Building a Culture Where the Honest Answer Is Easiest

Part of Clinical Data Management and Integrity: How Data Integrity Failures Are Found.

About 5 minutes. Free, in your language, no account needed.

Why this matters

Every failure in the previous lesson was committed by someone doing what the system made easiest. That is the whole of the practical conclusion.

An organization can train people extensively on data integrity and still produce these failures, if the daily experience of doing the honest thing is harder, slower or more likely to attract criticism than the alternative. People respond to the situation in front of them far more than to the training they had last year.

So the design question is not how to make people care. It is how to make the honest action the path of least resistance.

The asymmetries to remove

Reporting versus not reporting. If recording an adverse event triggers a long cascade of required fields and recording nothing requires no action, the system has made under-reporting easier. Measure the effort of each path and equalize it.

Explaining versus changing. If closing a query by changing a value takes one click and explaining why it is correct requires a free-text justification and a supervisor's review, values will change. Make the explanation as easy as the change.

Admitting versus concealing. If reporting a deviation produces scrutiny, a corrective action plan and a difficult conversation, while not reporting it produces nothing, deviations go unreported. This is the hardest asymmetry to remove because the scrutiny is legitimate, and the answer is to make the response to a self-reported problem visibly different from the response to a discovered one.

Waiting versus guessing. If a site cannot save a record without a value it does not have, it will guess. Every mandatory field needs an honest escape.

What to measure, and what not to

Measure the things whose improvement means the system is working: query yield, the proportion of discrepancies escalated rather than closed, self-reported deviation rates, and the time between an issue occurring and being identified.

Note that two of those go up when things improve. A rising self-reported deviation rate usually means reporting is improving rather than conduct worsening, and an organization that cannot tolerate that number rising will get a number that does not rise.

Do not measure, as targets: open query counts, deviation counts, compliance percentages. Each is a number people can move without improving anything, and each has an easier route through the data than through the work.

The response to a self-reported problem

This is the single highest-leverage thing an organization controls.

When someone reports that they made a mistake, entered something wrong, or discovered a problem in their own work, what happens next determines whether the next person reports.

A response that treats the report as the system working, focuses on the cause rather than the person, and visibly differs from the response to a discovered problem, produces more reports. A response that is indistinguishable from the one a discovered problem gets produces fewer, and the problems do not stop happening.

What data management specifically can do

Design forms that cannot force a false answer. Every required field gets an honest escape.

Write queries as questions, with "the value is correct and here is why" as an acceptable answer.

Treat an unresolvable discrepancy as a legitimate outcome, documented rather than closed.

Review the audit trail routinely and act on what it shows, so that the trail is a working tool rather than evidence produced at inspection.

Report patterns rather than counts, so that a site that differs is identified as differing rather than as failing.

In practice

A data manager introduces one change: when a query is answered with an explanation rather than a data change, the explanation is recorded and the query closes as verified rather than as resolved.

Previously both outcomes recorded the same status, which meant a site's explanation and a site's correction were indistinguishable in every report, and the reporting emphasized closure.

After the change, the study can see that 34 percent of queries close as verified. That number is now visible, and it is a quality signal in both directions: too low suggests sites are changing values rather than explaining them, and very high suggests the checks are finding nothing.

The change cost a configuration setting. What it altered was that explaining became a visible, recorded, legitimate outcome rather than an unrecorded non-event.

What this looks like when it goes wrong

An organization runs annual data integrity training for every study team member, with a test and a certificate. Completion is 100 percent.

Its electronic data capture systems have mandatory fields with no unknown option, its query metrics are open counts reported weekly to management, its deviation numbers are compared across sites in a performance review, and reporting a mistake goes through the same process as a discovered finding.

The training is accurate and well delivered. Everyone who takes it agrees with it.

They then return to a system in which every incentive runs the other way, and the behavior that follows is the behavior the system produces, not the behavior the training described.

The training was never the lever. It was the thing that let the organization believe it had addressed the problem.

Key takeaways