Why a lab report is harder than a PDF

A number on its own is not a result. Haemoglobin of 13.2 is unremarkable in one patient and worth a phone call in another, and the thing that decides which is the reference range it is read against.

That range is not a constant. It moves with age and sex, and it moves with the assay — two laboratories measuring the same analyte by different methods legitimately publish different ranges for it. Which means the interesting engineering problem is not rendering a document. It is being certain, years later, which range a particular number was interpreted against.

The report is a snapshot, not a view

The tempting design is to store results and generate the report on demand. It keeps one source of truth and the document is always current.

It is also wrong, and the failure is quiet. The laboratory changes analyser or updates a range in line with new guidance. Every historic report regenerated after that point silently reinterprets old numbers against new ranges. A value flagged as high in 2024 renders as normal today. Nothing errored, no data was edited, and the clinical record now disagrees with the document the clinician actually read.

So the range in force has to be captured onto the result at the moment of reporting, and the rendered report has to be retained as an artefact rather than reconstructed. What was issued is a fact about the past. It does not get to change because the present did.

Specimen identity is the whole thing

Everything above assumes the number belongs to the right person, which is the assumption the domain spends most of its effort defending.

Between a patient and a result sits a physical object that changes hands several times. Accessioning — assigning an identifier to a specimen and binding it to a person — is where a mismatch enters, and once entered it is nearly undetectable downstream. The result is plausible. It is simply someone else's.

This is why barcodes are load-bearing rather than convenient, and why any workflow that lets a human retype an identifier deserves scrutiny. A wrong result is a bad outcome. A wrong result that looks entirely normal for the patient it was misfiled against is worse, because nothing prompts anyone to question it.

Amended results

Sometimes a result is corrected after the report has gone out. The system cannot treat that as an edit.

The original was issued, it was read, and decisions may already have been made on it. What is needed is a new version, an explicit relationship to the one it supersedes, and a record of who received the original — because the clinical obligation is not to change the number quietly but to make sure the people acting on it know.

A design that overwrites the value and updates the timestamp destroys exactly the information that matters when someone asks why a patient was treated the way they were.

Units, and the conversion nobody logs

The same analyte is reported in different units in different countries, and conversion factors are not always simple ratios.

Convert on display and store canonical values, never the other way round. The failure mode of storing converted values is that a rounding decision made for a screen becomes the stored number, and the original precision is gone. Nobody notices for years, and then someone runs a trend across a period spanning the change.

Critical values need a path, not a flag

A result outside critical limits is not a rendering concern. It is an event that requires a person to be reached and the contact to be recorded.

Systems that treat this as a highlight on a page have implemented the display and not the obligation. The requirement is escalation with acknowledgement — who was told, when, and by what means — and it is one of the first things a reviewer asks to see, because it is the point where a laboratory's duty becomes operational rather than analytical.

Why this gets underestimated

Scoped as a document, this looks like a fortnight of layout work.

It is actually a versioned clinical artefact with a chain of custody, an interpretation context frozen at the moment of issue, and an escalation obligation attached. The PDF is the least interesting part, and it is the part everyone estimates.