Catching Data Quality Problems at the Point of Entry | AI Learning Guide

These checks are not about the science of the trial. They are about what a valid record looks like. A heart rate of nine hundred is not a subtle finding — it is impossible, and a range rule catches it instantly. A visit dated before the participant enrolled contradicts the record itself. A duplicate entry would quietly double-count a measurement. A blank required field leaves a gap. The reason catching these at entry matters is not that they are hard to fix; it is that fixing them late is expensive, because the site has to reconstruct what happened months after the visit.

0:00 / 0:00

Auto-play

Common data quality problems caught automatically

A value outside the physically possible range, such as an impossible heart rate or body temperature
A date that contradicts the record, such as a visit dated before the participant enrolled
The same measurement submitted twice, which would double-count it in the analysis
A required field left empty, which would leave a gap the analysis cannot use

The reason timing matters here is not that these errors are hard to fix — most are trivial once spotted. It is that spotting them late is costly. A query sent to a site six months after a visit often arrives when the staff who conducted it have moved on, the source documents are archived, and the answer requires reconstructing what happened. Catching the same error at entry turns a reconstruction into a correction.