EHR data entry, and the cost of a wrong field

A weight entered in pounds into a field expecting kilograms does not look like an incident. It looks like a number. It becomes an incident later, when a dose is calculated from it, and by then the person who typed it has seen forty other patients and the record carries no trace of the moment it went wrong.

We worked with a healthcare provider on this exact class of problem: frequent data entry errors and slow record retrieval across their electronic health record system. The brief arrived labelled as data quality. Data quality is a report that comes out wrong and gets corrected next month. This is a clinician making a decision from a value that is not true.

Where the errors actually come from

Almost none of them are carelessness. They come from the shape of the capture path. Free-text fields where a coded value already exists. Units left implicit because the form was designed for one department and then reused by another. Defaults that are correct nine times out of ten, which means the tenth gets confirmed without being read. Copy-forward, which carries yesterday's assessment into today's note along with anything wrong in it. And the worst of them: a form slow enough that staff batch their entry to the end of a shift and reconstruct it from memory.

Retrieval speed is a safety control

Record retrieval time gets treated as a comfort metric, something to put in a satisfaction survey. A clinician who needs a value in the next thirty seconds and cannot have it in the next thirty seconds does not wait. They ask the patient, or they work from what they remember, or they order the test again. Slow retrieval does not produce a queue. It produces a workaround, and the workaround leaves no audit trail at all.

The work behind that figure was unglamorous. Coded values replacing free text wherever a code existed. Automation of the routine entry that had been retyped by hand from another system. Validation at the point of capture rather than in a nightly report. And an honest review of which fields were mandatory because they were clinically necessary and which were mandatory because somebody once wanted a metric.

Validation that argues with the clinician loses

Hard blocks are tempting and they fail in a predictable way. Refuse an unusual value and staff will enter an acceptable one to get past the screen, then put the truth in a free-text note where nothing can read it. Physiology produces genuine outliers every day. A rule that treats every outlier as an error is teaching people to lie to the system, and they learn quickly.

Soft challenge works better. Flag the value, show the normal range and the patient's own history beside it, allow confirmation with a reason, and record that the confirmation happened. The record then holds both the unusual value and the evidence that a qualified person looked at it, which is what an auditor and the next clinician each need for different purposes.

Auditing means reading records, not counting them

Error rates computed by the system will only ever surface the errors the system can detect. The dangerous ones are internally consistent and wrong. Sampling real records against source documents, on a schedule, with clinicians in the room, is the only method we have found that catches those. It is slow, somebody has to be paid to do it, and it is the part clients try hardest to cut.

The framing matters more than any individual fix. Filed as data quality, a wrong field gets a severity rating and a position in a backlog. Filed as patient safety, it gets the scrutiny given to any other route to harm, which is what it deserves, given that the harm turns up weeks later and almost nobody traces it back to a form.