By Xiaoling Wu, Statistical Science Director, AstraZeneca
Not long ago, a PhD student asked me whether artificial intelligence would eventually replace statisticians. The question is increasingly common as tools such as ChatGPT, Gemini, and Claude demonstrate an impressive ability to generate code, summarize research papers, and even suggest statistical models.
For statisticians working in applied fields, particularly in pharmaceutical development, the question is both practical and immediate. Many of us have already experimented with AI tools to assist with programming, literature review, or documentation. In these contexts, the technology can be surprisingly useful. Routine tasks that once required hours of manual effort can often be completed much more quickly with AI-assisted workflows.
Yet after experimenting with these tools in real projects, I have come to a different conclusion than the one implied by the original question. AI may significantly improve productivity in statistical work, but it does not replace the central role of statistical reasoning in scientific decision-making. In pharmaceutical development—where statistical conclusions directly influence clinical trials, regulatory interactions, and ultimately patient outcomes—the work of statisticians involves forms of judgment that remain difficult to automate.
Several aspects of pharmaceutical statistical practice illustrate where the limits currently lie.
Where AI Still Falls Short in Pharmaceutical Statistics
Translating Clinical Insight into Empirical Rules
One important task in clinical trial analysis is identifying patient characteristics that influence treatment response. In oncology studies, statisticians often explore whether prognostic factors or post-treatment biomarkers can help predict which patients are most likely to benefit from a therapy.
Developing such empirical rules is rarely a purely algorithmic exercise. It typically requires integrating information from multiple sources: biological knowledge about disease mechanisms; clinical observations about response patterns; and statistical evidence from exploratory analyses across trials. These insights often emerge gradually through collaboration among statisticians, clinicians, and scientists rather than from a single dataset.
AI tools can assist with summarizing literature or generating analysis code, but they generally struggle to synthesize these diverse forms of knowledge into meaningful hypotheses. The reasoning that links patient characteristics, disease biology, and treatment effect remains fundamentally dependent on human expertise.
Understanding When Statistical Methods Do—and Do Not—Transfer
Another challenge arises when statistical approaches developed in one therapeutic context are applied to another.
In oncology, for example, many statistical methods were developed in the context of immune checkpoint inhibitors such as PD-1 therapies, where treatment effects may emerge gradually and survival curves can cross. These approaches may work well in that setting but may not apply directly to newer treatment modalities such as ADC (antibody–drug conjugates) or cell therapies, where mechanisms of action and response dynamics differ substantially.
Determining whether a method developed for one class of therapies can be applied to another requires understanding disease biology, treatment mechanisms, endpoint behavior, and the historical performance of analytical strategies in similar trials. These judgments rely heavily on domain expertise and accumulated experience, rather than pattern recognition from published data alone.
Decision-Making Based on Proprietary Institutional Knowledge
Statistical work in pharmaceutical development also relies heavily on internal organizational knowledge.
When responding to health authority queries or evaluating development strategies, statisticians frequently draw on internal sources such as prior clinical study reports, unpublished analyses, and institutional experience with specific endpoints or patient populations. These materials often contain insights that are not publicly available but are highly relevant to future trial design and analysis.
Most current AI systems operate primarily on publicly accessible information. As a result, their recommendations may be based on published literature while overlooking important context derived from a company’s internal development history.
Data Privacy and Regulatory Constraints
Finally, the use of AI in pharmaceutical statistics is shaped by strict privacy and regulatory requirements.
Clinical trial datasets contain sensitive patient-level information and are subject to rigorous confidentiality protections. For this reason, many pharmaceutical organizations impose strict restrictions on how AI tools can interact with patient data. In some environments, AI systems may assist with documentation or programming but are not permitted to directly analyze patient-level clinical datasets.
This constraint makes the application of AI in pharmaceutical statistics quite different from its use in industries such as technology or marketing analytics, where large behavioral datasets may be more freely used to train and deploy machine learning models.
How the Role of Statisticians May Evolve
If AI is unlikely to replace statisticians in pharmaceutical development, it will almost certainly reshape how the profession is practiced.
Many routine tasks—such as generating code, drafting reports, or summarizing literature—may increasingly be supported by AI tools. As these capabilities expand, statisticians may spend less time on mechanical aspects of analysis and more time on the scientific reasoning that underlies statistical decisions.
Paradoxically, the rise of AI may therefore make the core strengths of statisticians even more important. As automated tools produce analyses more quickly, someone must evaluate whether those analyses are scientifically meaningful, whether model assumptions are appropriate, and whether the resulting conclusions support sound decision-making. In drug development, where statistical conclusions may influence regulatory approval and clinical practice, these responsibilities cannot be delegated entirely to automated systems.
In this evolving environment, statisticians may increasingly serve as interpreters of evidence and scientific advisers, helping multidisciplinary teams navigate uncertainty and choose analytical strategies that align with clinical and regulatory realities.
Advice for Early-Career Statisticians
For students and early-career statisticians, the rapid development of AI can create uncertainty about the future of the profession. Yet the skills most likely to remain valuable are those that have long defined strong statistical practice.
First, master the fundamentals of statistics. A deep understanding of study design, model assumptions, bias, and uncertainty is essential for evaluating analyses—whether those analyses are generated by humans or machines.
Second, develop domain expertise. In pharmaceutical research, understanding disease biology, treatment mechanisms, and clinical trial objectives is often what determines whether a statistical method is appropriate. These contextual judgments are difficult for generalized AI systems to replicate.
Finally, cultivate the ability to communicate statistical reasoning clearly. Many critical decisions in drug development involve explaining uncertainty, comparing competing analytical approaches, and helping nonstatistical collaborators interpret complex data. The ability to translate statistical insight into practical decisions will remain a defining strength of the profession.
Looking Forward
Artificial intelligence will undoubtedly continue to transform many aspects of scientific work. In statistics, it will likely automate parts of the analytical workflow and expand the tools available to practitioners.
However, in pharmaceutical development, statistical practice extends far beyond model selection or computation. It involves integrating scientific knowledge, regulatory considerations, and institutional experience to interpret uncertain evidence and guide consequential decisions.
For this reason, the rise of AI is unlikely to diminish the importance of statisticians. Instead, it may shift the focus of the profession toward its most essential function: applying statistical reasoning to complex real-world problems where data alone cannot provide all the answers.
Leave a Reply