Rescue Doesn’t Start with Bad Quality Data. It Starts with a Question Nobody Can Answer.

By Christine Kanalis, Vice President, Clinical Data Management and Aga Rasinska, Director of Strategy

September 11, 2026

A sponsor asks a simple question: where are we? It’s often asked quietly, with no panic at all.

Let’s take a look. Enrollment is holding. The weekly status email arrives on schedule. The dashboard is green. But when someone asks how many patients are clean, which queries are aging and why, and what stands between today and database lock, the answer takes three days to assemble and arrives as a spreadsheet that contradicts last week’s numbers. The level of concern starts to increase and curiosity turns to suspicion. At that moment, nobody can say where the study is.

You may think that conversations about rescue studies start with quality: errors, unlocks, biostatistics sending the same findings back for the third month running. Those are real too, and they matter, but they are usually the second thing a sponsor notices. The first is a lack of visibility.

The oversight model stopped scaling

In a scenario like the one above, it’s often the case that no one has done anything wrong. The way studies run has changed faster than the way they are overseen. A single study now draws from EDC, central and local labs, eCOA, imaging, and a handful of other external providers, each on its own transfer schedule and its own working definition of ready. Amendments land mid-flight, and programs get spread across two or three CROs for reasons that made sense at contracting and stop making sense at execution.

What has not changed is the layer the sponsor sees: a weekly status email or an Excel tracker maintained by hand. The sponsor is accountable for all of it and can see almost none of it. “Where are we?” isn’t as simple to answer as it once was, but teams keep trying the old tricks.

What the signal looks like

We have seen this pattern often enough to recognize it before the sponsor names it, and it rarely begins with a catastrophe. There is no missing dataset, no failed audit, no regulatory letter. Instead, there are reconciliation listings that were requested and never arrived. Queries aging in the EDC with no detail about why, and no evidence anyone is chasing them. The same findings come back from biostatistics after every dry run, month after month, unresolved. The same status report is reissued with a new date. Change orders arrive faster than deliverables. Taken one at a time, each of these is a challenging week in the study. Taken together, they describe a study nobody is steering. And the question that finally triggers the call is rarely is my data clean, but is anyone doing anything? Curiosity turns to suspicion; then, it becomes panic.

Rescue is a decision, not an emergency

As an industry, we have trained ourselves to treat rescue as a category of emergency — the thing you do when a study is already burning. That framing is expensive, because it means the decision only gets made once the options have narrowed. Sometimes it is genuinely urgent. Far more often, the work is turning the ship before it reaches the iceberg. This is a management decision, and it needs time, which is why the assessment matters more than the intervention itself.

When we come into a study already underway, the first question is what can be trusted. The EDC build and the programming around it are usually salvageable, because they have already been validated. Unless there is something in the specifications that negatively impacts the study — such as edit checks that never fire on a primary endpoint, an amendment that was never built into the eCRF, reconciliation that was never configured, or coding that will not map cleanly at submission — rebuilding them usually spends valuable time with little payoff.

The data is a different matter. We look at all of it, because that is how data cleaning works. Reports and edit checks run across the whole study; there is no option to clean only the part that arrived after you did. That is also why timing drives cost more than severity does. The volume of data that must be re-earned is a function of how long the study ran before anyone asked the question.

The solution is not always a transition. Sometimes the finding is one report nobody was producing, or a reconciliation process that was never properly defined. Sometimes it is structural: a sponsor that grew quickly, stood up a program across several CROs, and now has no consistent view across any of them. In that case, adding people does not fix it, but structure does.

The question the sponsor asked at the start — where are we — is the one the Clean Patient Tracker, built by Atorus, answers. It sits on a clinical data review framework built in R Shiny, with curated, analysis-ready data at the center of the work: one environment, shared definitions, and integration of EDC data and metadata to enable interactive review and querying without manual handoffs. Every review activity is tracked, and that tracking rolls into a live patient-level clean status. Study and program health is available at any moment as part of the structure that keeps a study on track, rather than something reconstructed the week before a milestone.

The question

Siloed validation output tells you what is wrong, not what is ready. Ensuring clean data does not happen on its own, but it should not be a crisis. Instead, you need visibility earlier, when you still have options. So, the question worth asking now is: if you asked where your study is today, how long would it take to get an answer you would be willing to stake a submission on?