Novo Nordisk accelerates critical pharmaceutical reports with MongoDB Atlas

THEIR CHALLENGE

Using MongoDB Atlas to accelerate time to market

Denmark-based Novo Nordisk is one of the world’s largest pharmaceuticals operators, and one of Europe’s most valuable companies. It’s also expanding at an unprecedented rate, so to achieve its target of 1,000% growth by 2030, it needs to scale, accelerate, and digitize its drug-development processes.

Having a drug approved for market access is a process that requires, among other necessities, creating a common technical document (CTD) detailing the results, efficacy, and safety of all the relevant clinical trials. One of the CTD’s most crucial parts is the clinical study report (CSR), which summarizes the trial results in a document that can be up to 300 pages long.

Producing a CSR is a long cycle of writing, reviews, rewrites, and approvals that can take the company up to 19 weeks. As a largely manual process, it’s also prone to errors. Therefore, Novo Nordisk looked to streamline production with MongoDB Atlas: By breaking text down into common modular phrases and “snippets” that can then be combined with trial-specific data and copy, the company could significantly accelerate the production an initial drafts.

OUR SOLUTION

Producing accurate copy with a streamlined process

Novo Nordisk’s solution is NovoScribe, a high-power tool that relies on large language models (LLMs), vectors, and retrieval-augmented generation (RAG), all enabled by MongoDB Atlas. The context-based tool uses a process of prompt, search, and retrieval to develop accurate copy based on previously approved text combined with case-specific variables.

“LLMs are probabilistic models, but pharmaceuticals [are] very tightly regulated, so we can’t just assume that the model does the right thing,” explained Tobias Kröpelin, Structured Authoring Tech Lead and Statistical Programming Specialist at Novo Nordisk. “We need deterministic and clear outcomes, so we introduced an intermediate step where a human reviews five possible pieces of text and chooses the best one, which then goes into our database.”

By fusing the approved text with the correct data—such as tables, lists, and statistical and analytical results—NovoScribe can then, section by section, produce a complete and accurate document. In essence, structured authoring enables the team to use and reuse text, almost reaching the point of producing documents automatically.

OUTCOME

Delivering fast, valuable, and regulator-friendly results

Thanks to the implementation, Novo Nordisk has seen a dramatic reduction in the time needed to produce a CSR.

“MongoDB has helped us cut writing times by 90%,” said Kröpelin. “The savings could actually be much higher than that, but we still need human stakeholders to review each one. Technically, we could produce a CSR in just a few minutes, but we can’t have a machine owning the document.”

Previously, one staff writer would produce around 2.3 CSRs in a full year. Now, Novo Nordisk can manage its writing workload with just three full-time staff instead of 42—and with significantly reduced error rates. Replacing manual processes and subjective writing with NovoScribe is fast, accurate, and—most importantly—gaining very positive feedback from regulators. What’s more, the company is branching its use of NovoScribe to other areas, with the goal of ultimately automating the creation of entire CTDs.

“MongoDB generates important value that [enables] our people to focus on the science,” said Kröpelin. “There are plenty of other parts of the process that still need human thought and scientific input, so it’s great to be able to pass the repetitive tasks on to computers.”

To learn more, visit MongoDB Atlas.