Collect and analyse a survey

One survey question analysed: response rate, missing values and distribution, in a single widget. Click to enlarge.

Summary

An eCRF export — electronic case report form — imports as it is: Linkr recognises the Goupile or REDCap format, joins the form sheets back together and picks up the labels from the dictionary. The Survey question plugin then analyses each question with its response rate, which is the first thing to check on any collection.

An eCRF is the data collection tool of clinical studies: a structured form, filled in centre by centre. What comes out of it is rarely a clean table — which is usually where the work starts.

Importing an eCRF export

Importing a file opens a dialog that detects the format and offers the matching preset. Here, Goupile (eCRF) — an open-source eCRF developed in France — was recognised from the file.

Importing ecrf_export.xlsx: 240 rows, 49 columns. The Goupile preset is detected automatically, with a preview of the first rows below.

That preset does three things you would otherwise have to code: it joins every form sheet on the record identifier into a single table, it labels the columns from the form's dictionary, and it handles missing value tokens (na, n/a, null…) so a numeric column containing them stays numeric.

Without that last point, a single cell containing "n/a" is enough to turn a whole column of ages into text — and to make any calculation on it impossible.

The resulting dataset holds 240 records and 49 columns: investigating centre, screening date, eligibility, allocation arm, and the questionnaire answers.

Analysing the answers

Like any dataset, this one takes analyses. The plugin library is the same as for dashboards — but one of them is built for this exact case.

Choosing the analysis plugin. Survey question is designed for eCRF exports (Goupile, REDCap).

The Survey question plugin takes one question and gives the reading a collection calls for: the response rate, the number of missing values, and the distribution of answers — histogram and median for a numeric question, breakdown by category for a multiple-choice one.

In the screenshot at the top of this page, the age question shows a 81.3% response rate: 195 answered, 45 missing, with a median of 52. That is exactly what you want to know first about a collection — before analysing the answers, knowing whether they are there.

The other plugins remain available on this data: a descriptive table to compare allocation arms, a statistical test, or a map if the collection carries coordinates.

From collection to dashboard

An analysis attached to the dataset answers a one-off question. To follow a collection while it runs — inclusions per centre, completeness per form — those same widgets go on a dashboard, which recomputes with every new export.

Going further

Build a dashboard — follow a collection continuously rather than at a single point in time.
Manual data collection — why an eCRF rather than a spreadsheet, and how to structure your form.
Data quality — what the response rate tells you, and the other checks worth running.