Verify a new device
Device, reference (a connected gold device, the main lab, or manual), analytes and allowable error. Pairs are drawn from results, precision from QC, and the pack builds itself.
Home / Platform / Statistics Workbench
Add-on module
Verify a new analyser in an afternoon, with the file to prove it.
Precision, comparison, linearity, detection limits, reference intervals, measurement uncertainty and sigma, run on data the platform already holds, written up as a signed verification pack. No spreadsheet, no statistics package, no arguing about the maths.
- 30 tools, 12 guided scenarios
- CLSI EP05 / EP09 / EP15 / EP06 / EP17 / EP28 methods
- Signed packs locked with SHA-256
- Verified against textbook values
Start from the question
Twelve scenarios, each a wizard.
Precision
Five-day EP15 style or twenty-day EP05 style estimates of repeatability and within-laboratory precision.
Comparability
Two devices, two sites or device versus main lab, paired by patient, with Deming regression and bias at decision levels.
Operators
QC grouped by operator, so a training need shows as a statistic rather than an argument.
A QC problem
Shift, trend, outliers and rule power on the affected series.
Measurement uncertainty
ISO/TS 20914 approach from IQC and EQA data, per analyte.
QC design
Sigma per analyte, the method-decision chart and Westgard Sigma Rules, with Monte-Carlo rule power and a patient-based QC what-if.
Lot change
New lot against old with the acceptance decision recorded.
Linearity
EP06 polynomial evaluation across the measuring range.
Reference interval
Establish, transfer with 20 samples, or partition with Harris-Boyd.
Qualitative
2x2 diagnostic accuracy with Wilson intervals, kappa, Fisher exact, ROC.
Free analysis
Any tool on any dataset, pasted or pulled from the platform.
What comes out
One result contract, one signature, one file.
The numerics are pure, unit-tested code verified against published worked examples, so the same input always gives the same answer.
- Every tool answers the same way: summary, verdict, interpretation, tables, charts and the raw data used.
- Run, finalise, sign: a finalised analysis is locked with a SHA-256 hash. Unlocking needs an administrator and leaves an audit row.
- Verification packs: an analyte by tool matrix of verdicts, one signature over the lot, and a combined print-to-PDF report with the charts drawn server-side.
- Allowable error from the right place: your own specification first, then CLIA 2024, RCPA or EFLM biological variation from the library, with entries that need checking flagged.
- Sigma dashboard: CV from IQC, bias from EQA or an assigned target, TEa from your spec or the library, per analyte, feeding the dashboard widget.
| Standard | What the workbench does with it |
|---|---|
| CLSI EP15-A3 | User verification of precision and estimation of bias |
| CLSI EP05-A3 | Precision over 20 days, repeatability and within-lab |
| CLSI EP09c | Method comparison, Deming with jackknife confidence intervals |
| CLSI EP06 | Polynomial linearity across the range |
| CLSI EP17 | Limit of blank, detection and quantitation |
| CLSI EP28 | Reference intervals, transference, partitioning |
| ISO/TS 20914 | Measurement uncertainty from IQC and bias data |
| Westgard Sigma Rules | Rule and control-number selection from sigma |
A verification, end to end
From a new analyser on the bench to a signed file in the folder.
Choose the device and the reference
The new analyser, and what it is being compared to: a connected device already trusted, the main laboratory, or values entered by hand.
Pick analytes and allowable error
Your specification if you have one; otherwise CLIA, RCPA or EFLM biological variation from the library.
Let the data come to you
Paired patient results are drawn from both devices automatically; the 5x5 precision grid fills from QC runs. Anything missing can be pasted.
Read the acceptance table, sign the pack
Each analyte gets a verdict for precision and comparison. The pack is signed once, locked, and printed as the verification file.
A clear boundary
The workbench describes methods, processes and datasets. It never interprets a patient result and sits outside the scope of software as a medical device. Catenix is connectivity, workflow, record-keeping and data display, and provides no clinical decision support.
Questions, answered
The questions teams ask first.
Do we have to export data to a spreadsheet first?
No. The workbench pulls patient series, QC series (with target and SD), QC grouped by operator, device, lot or month, and paired-by-patient comparisons between two devices straight from the platform. You can also paste from Excel into a grid.
Which methods are implemented?
Precision (EP05 and EP15 style), method comparison with ordinary least squares and Deming regression with jackknife confidence intervals, EP06 polynomial linearity, EP17 limit of blank, detection and quantitation, EP28 reference intervals including 20-sample transference and Harris-Boyd partitioning, ISO/TS 20914 measurement uncertainty, sigma metrics with the method-decision chart and Westgard Sigma Rules, Monte-Carlo rule power, patient-based QC what-if, Broughton carryover, drift, sample-size planning, diagnostic accuracy with Wilson intervals, kappa, ROC, and the usual parametric and non-parametric tests.
Where do the allowable-error limits come from?
From your own specifications first. If none is set, the built-in library offers CLIA 2024, RCPA and EFLM biological-variation limits (desirable, minimal and optimal) keyed by analyte. Library entries that need checking before a formal verdict are flagged.
Can an analysis be tampered with after sign-off?
A finalised analysis is signed and locked with a SHA-256 hash. Unlocking needs an administrator and is audited. A verification pack carries one signature over every analysis in it.
Does the workbench interpret patient results?
No. It describes methods, processes and datasets. It never interprets an individual patient result and sits outside the scope of software as a medical device.
Bring the analyser you are about to verify.
We will run the verification scenario on a live tenant with your analytes and your allowable-error limits.