How AI is reshaping pharmaceutical production processes

On a filling line at Medivant Healthcare, 20,000 single-dose vials for injection are produced every day. These vials end up at patients' bedsides, in hospital crash carts, and in emergency rooms, often for patients who need anesthesia.

The U.S. Food and Drug Administration (FDA) requires that every vial of medication undergo visual inspection before leaving the factory. Currently, Arizona-based Medivant relies on human workers to confirm that no visible particles or label defects exist in the vials—issues that could render the medication unusable.

But soon, the company's vial inspection equipment will take over this process. "People get tired and bored," said Viraj Gandhi, CEO of Medivant, which can cause them to miss certain defects, while machines will not.

As part of the equipment's AI training process, Medivant sets the appearance standards for prohibited particles for algorithm training. Gandhi added that AI is permeating many aspects of the drug manufacturing process, and the FDA is closely monitoring this. Earlier this year, the agency released a discussion paper titled "Artificial Intelligence in Drug Manufacturing," seeking input from stakeholders and the public on the best ways to apply the technology in this field.

Thomas Hartman, President and CEO of the International Society for Pharmaceutical Engineering (ISPE), noted that integrating AI into drug manufacturing is difficult because the industry is highly regulated. Any process change—including the direct or indirect introduction of AI—can affect product quality, requiring manufacturers to undergo comprehensive change control procedures, which are both time-consuming and resource-intensive.

How AI is currently used in drug manufacturing

Beyond quality control, AI can also help manufacturers understand how to optimize production processes. Sue Marchant, Product Executive Vice President at MasterControl, said AI algorithms can help companies understand how trends and facility temperatures affect the manufacturing process, including impacts on product quality or yield. MasterControl sells quality management and manufacturing execution software.

According to Marchant, some manufacturers use AI to focus on a single manufacturing step before commercialization. During the R&D phase, they change one step in the manufacturing process at a time to determine whether it improves quality, outcomes, or efficiency.

AI can also be used to enhance a company's visibility into operations, such as unplanned deviations at a manufacturing site. It can also provide contextual data, such as sharing when and why similar situations previously occurred at the facility.

"We are providing information to (employees) to potentially guide their actions and assess risk," Marchant said, "but we are not yet telling them in a prescriptive way, 'We think you should take this action,' so we are introducing AI into these processes in a very cautious manner."

Marchant noted that the technology can also be integrated into the quality side, making it easier for manufacturers to comply with regulations and ensure their operations are safe and reliable by reviewing data and flagging high-risk areas.

Challenges of implementing AI in drug manufacturing

Gandhi said the biggest obstacle to implementing AI in drug manufacturing is cost. Medivant spent millions of dollars on its semi-automated vial inspection equipment, which he said is 10 times more expensive than the labor cost for the same work.

"Everyone wants to do it, but you need sufficient capital," Gandhi said.

Medivant purchased the equipment four months ago and is currently training and validating it with sample vials over a four-to-five-month period, after which it will be tested with commercial batches. Once in production, the company will also conduct six to nine months of human observation to ensure there are no false negatives in the inspection. After that, the equipment will become the manufacturer's primary inspection method.

For small manufacturers like Medivant, which produces low-cost generic drugs such as epinephrine and lidocaine, investing in automation and AI is challenging. Gandhi said the company's products mostly sell for a few dollars per vial.

However, AI tools have seen broader adoption among larger manufacturers. "Big companies typically take on the risk of new equipment to see how it works," Gandhi added.

Another limitation to adopting AI is whether companies have the right digital data to train algorithms and run analyses. Marchant said this could restrict AI options to larger enterprises, as many smaller manufacturers still use paper logs or rely primarily on local data storage rather than cloud services.

Introducing AI may require "the entire manufacturing process to be digitized, where you have outlined every step in the manufacturing process, you are capturing each step digitally, recording the date and time it occurred, who was involved, and the parameters of each step," Marchant said. Manufacturers may also need to merge and accurately integrate data from different sources.

The role of the FDA

An FDA spokesperson told Manufacturing Dive that the agency "has recognized and embraced the potential of advanced manufacturing to bring benefits to patients and consumers."

Its Center for Drug Evaluation and Research (CDER) established the Emerging Technology Program in 2014 to work with stakeholders to support advanced manufacturing, including technologies like AI. According to the spokesperson, CDER recognizes that regulatory policies and programs may need to evolve to enable the timely adoption of new technologies.

"A lot of the action we see from (the FDA) is them trying to get their arms around this technology," Marchant said. "There is not a lot of strong guidance yet on what companies should do."

She said the result is that companies may be hesitant to use AI because they are unsure what it means for the future.

"A lot of the action we see from (the FDA) is them trying to get their arms around this technology. There is not a lot of strong guidance yet on what companies should do."

Marchant said her clients are leaning conservative, asking in requests for proposals how AI is used in MasterControl software. They worry that if they adopt AI tools now, future FDA requirements could interfere with their processes.

"Everyone is in a wait-and-see mode," Marchant noted. "The FDA definitely wants to encourage innovation, because their message is that they want to embrace innovation, but what does that actually mean? No one really knows yet."

The FDA is interested in incorporating AI into drug manufacturing and highlighted to Manufacturing Dive several ways the technology could improve processes:

• Identifying optimal process design and scale-up strategies, potentially reducing development time and waste
• Better controlling product and process quality while reducing human involvement and human error
• Better monitoring of manufacturing processes and detecting performance changes, thereby triggering preventive maintenance activities and reducing process downtime
• Preventing more process deviations and better identifying root causes when deviations occur

In addition to seeking input from stakeholders and the public, the FDA also co-hosted a workshop in September with the Product Quality Research Institute (PQRI) on a regulatory framework for AI use in drug manufacturing.

The spokesperson said CDER's goal is to help patients realize the benefits of using AI in drug manufacturing while minimizing risks, such as those arising from AI being used for unethical purposes.

Gandhi said the biggest concern is that data could be tampered with. In terms of inspection, the FDA already has strict regulations to verify the accuracy and safety of equipment and inspection results.