Why The Biostatistician's Seat At The Table Keeps Vanishing

By John Oncea, Chief Editor, Clinical Tech Leader

Natasa Rajicic, ScD, has a line she used to put in presentations, one she still thinks about when a client brings her in too late. “If you call a statistician after the experiment is done, it’s too late. The patient’s already dead; you’re not going to save them.”

It’s a blunt way of making a point that, in her 25 years across Pfizer, Cytel, and independent consulting, she’s found herself repeating in gentler language to almost every client she’s worked with: biostatistics isn’t something you apply to a trial after it’s over. It’s something that has to shape the trial before it starts.

What surprised me in our conversation wasn’t that she believed this. It’s how specific she got about why the alternative is so common, and what’s actually being lost when it happens.

What the Statistician Does Before Anyone Enrolls

“I wouldn’t even engage with clients if they haven’t had a biostatistician before the study start,” Rajicic told me. “That’s just impossible.” In her framing, a trial’s statistical analysis plan needs to be written and largely locked in before the first patient enrolls, which means a biostatistician has to be involved in decisions that most people outside the field assume are purely clinical or operational.

She offered a concrete example. Say a trial is measuring drug effect after ten weeks of treatment, but some patients won’t complete all ten weeks. Do you keep collecting data after they stop treatment? Do you stop collecting when they do? “That influences what you can say at the end of the study, what your targeted treatment is,” she said. The decision touches clinical operations, shapes what claims the study can support, and determines how the data gets analyzed. “You can’t do that without a biostatistician,” she said. “Biostatistician is the only one who would have the appropriate expertise to understand the intricacies between this study's clinical question being asked, how the data is collected, and the analysis that needs to be done.”

That’s a different job description than the one most non-statisticians carry around in their heads. As Rajicic put it, people tend to think of statisticians as the ones who “analyze the data at the end,” missing the chance to shape a better trial by bringing that expertise in before it begins.

Why the Misconception Persists, and Why AI Makes It Easier to Miss

Part of the reason that misconception has staying power, in Rajicic’s view, is that the mechanical side of a statistician’s job has genuinely gotten faster. “In today’s world, you can write up all the programs before study start, automated, to almost just have a button clicked and studies analyzed, tables produced,” she said. Technology, including newer AI-assisted tools, has meaningfully reduced the time statisticians spend on what she called “minutia and mundane stuff,” something she said had bothered her since early in her career. “Thank God it’s finally reducing,” she told me. “That thing improved a lot, which is nice.”

But the same automation that frees up a statistician’s time also makes it easier for organizations to mistake the analysis-and-reporting output for the entirety of the job, when the harder, more valuable work – the design decisions that determine what those outputs can and can’t say – still requires a person in the room early. Faster reporting doesn’t reduce the need for design expertise. If anything, it raises the stakes on getting the design right, since a flawed design now gets analyzed and reported faster too.

The Structural Reason: Who’s Actually on Contract

The second half of the problem is economic and structural, not just a matter of awareness. Rajicic pointed to how smaller companies typically staff biometrics: through a CRO, rather than with an in-house statistician collaborating closely with the clinical team. “There’s a once-removed statistician somewhere into CRO,” she said, “and the relationships can then be different because it’s like on a transaction basis: contracts are signed, and you include a biostatistician for this many hours and for these tasks only.” The result, in her words, is that “there’s no free flow of conversations and intellectual conversation.”

She contrasted that with her years at Pfizer, where the company employed well over 200 statisticians. “At that time that was not an issue because the processes were such that by default you have to be included,” she said. That default has eroded industry-wide as sponsors, particularly smaller and mid-sized ones, have shifted toward outsourcing biometrics functions rather than building them in-house. It’s a trend she describes with some ambivalence rather than nostalgia: “A lot of things are being outsourced now in this industry, although that goes through difficulties sometimes. The trend is to outsource, sometimes to bring it in-house.”

The consequence she sees most often isn’t a single catastrophic failure. It’s smaller erosions: a manuscript describing primary study results drafted without a biostatistician’s input, or a regulatory interaction where statistical framing gets decided without the person who understands the analysis plan’s implications. “That I see a lot,” she said, “and that can be kind of frustrating.”

A Consultant’s View From Outside the Org Chart

Rajicic’s own career path adds an interesting wrinkle to this story. As an independent consultant since 2021, she’s essentially the outside expert that a transactional CRO relationship is supposed to replace, brought in specifically because a sponsor needs the kind of deep, design-stage collaboration that an hours-capped contract often can’t provide. She was careful not to overstate her own vantage point. “I’ve been working as a consultant for the last five years,” she told me, “so my perspective, I’m kind of over time losing perspective of somebody who’s within a company and actually dealing with the corporate stuff, which is on purpose, thank God.” Even so, the pattern she described, being called in precisely because an internal team lacked the statistical partnership it needed at the design stage, is itself evidence of the gap her larger point describes.

What This Means for Clinical Technology Leaders

For a technology leader who doesn’t control the org chart but does influence how teams are structured around a given program, Rajicic’s experience suggests a few practical checks. Confirm a biostatistician is involved before the statistical analysis plan and protocol are finalized, not after. If biometrics are outsourced to a CRO, evaluate whether the contract structure supports genuine back-and-forth collaboration or only a fixed set of deliverable hours. And don’t mistake faster, more automated statistical reporting for a reduced need for statistical judgment earlier in the process. The tools have changed. The timing problem Rajicic has been flagging for decades hasn’t.