Biodesign in the generative AI era: enhancing innovation and equity with NLP and LLM tools

Over the last two decades, the biodesign process1 has inspired a multitude of innovations in medical devices and digital health,2 benefiting millions of patients.3 Building on this foundation, a concept exploration and testing framework for machine learning medical devices based on biodesign principles has also been proposed.4 Nevertheless, traditional biodesign and biomedical innovation are presented with challenges as well as opportunities with the advent of generative artificial intelligence (AI).5 The biodesign process starts with identifying unmet clinical needs through clinical immersion in a clinical setting. This step is crucial to producing high-value needs but is also often time consuming, resource intensive and constrained by the accessibility of healthcare environments and professionals.6

Park et al developed a natural language processing (NLP) pipeline that extracts breast cancer treatment outcomes from electronic health records of women from under-represented populations.7 The pipeline has been shown to perform well in extracting breast cancer treatment outcomes with a high area under the curve. The study demonstrated that NLP could process unstructured data by analysing clinical notes and revealing health disparities that might have been previously overlooked.

These qualities highlighted NLP’s immense potential for the clinical immersion phase of biodesign, where teams observe clinical settings, interview stakeholders and identify unmet needs. NLP could analyse clinical notes and interview transcripts to uncover meaningful findings, challenges and unmet needs. It could also summarise large volumes of qualitative data and uncover inequities in care delivery that might be missed during brief clinical observations, allowing teams to focus more on creative problem solving. While clinical immersions in real clinical settings are essential for innovators to gain first-hand insights, NLP could augment them, providing richer, data-driven insights that complement observations and interviews. The combination of human empathy and AI-driven analysis ensures a more comprehensive understanding of clinical needs.

Moon et al demonstrated how large language model (LLM) tools like GPT-4 simplify tasks such as trend identification and regulatory analysis, lowering barriers to medical device development.8 The review illustrated that LLM tools could provide up-to-date, contextual knowledge that significantly reduce the effort and costs associated with these steps. The LLM tools could be even more powerful when leveraging retrieval augmented generation. It could empower teams from diverse backgrounds and resource levels to participate in innovation, fostering inclusive solutions that reflect a broad range of perspectives and needs.

While NLP and LLM tools are powerful means that facilitate and enhance the biodesign process, their role extends beyond merely supporting the identification and assessment of clinical unmet needs. During the concept generation phase of biodesign, teams should be aware of the transformative potential of these generative AI era technologies during the brainstorming process. Just as robotic surgery has become a core component of modern surgical practice, the two editors’ choice papers in this issue of BMJ Health & Care Informatics demonstrate the potential of NLP and LLM tools in innovative medical devices and healthcare systems. These tools can extract insights from both unstructured data and structured databases, eventually becoming integral to future innovations.7 8

As demonstrated in the review, LLMs can support innovation in diverse fields, such as cardiovascular intervention and hepatobiliary diagnostics.8 However, the effectiveness of these tools is highly dependent on the quality of input data provided by innovators—NLP tools rely on the accuracy and completeness of clinical records, while LLM outputs are shaped by the reference data used for analysis. Although technologies can broaden access to biomedical innovation, particularly for resource-limited innovators, disparities in access to high-quality data sources may perpetuate gaps in the innovation process, underscoring the need for equitable data sharing and support.

Nonetheless, over-reliance on these technologies could introduce additional risks in healthcare. AI models may inadvertently reflect biases found in their training data. Inadequate data privacy safeguards for AI tools usage may result in unauthorised exposure of sensitive information, breaching patient consent and raising serious ethical concerns. Excessive dependence on AI could also erode human vigilance, compromising critical judgement and increasing the likelihood of harmful design flaws or clinically detrimental decisions. A balanced approach, combining AI insights with rigorous human oversight, is essential for safe and ethical healthcare.

By embracing NLP and LLM technologies, biodesign can evolve into a more efficient, inclusive and data-driven process. This new approach lowers barriers to innovation, deepens insights through AI-enhanced human observation, promotes equity and drives innovative solutions incorporating advances in generative AI. The future of biodesign lies in a dynamic collaboration between human ingenuity and AI-driven analysis, resulting in healthcare solutions that are more innovative, inclusive and impactful. This marks the dawn of ‘Biodesign in the Generative AI Era’, where NLP and LLM tools facilitate needs identification, enhance concept generation, democratise innovation and advance equitable healthcare solutions.