expert Determination mode. Solid: available. Outlined: ROADMAP.
SOC 2 Type I and HITRUST e1 in progress · Runs in a cloud account you control · BAA whenever our engineers access PHI · Never buys or sells data
AFTER DISCOVERY Faster discovery means more evidence to produce.AI is speeding up discovery, so more candidates need trial linkage, external control arms, long-term follow-up and post-approval evidence. Each depends on linkage someone has to defend.
Path to live

From one study to your evidence program

  1. 01BriefWe walk your RWE and data science teams through the Tidegate engine on fictional records and a sample referential integrity report (release-pack document 05). An evaluation on one of your datasets follows once your agreements are in place.
  2. 02PilotOne study or cohort runs inside your data platform. Your team checks the linkage report and coverage gaps before the data is relied on.
  3. 03ProgramMore studies and data sources are added, with match quality within your data and coverage reported for each release. From v1.1: quality broken out by patient group.
AgreementsWhenever Tidegate engineers access PHI, a Business Associate Agreement is in place first. You can buy directly or through your real-world data provider.
What you get

What changes on Monday morning.

For RWE and data science leaders

Evidence whose linkage you can explain and defend.

TodayTeams commit to datasets before anyone knows what they hold or whether they will link. No one can say how well a linkage worked, or for whom. Consented participants drop out at the link, relatives’ conditions sit in the wrong history, and the reasons that matter are buried in free text.
With TidegateYou see real counts of what a dataset holds, measured in place, before you rely on it. Each linkage within your data comes with a referential integrity report, documented for review. Every field carries its permitted uses, and every fact is cited to its source with its condition status.
Participants accounted for at the linkMissed links are measured and recovered where a record exists. Participants with no record in a source are reported as coverage gaps, not hidden.Track: consented participants linked
Linkage you can documentMatch quality within your data, reported in every release in release-pack document 05. From v1.1: quality broken out by patient group (document 11).Track: regulator questions on linkage
Traceable to the pageEvery finding cites its note, page and words, so your inclusion and exclusion logic can rest on evidence.Track: reviewer queries per study
Reproducible releasesVersions and replacement keys are pinned, so a cohort or a release can be rebuilt exactly.Track: time to rerun a cohort
Before you rely on it

See it first, then decide.

What it containsWhen your data provider runs the Tidegate engine, you see aggregate counts of identifiers, clinical entities and the facts in notes and documents before you license a dataset. Record-level data stays with the provider.
Linkage within your dataRecords linked within your data, with match quality reported. On the roadmap: linkage across organizations.
Synthetic data to build againstSynthetic data for development and testing, released only after memorization tests, with a synthetic data report, so your engineers can build before the real data is cleared. Feasibility comes from the real counts, not the synthetic data.
Unstructured data

The evidence is in the notes.

Reasons for stopping therapy, family history, symptoms and social context are rarely coded. They sit in unstructured clinical and case notes. Tidegate reads them, marks each fact’s condition status and cites the note, page and words. On the roadmap: each fact tied to the person it describes.

What it reads
Progress and clinic notesfree text
Discharge summariesfree text · PDF
Pathology and imaging reportsfree text · PDF
Care and case management notesfree text
Outside recordsPDF with a text layer · scans from v1.1
Alongside structured dataOMOP · claims · CSV · Parquet
The problem

Where evidence leaks

WHERE EVIDENCE LEAKS04 GAPS
01Consented participants lost at the linkRecords that should join do not, and the study quietly shrinks.
02Relatives’ conditions in the wrong historyA family history line becomes the patient’s own diagnosis.
03Reasons for stopping therapy written only in notesThe answer exists, but no structured field holds it.
04Linkage quality no one has measuredNo report says how well the linkage worked, or for whom.
Deployment

Inside your data platform

Tidegate runs where your records already sit, alongside the linkage and tokens your data providers already use. No data is pooled or sold. Buy it directly, or through your real-world data provider.

Review

Built for evidence reviewers

Referential integrity reportMatch quality within your data for each study, with coverage gaps reported. From v1.1: quality broken out by patient group.
Cited, versioned findingsEvery finding cites its note, page and words, so your inclusion and exclusion logic can rest on evidence. The release manifest pins engine and rule versions.
Risk summaryA re-identification risk summary with every release, with one replacement per person across every table and note.
Privacy review

Ready for your privacy office.

Protected output is prepared for HIPAA expert determination, with every protection logged and a re-identification risk summary with every release. Tidegate does not make determinations. When your study calls for one, we introduce you to independent, qualified experts and hand off a release ready for their review.

The release pack

What ships with a release

01Field labels with permitted usesAPRIL 2027 RELEASE
02Safe Harbor checklist, or the qualified expert’s determinationAPRIL 2027 RELEASE
03Residual identifier report from a reviewed sampleAPRIL 2027 RELEASE
04Re-identification risk summaryAPRIL 2027 RELEASE
05Referential integrity report: one patient, every mention, every releaseAPRIL 2027 RELEASE
06Release manifest with engine and rule versionsAPRIL 2027 RELEASE
07Date policy and checksFROM V1.1
08Surrogate quality reportFROM V1.1
09Clinical retention reportFROM V1.1
10Synthetic data report (when synthetic data is included)FROM V1.1
11Quality broken out by patient groupFROM V1.1
12Data card: sources, time span, population, transformations, known gapsFROM V1.1
See samples of documents 01 to 06
Coverage

What this means for your data

Real-world data spans decades of coding changes. The Tidegate engine recognizes legacy ICD-9-CM alongside ICD-10, and drug and lab codes including NDC, RxNorm, LOINC and SNOMED CT, so study data can be classified consistently across time and sources.

What the engine recognizes
Designed to support review under FDA real-world evidence guidance Audit-trail evidence for 21 CFR Part 11 programs HIPAA Expert Determination
Sample output

Questions we hear first

Are date intervals kept?
In Expert Determination mode, yes: one date offset per person keeps intervals and ages. Safe Harbor mode keeps only the year.
How is linkage quality measured?
Within your data, reported in every release; see release-pack document 05. From v1.1: broken out by patient group.
We hold trial and registry data; does this apply?
Yes. The same engine runs on data you hold.
Technical fitBatch jobs inside your data platform.
MODEBatch jobs or SQL over notes and OMOP tables
INPUTSUnstructured clinical notes, documents with a text layer, OMOP, FHIR, claims
OUTPUTSOMOP copy with the same schema and field labels, NOTE_NLP rows with citations, CSV and Parquet, link table, referential integrity report and risk summary. From v1.1: FHIR with Provenance
Engine details