Clinical trials comprise the most significant portion of drug development costs. In some cases, drug R&D costs can approach as high as $2 billion per drug and are accompanied by protracted timelines of 10-15 years. Pharmaceutical companies are turning to AI to shorten discovery timelines while dramatically upholding rigorous standards.
AI-driven approaches have demonstrated meaningful gains in efficiency and diagnostic performance within controlled clinical research settings. However, evidence increasingly shows that success in trials does not guarantee real-world impact.
A 2025 peer-reviewed narrative review highlights a persistent gap between AI’s strong performance in tightly controlled environments and its variable effectiveness in routine clinical practice, where scalability, workflow integration, data heterogeneity, and governance often limit adoption. Realizing AI’s full potential, therefore, depends not only on technical validation but on translating these tools responsibly into everyday healthcare systems.
In a controlled oncology setting, an AI-driven mortality prediction system increased serious illness conversation (SIC) rates from 3.4% to 13.5%. Yet, the same body of evidence shows that such gains often do not generalize to heterogeneous, real-world clinical environments due to workflow, data, and scalability challenges.
Emerj Editorial Director Matthew DeMello sat down with Patricio La Rosa from Bayer on the ‘AI in Business’ podcast to continue their prior conversation discussing AI’s role across the clinical continuum from trial design to long-term efficacy tracking.
The following article will focus on three key takeaways from their conversation:
- Establishing scaling sensing modalities: Building AI-driven diagnostics that enable seamless transition from R&D to clinical practice without prohibitive compute costs.
- Navigating regulatory and recruitment bottlenecks: Leveraging probabilistic AI to identify robust biomarkers faster, while acknowledging unchanged approval timelines and patient engagement hurdles.
- Cybersecurity and data monetization: Protecting patient privacy amid rising threats, and exploring novel incentive models where patients share in the value their data creates.
Listen to the full episode below:
Guest: Patricio La Rosa, Head of End-to-End Decision Science in Seed Production Innovation, Bayer Crop Science
Expertise: MLOps for Industrial Analytics, Biostatistics, Biophysical Modeling
Brief recognition: Patricio is a decision science and AI leader with over 20 years of experience applying machine learning and quantitative modeling to large-scale scientific and operational challenges. At Bayer Crop Science, he leads global decision science initiatives that have embedded AI into seed production, supply chain planning, and manufacturing, driving significant business impact. His work spans industry and academia, with peer-reviewed research, extensive teaching at Washington University in St. Louis, and a focus on building scalable, responsible AI systems.
Establishing Scaling Sensing Modalities
Host Matthew DeMello opens the conversation by asking about how and where they are applying AI in clinical trials, not just for basic efficiency and automation, but also to capture meaningful data and insights that improve future trials. La Rosa responds by emphasizing that there is always an upfront base cost for infrastructure, regardless of the sensing modalities used.
He details how the initial cost in drug development might be focused on estimating a particular effect of a drug. However, once the drug is developed and the effect has been detected, the focus shifts. He goes on to explain that the new challenge is determining how to continue reliably measuring and confirming a drug’s effect in real-world use.
According to La Rosa, even though AI enables us to analyze more data than ever before, it must lead us to scalable technology that supports proper clinical practice and economic gain within reasonable economic constraints. He cautions the Emerj executive podcast audience about the need to pay for models in clinical practice that have a high computational burden during R&D.
La Rosa concludes by underscoring the importance of sensing modalities not only being scientifically valid in research environments but also being industrialized so they can be utilized in healthcare without placing unsustainable costs on patients, providers, or even insurers. He provides an example of using an fMRI that might work well during development, but ends up being too expensive in practice for ongoing patient monitoring.
Navigating Regulatory and Recruitment Bottlenecks
La Rosa goes on to explain that even with dramatic gains in biomarker discovery, the regulatory path remains unchanged. Regulatory reviews take time and are in place for a reason. Even if the regulatory path is expedited to match increased drug discovery capability, companies are still subject to limitations on their ability to find and recruit patients.
According to La Rosa, it’s not even necessary to solve these issues just yet. He thinks it’s essential to prioritize leveraging the technology first, before addressing how regulatory processes can evolve in step with its advancement and maturity.
La Rosa agrees that many different techniques and approaches are often referred to as AI, even though there are differences, and that differentiating helps build transparency and gain buy-in from the business side. Overall, he considers deep learning techniques as AI and groups the remainder under machine learning and statistics:
“We tend to use the word AI for a lot of different things, even though they are not exactly the same. I do think it’s important to differentiate, because that transparency helps with trust and buy-in from the business.
Today, I would personally reserve ‘AI’ for deep learning techniques, and then group everything else under machine learning and statistics. At the end of the day, that clarity matters less to patients, but it matters a great deal internally when organizations are deciding how to deploy, govern, and scale these systems responsibly.”
– Patricio La Rosa, Head of End-to-End Decision Science in Seed Production Innovation at Bayer Crop Science
La Rosa further explains that most patients won’t mind the blanket usage of AI to refer to seemingly different techniques. He concludes that ultimately, patients are interested in what works because they want their problem solved, particularly if they’re desperate for a solution.
Cybersecurity and Data Monetization
When asked how security considerations intersect with patient targeting in clinical research workflows and how they affect the broader system, La Rosa emphasizes the responsibility that every R&D workflow and critical assessment carries to ensure patients’ data is protected.
La Rosa also explains that the better we become at analyzing information, the more proficient attackers will become at circumventing barriers. The arms race he describes makes it essential to advance in cybersecurity as well, according to La Rosa.
The conversation shifted to the possibility that patients might eventually expect financial compensation for their data, thereby contributing to breakthroughs in therapy. LaRosa frames this as an investment in the cure, suggesting that patients could participate and benefit from the value created by their data. He also underscores a broader ethical question: how to balance individual ownership of information with collective contributions to advancing health and humanity.
Source link
#Enhancing #Clinical #Workflows #Optimizing #Efficiencies #Patricio #Rosa #Bayer
