The Beatles Revolutionized the Music Industry by Using Recording Studios — Will AI Revolutionize the Medical Writing Industry Next?

Artificial intelligence (AI) is revolutionizing industries worldwide, and medical writing is no exception. With the integration of human intelligence (HI) with AI, the field is discovering new efficiencies and maintaining, or even improving, the rigor required for quality and compliance.

In this Xtalks Spotlight interview, Maria Hopfgarten, Head of Global Medical Writing at the PPD Clinical Research Business of Thermo Fisher Scientific, shared insights into how humans and AI can effectively collaborate.

She discussed the benefits, opportunities, challenges and limitations of this collaboration, along with the roles of medical writers in an AI-assisted environment and the future of medical writing driven by advancements in natural language processing (NLP).

A Hierarchical Collaboration Between Humans and AI

When evaluating the synergy between HI and AI, Maria explained, “The way we think about it is in a hierarchical structure where humans and AI continue to interact with each other, and at the top of that hierarchy is always the human making final executive decisions.”

This structure ensures that AI’s contributions are driven by human oversight as it continues to learn from medical writers, creating a reiterative learning process.

To evaluate the potential synergies between traditional human processes and AI integration in writing methods, it is first important to understand the roles of medical writers. According to Maria, medical writers play three crucial roles in any AI-assisted environment. These include quality assurance, oversight and training AI.

“The way we think about it is in a hierarchical structure where humans and AI continue to interact with each other, and at the top of that hierarchy is always the human making final executive decisions.”

— Maria Hopfgarten

Medical writers are the ultimate guardians of quality, ensuring accuracy, style and compliance with regulatory guidelines in every written document, Maria explains.

Writers act as the ultimate “quality controllers” to verify the clinical relevance and regulatory compliance of AI outputs. They are tasked with the design of quality processes for AI integration, which includes multi-layered verification where secondary algorithms validate AI outputs.

This also involves some form of human validation checklists focused on critical accuracy factors like medical claims, clinical endpoints and patient safety information. Automated error detection tools that flag ambiguous terms or suspect data for further review are also key to verification.

Additionally, medical writers provide contextual judgment, which is vital for nuanced tasks such as interpreting complex medical data or crafting patient narratives. They serve as the final approval, “ensuring that content meets ethical standards, reflects patient safety priorities and aligns with regulatory guidelines,” says Maria.

With respect to oversight, “Processes and oversight are created by humans for humans, and we need to rethink this a little to not minimize the benefits of AI,” notes Maria.

Some strategies for effective oversight include automated review checkpoints that focus human review on aspects like accuracy, readability and regulatory compliance. Another strategy involves tiered review levels based on document complexity and type, which will determine the level of oversight and review required.

Structured feedback loops where recurring AI errors are flagged and corrected to refine the system are also important to implement. This can involve using a rating system that focuses on accuracy, readability and regulatory guidelines.

Correcting the AI during the process is essential to prevent the repetition of errors. Employing human-in-the-loop approaches allows writers to review and refine AI-generated outputs. By correcting these outputs, the improvements are fed back into the AI system, effectively reducing the likelihood of recurring errors and enhancing the model’s accuracy over time.

Training AI tools is another critical responsibility for medical writers, namely with a focus on medical terminology. By injecting domain-specific terminology and updating models with new research, AI can better grasp the nuances of various therapeutic areas, including rare diseases.

Domain-specific models can be implemented and continuously updated with the latest research and terminologies, particularly in rapidly evolving fields such as oncology and gene therapy, says Maria.

Addressing Risks in AI-Assisted Medical Writing

The integration of AI into medical writing isn’t without risks. Data privacy breaches, accuracy lapses and lack of transparency in AI decisions are primary concerns.

Minimizing risk for AI is key in handling sensitive medical information, and therefore, risk mitigation is an integral part of final quality, Maria says.

Strategies to tackle risks include:

• Data Privacy Safeguards: Anonymizing protocols and encrypting patient information within AI workflows.
• Risk-Based Content Reviews: Applying higher scrutiny to documents with critical content like adverse event summaries.
• Transparent AI Models: Ensuring AI outputs are explainable, particularly when dealing with clinical or safety data, to support informed human oversight.

Beyond Buzzwords: Using AI to Measure Efficiency and Productivity

While there are many opinions on the topic, Maria says the only way to assess AI’s impact on efficiencies is to gather relevant data on productivity gains.

This data must come from appropriate metrics that focus on ratings for accuracy, readability and compliance with regulatory guidelines.

For example, time-to-completion metrics to provide a measure of time savings can be used to quantify AI’s input on efficiency when comparing the use of AI-assisted writing processes versus traditional writing methods.

Another useful metric is the reduction in manual corrections required after initial AI drafts. Quantifying these changes can help determine whether the integration of AI is truly effective.

Additionally, tracking cost-per-document metrics can highlight reductions in labor and time expenses. From a quality perspective, reviewer feedback scores are essential for assessing the accuracy, readability and overall quality of the documents generated by AI.

Establishing a quality framework based on these criteria can help measure the effectiveness of AI and assess whether AI is simply a buzzword, or whether it can deliver tangible results. To address this, Maria recommends implementing a continuous evaluation framework. This can include periodic performance audits to compare AI-generated content against human-written content, identifying differences between the two processes.

An ongoing feedback loop with human reviewers is crucial to refine the AI model by addressing observed weaknesses. Additionally, setting clear key performance indicators (KPIs) for AI output is necessary. While there may be resistance to tracking savings due to initial results not meeting expectations, having a baseline for measurement is vital, says Maria. This baseline allows the assessment of whether the chosen KPIs are appropriate and if they reflect target benefits.

Establishing these processes and metrics ensures a robust system for evaluating and improving AI integration.

The Future of NLP in Medical Writing

“I still see that medical writing will likely remain a human-centered profession for the foreseeable future. We will use AI to assist the medical writing.”

— Maria Hopfgarten

Advancements in NLP are poised to revolutionize medical writing, offering enhancements in clinical context understanding, multilingual support and nuanced language generation.

“AI will enhance its ability to understand clinical context deeply, making outputs more accurate and relevant, especially in complex therapeutic areas,” predicted Maria.

AI can offer multilingual support, which can assist in simplifying region-specific regulatory submissions and creating translated clinical trial documents. It can also help improve language generation for areas that require more nuance, such as patient narratives or physician summaries that require contextual awareness.

Explainable AI systems can offer insights into AI processes to help medical writers evaluate recommendations. And contextual training can enable AI to better grasp intricate relationships in medical data to ensure accurate reporting of clinical findings.

Despite these advancements, Maria says there are still some limitations of AI integration in medical writing currently and in the near future. She emphasized that the role of human oversight in the accurate interpretation of data is indispensable, as it helps ensure the appropriate tone and context for regulatory and clinical documents.

And while automation can streamline tasks, “I still see that medical writing will likely remain a human-centered profession for the foreseeable future. We will use AI to assist the medical writing.”

While AI is undeniably transforming medical writing by enhancing efficiency, accuracy and productivity, the human role remains central, guiding AI to meet the stringent standards required in medical and regulatory contexts.

By embracing a collaborative, synergistic and integrative approach, medical writers can leverage AI to focus on high-value tasks, ensuring the field continues to evolve as a human-centered profession supported by advanced technology.

This article was created in collaboration with the sponsoring company and the Xtalks editorial team.