What a case report form is
The case report form is a study's collection grid: the list of information to be recorded, in the same order and the same shape, for every included record. In a medical thesis it usually sits in an appendix; yet it is the form, not the manuscript, that determines what the analysis will be able to say.
French-speaking thesis work calls it « fiche d'exploitation »; regulated research calls the same object a case report form, or CRF. Two names for one thing: a structured form, filled once per included subject. See the electronic case report form for the computerised version and what it changes.
What it contains
The structure varies little between specialties, because it follows the patient's path. Six sections cover nearly all single-centre work:
| Section | What goes in | Common trap |
|---|---|---|
| Identification | Anonymity code, inclusion date | Putting the name in — it has no business there |
| Demographics | Age, sex, BMI, history | Age "today" rather than at diagnosis |
| Baseline clinical | Presenting complaint, examination, validated scores | A scale cited without its name or version |
| Investigations | Laboratory, imaging, pathology | A value without its unit |
| Management | Strategy, procedure, approach, durations | Free text where three choices would do |
| Outcome | Complications, follow-up, status at last point | No follow-up date — hence no measurable follow-up |
What separates a good form from a bad one is not the number of fields. A thirty-variable form with well-defined variables beats a hundred-field form half of which will read "not stated".
The seven costliest mistakes
They are all paid for at the same moment — when the spreadsheet is opened for analysis, three months into collection, and it is too late to go back to the records.
- The open question. "Type of complication: ______". Two hundred records later there are a hundred and eighty phrasings for twelve situations. A closed list, with "other — specify" if needed, can be counted; free text has to be recoded by hand.
- The untyped variable. "BMI" without saying whether it is a number to one decimal or a class. The column ends up holding "27.4", "27,4", "27" and "overweight", and no software knows these are the same variable.
- The implicit unit. "Creatinine: ___". In mg/L or µmol/L? The ratio is 8.84. A form that omits the unit produces a column where two scales coexist unseen. The UCUM code system exists for this.
- The unqualified missing value. An empty cell may mean "not measured", "not applicable", "lost to follow-up" or "the record did not say". Those are four different situations at analysis, and a blank tells them apart not at all. One code per case is needed.
- The undeclared endpoint. Without a primary endpoint written before collection, the study will pick at the end whichever variable comes out favourable — and a reviewer will see it.
- The hand-entered derived field. BMI, a score delta, the interval between two dates: anything derivable must be derived. Entered by hand, it eventually contradicts the fields it comes from.
- No dictionary. The form says what to record; it does not say what each field means. Six months later nobody — not even its author — will know whether "recurrence" counted biochemical recurrence. See the data dictionary.
How to build one — in order
The order matters: each step narrows the next, and skipping the first makes the form swell without anyone knowing why.
- Write the question, in one sentence. "In patients operated on for BPH in the department, which preoperative factors are associated with a grade ≥ II complication?" If the question does not fit in one sentence, neither will the form.
- Derive the primary endpoint from it. It is inside the question. It is phrased with its population, measure, threshold and window.
- List the variables serving that endpoint — the factors to test, the known confounders, and nothing else. This is where you resist "let's take that too", the leading cause of hundred-field forms.
- Type every variable. Technical name, type, unit, possible values, plausible bounds. Five columns in an appendix table: that is the dictionary, and it gets filled now.
- Decide how missing values are coded, once and for all, and write it on the form.
- Test on five real records. Always. This is where you discover that "surgical approach" is meaningless for a patient who was not operated on, and that a "not applicable" box was needed.
- Freeze. From the sixth record on, the form no longer moves — or it moves, and the change is dated. A form altered mid-collection without a trace makes the records before and after incomparable.
From paper to dataset: where the path breaks
The classic thesis path has three steps, and a loss at each.
| Step | What is lost |
|---|---|
| Record → paper form | Origin: nothing says which page of the record the figure came from |
| Paper form → spreadsheet | Consistency: typos enter unresisted, units mix |
| Spreadsheet → analysis | Memory: who corrected what, when, and on what basis |
None of these losses is fatal on its own. Together they produce the situation every supervisor knows: a spreadsheet in which nobody can certify a single value, and a defence where "how did you measure that endpoint?" has no good answer.
The detailed comparison between the two approaches is treated separately: spreadsheet or platform.
In AnzarSeha
The form is not typed in: it is generated. When a study is created, the chosen condition's catalogue already supplies the usual variables — staging, work-up, validated scores, laboratory markers — each typed, with its unit and bounds. The work is removing what does not serve the question, then adding what is missing, rather than writing everything from a blank page.
Every variable carries its technical key, type, measurement scale, possible values and registry source. The primary endpoint is declared by designating one of them. The protocol locks before collection, and the database lock precedes analysis.
Frameworks and conventions
CDISC CDASH standardises the structure of collected variables and is worth knowing even outside regulated trials: it saves inventing a naming scheme per study. ICH E6(R3) places the form within a trial's documentation. For publication, STROBE and the EQUATOR network ask for a precise description of variables and their measurement — which writes itself when the dictionary already exists. UCUM codes units unambiguously.