Generative Artificial Intelligence as Modern-Day Electricity – One Step at a Time
Hong Tian and Tony Guo (BeOne Medicines USA, Inc., formerly BeiGene USA, Inc.)
Highlights
Discover how AI tools like Open AI GPT are reshaping the role of statisticians in pharma and biotech - boosting efficiency, accelerating learning, and sparking innovation.
See real-world examples of AI assistants designed as medical and regulatory experts, providing domain-specific support.
Explore how prompt engineering and generative AI can accelerate knowledge acquisition, enhance productivity, and uncover deeper insights
In pharma and biotech, statisticians play a pivotal role in shaping data generation strategies to answer scientific questions around safety, efficacy, and quality. Their work spans study design, data collection, analysis, and interpretation—requiring both strategic vision and operational excellence.
Today, AI tools such as Open AI GPT can assist with nearly every aspect of that work. With minimal effort, statisticians can create custom AI assistants to enhance efficiency, accelerate learning, and spark new ideas. We aim to share practical examples and inspire broader applications.
“The best thing about being a statistician is that you get to play in everyone’s backyard.” — John Tukey
To make a meaningful impact, statisticians must understand the evolving context of their work: disease biology, treatment algorithms, risk stratification, competitive landscapes and regulatory policy. These “backyards” are data-rich—and constantly changing.
Chat GPTs can dramatically speed up knowledge acquisition and elevate both productivity and creativity. For statisticians already fluent in programming, learning prompt engineering is a natural and rewarding next step. White et al. (2023) outlines many helpful prompt patterns.
It takes surprisingly little effort to bring expert-level AI support to your fingertips. In this article, we share a few simple use cases—hoping to inspire many more.
The first two examples demonstrate how Chat GPTs can be configured to act as a medical assistant and a regulatory assistant, delivering domain-specific, reliable support. The following three examples highlight how to enhance efficiency and quality through conversational prompting—drawing clear parallels to R programming constructs to make prompt engineering more intuitive for statisticians.
Example 1: Medical Expert Specialized in Lymphoma
# Persona
You are a digital medical expert specializing in lymphoma. You integrate current clinical practice, trial data, and regulatory guidelines to produce the most reliable and explainable answers.
# Output Expectations
- Respond concisely and clearly.
- Justify conclusions using trial data or official guidance.
- Include links to references for every key claim (e.g., PubMed, FDA, NCCN, ASCO).
- State when no definitive evidence exists.
# Upload files
Include latest disease and treatment guidelines.
Example 2: Global regulatory strategist
# Persona
You are a global regulatory strategist with deep expertise in oncology drug and device development. You are highly knowledgeable about regulations, processes, and precedents across major health authorities: FDA (U.S.), EMA (Europe), PMDA (Japan), and Health Canada. You synthesize current guidance documents, historical approvals, and jurisdiction-specific practices to deliver reliable, evidence-based regulatory insights.
# Output Expectations
- Provide **clear**, **concise**, and **well-reasoned** responses.
- Ground all conclusions in **official regulatory guidance**, **publicly available review documents**, or **historical approval records**.
- **Cite and link to primary sources**, prioritizing the following approval databases:
- **PMDA (Japan)**
- [Approved Drugs (English)] (https://www.pmda.go.jp/english/review-services/reviews/approved-information/drugs/0001.html)
- [Japanese Drug Search (in Japanese)] (https://www.pmda.go.jp/PmdaSearch/iyakuSearch/)
- **Health Canada**
- [Clinical Information Portal] (https://clinical-information.canada.ca/search/ci-rc)
- **FDA (U.S.)**
- [Drugs@FDA Database] (https://www.accessdata.fda.gov/scripts/cder/daf/index.cfm)
- **EMA (Europe)**
- [EMA Medicines Search] (https://www.ema.europa.eu/en/search?f%5B0%5D=ema_medicine_bundle%3Aema_medicine&f%5B1%5D=ema_search_categories%3A83)
- Use **regulatory terms and timelines correctly**, and specify **jurisdictional differences** when applicable (e.g., Fast Track vs. PRIME vs. Sakigake).
- If no definitive guidance exists, **state this clearly**, and offer relevant analogs or case precedents.
Generative AI has made learning and knowledge acquisition remarkably more accessible. With genuine curiosity and a basic grasp of prompt engineering, anyone can unlock its potential.
Yet, hallucination remains a persistent challenge. It quickly becomes clear that links may be inaccurate, and responses—though confident—can shift under scrutiny. In this era, human expertise and critical thinking are more essential than ever.
GPT models offer useful coding support—and better yet, we can use a conversational style to carry out tasks that mimic programming logic.
Example 3: Repetitive Task — Iterating Through Multiple Drug Approvals
for (i in 1:n) {
# repeat task for each element
}
I will provide links to three FDA drug approvals. For each, please summarize the following in a table: medication name, indication, study objectives, primary and key secondary endpoints, and main efficacy results.
Links:1…, 2… and 3…
Example 4: Conditional Logic — Tailoring Search Based on Region
if (condition) {
# code if condition is TRUE
} else {
# code if condition is FALSE
}
Example: If I ask about drug approval in Japan, please first search:
site:https://www.pmda.go.jp/PmdaSearch/iyakuSearch/
If I ask about drug approval in countries other than Japan (e.g., U.S.), please first search:
site:https://www.accessdata.fda.gov/scripts/cder/daf/index.cfm
Example 5: Conditional Loop — Repeating a Behavior Based on Ongoing Instruction
while (condition) {
# code to run
}
Example:
From now on, please whenever I ask you a question. Please write me alternative prompts to address my question.
We encourage statisticians to explore generative AI tools—not just as users, but as builders. Using your own GPT assistant can accelerate domain understanding and unlock deeper insights, ultimately contributing to more efficient drug development.
The barrier to entry is lower than many assume. With just curiosity and a willingness to experiment, statisticians can begin integrating AI into their daily work.
By sharing our early experiences, we hope to spark more use cases, foster collaboration, and inspire others to share their journeys. In a rapidly evolving landscape, the ability to adapt and embrace new technologies is as vital as technical skill. When combined thoughtfully, human expertise and artificial intelligence can propel us further than alone.
Reference:
White, Jules, et al. "A prompt pattern catalog to enhance prompt engineering with chatgpt." arXiv preprint arXiv:2302.11382 (2023).