Office files & health data (Word · Excel · CSV)

In one line

Every hospital in the world runs on Excel somewhere. It is the most widely used health analytics tool on earth, it is genuinely useful, and it corrupts data silently and confidently — which is a combination worth understanding properly rather than sneering at.

Why this belongs in a health informatics knowledge base

Not to teach you to bold text. Because the interface between the office suite and clinical systems is where an enormous amount of real health data actually lives and dies.

The typical flow, in every health system on the planet:

HIS / EHR  →  CSV export  →  Excel  →  pivot table  →  chart  →  Word / PowerPoint  →  a decision

Every arrow is a place data changes shape, loses provenance, or breaks. The clinician making the decision at the end sees a clean chart. Nothing tells them what happened in between — and the answer is usually "more than you'd like."

What Excel does to clinical data, without asking

These aren't hypotheticals. They are the default behaviour:

It converts things to dates. The famous case: genes. SEPT2 becomes 2-Sep. MARCH1 becomes 1-Mar. Ziemann et al. found gene-name errors in a fifth of papers with Excel supplementary files — in the peer-reviewed literature. If it does that to genetics, consider what it does to an ICD-10 code, a specimen ID, or a lot number.

It strips leading zeros. A patient ID of 007412 becomes 7412. Silently. Now it doesn't match the source system, and your join drops rows — the wrong rows, since only some IDs have leading zeros.

It mangles long numbers. Anything over 15 digits loses precision — an ABHA number, a device serial, an accession number — because it's stored as a float. The last digits become zeros and look plausible.

It guesses encodings. Indian names in a UTF-8 CSV opened by double-clicking become mojibake, because the default import assumed something else.

It has a row limit. ~1,048,576. Exceed it and older Excel just stopped, without an error — which is precisely what happened in the UK's Test and Trace incident in 2020, where thousands of positive COVID cases went unreported because a file silently ran out of rows during a pandemic.

Notice the pattern: none of these produce an error message. The file opens. The numbers look fine. That's what makes it dangerous rather than merely annoying.

CSV is not a format

The other half of the problem. "CSV" is a convention, not a standard people follow — RFC 4180 exists and is widely ignored:

• What's the delimiter? Comma — unless the locale uses comma as a decimal separator, then semicolon. Same file, different meaning by machine.
• What about commas inside fields? "Sharma, Anita" needs quoting; a naive exporter won't.
• Line endings, encodings, BOMs. Windows and Unix disagree. A BOM breaks naive parsers.
• There are no types. Everything is text until someone guesses. Excel is the someone.
• There is no schema. Column order changes between exports and nothing notices.

CSV survives because it is universal and trivial. Both remain true, and neither makes it safe.

The honest defence of Excel

It would be easy — and wrong — to end there. Excel is not the enemy:

• It is immediate. A ward manager answers a question in four minutes that would take IT four weeks.
• The formulas are visible. You can see the logic, unlike most dashboards.
• Everyone already has it and knows it. That's worth more than most tooling arguments admit.
• It is frequently the only tool available in a district hospital, and the analysis still needs doing.

The failure isn't Excel. It's Excel as production infrastructure — a spreadsheet emailed around, versioned by filename (final_v3_FINAL_updated.xlsx), with no audit trail, no access control, and no one certain which copy is real. That artefact holding patient data is a governance problem wearing a familiar icon.

The rules that actually help

• Export as text, import deliberately. Never double-click a clinical CSV. Use Data → Import and set every ID column to Text before it loads. Once Excel has guessed, the original is gone.
• Treat identifiers as text always. They are labels, not numbers. Nobody ever needs the average of an ABHA number.
• Check row counts at both ends. Exported 8,432; opened 8,432? That single habit would have caught Test and Trace.
• Keep the export query, not just the spreadsheet. Provenance is the difference between a number and a claim.
• Don't email PHI in a spreadsheet. Obvious, universal, and it happens daily.

For the INC Unit I student

The curriculum asks you to use computer applications for patient care. The genuinely useful version of that skill is not knowing where the toolbar buttons are — it's knowing what happens to a patient identifier when it crosses from the hospital system into a spreadsheet, and being the person in the room who checks.