Clinical data is the most valuable asset in drug development. Preparing it is still one of the most manual stages of the entire process.

Before trial data can feed a submission, it has to be converted into the standardised format that regulators review. The work is skilled but repetitive, and anyone in clinical data management or statistical programming could describe it in detail. Few expect it to change. It has been filed, quietly, under the cost of doing business.

That trade-off is no longer necessary. AI, applied with experts firmly in control, can now take on the heaviest part of the work without loosening the rigour regulators expect. The clearest gains are at the very start of the pipeline, where trustworthy evidence begins.

Data arrives from electronic data capture systems, laboratories, devices and research partners, each in its own format. The target is the Study Data Tabulation Model, or SDTM, the industry standard that gives trial data a common structure for regulatory review.

The conversion is painstaking. Someone has to work out, variable by variable, where each piece of source data belongs, then write and check the code that moves it there. On a single study, that source-to-target mapping can take weeks of skilled effort. When a protocol amendment lands or a late data transfer arrives, the rework begins and timelines can start to slip.

It also resists scale, at a time when each trial draws on more data from more sources. More studies have simply meant more people doing more mapping. And because the work is so manual, it drifts: the same lab test can be mapped differently by different teams or vendors, and hard-won knowledge rarely carries over to the next study.

This is where AI changes the economics, and where many experienced practitioners grow wary. Regulated data is no place for tools nobody can inspect, and that scepticism is healthy.

The answer is to keep experts in charge. In a human-in-the-loop model, AI proposes the mappings and generates the transformation code, and specialists review and approve it, adjusting where needed, before anything is used. The AI does the repetitive groundwork but people make the decisions, and every decision can be seen and checked.

Nobody is arguing for fewer statistical programmers. The point is to move their time from repetitive mapping to the complex derivations and judgement calls that genuinely need them.

The starting line

Much of the conversation about clinical data integrity focuses on the finishing line: proving how results were produced and getting submissions right.

Both depend on what happens here first. SDTM feeds the analysis data sets, tables and figures that eventually reach a regulator, so a weakness at the start travels all the way through, and so does a delay.

Speeding up one step only helps if the chain stays connected. Inside a validated, governed environment, versioning and auditability are part of the work from the outset. SAS Clinical Data Transformation, where AI recommends mappings and generates code for experts to review, is one example of where the industry is heading.

The bigger shift is in how clinical data is seen. For too long it has been treated as plumbing. Its quality and speed shape how quickly a programme learns and how confidently it can go to regulators, which makes transforming it a strategic decision, not a matter for the back office.

Clinical data begins with the patients and volunteers who give their time and trust to a trial. It is patients, too, who gain most when that data moves faster without losing its integrity. That is what gets therapies to them sooner.