The initiative, titled “Using Real-World Data to Derive Common Data Elements for Alzheimer’s Disease and AD-Related Dementias Research Through Ontological Innovation” (ReCARDO), represents a highly collaborative effort uniting expertise from ten premier academic institutions across the United States. UTHealth Houston stands at the helm of this consortium, functioning as the central hub coordinating multi-institutional research that integrates cutting-edge artificial intelligence (AI), natural language processing (NLP), and biomedical informatics.

Real-world data, which consists of information gathered from diverse sources including electronic health records, insurance claims, wearable devices, and mobile health applications, has emerged as a critical resource capable of offering granular insights into patient experiences, disease progression, and treatment efficacy outside the confines of traditional clinical trials. By systematizing and standardizing these data through common data elements derived via ontological frameworks, the ReCARDO project aims to create interoperable datasets that facilitate seamless sharing, comparison, and meta-analysis across disparate studies and populations.

Leading the charge at UTHealth Houston are three principal investigators: GQ Zhang, PhD, vice president and chief data scientist; Hongfang Liu, PhD, vice president of learning health systems; and Licong Cui, PhD, associate professor at the McWilliams School of Biomedical Informatics. These researchers bring a wealth of expertise in AI, digital innovation, and biomedical data science, positioning UTHealth Houston as an unrivaled epicenter of data-driven neurodegenerative disease research.

Dr. Zhang emphasized the unique positioning of UTHealth Houston given its concentration of multidisciplinary talents in neuroscience, informatics, and aging studies. He noted that development of a robust infrastructure to leverage real-world data is essential for enabling collaborative scientific discovery and accelerating the pace at which therapeutic strategies can be validated and deployed to patients. The vision is to harness advanced data science methodologies that not only analyze vast datasets but also improve the interpretability and utility of findings in clinical and translational contexts.

Alzheimer’s disease is an escalating public health crisis, affecting over 7 million Americans aged 65 and older, a figure projected to swell dramatically in the coming decades. This neurodegenerative condition is marked by progressive cognitive decline and poses profound challenges for patients, caregivers, and healthcare systems worldwide. The ReCARDO initiative seeks to catalyze novel insights by empowering researchers with data-centric tools and common standards that enhance reproducibility and cross-study integration.

A key scientific ambition of the project is to develop sophisticated AI algorithms and natural language processing tools that can extract meaningful information from unstructured clinical narratives, imaging reports, and other heterogeneous sources. These technological innovations will underpin the creation of a unified ontology guiding the derivation of common data elements, thereby standardizing variables across institutions and datasets to enable more precise and scalable analyses.

Dr. Hongfang Liu highlighted the transformative potential of this approach, describing it as a critical step toward translating real-world data into actionable evidence that can directly inform clinical decision-making and healthcare policies. By positioning UTHealth Houston as the national center for real-world data science in Alzheimer’s research, the project aims to establish a sustainable platform for continuous discovery, validation, and dissemination of findings with immediate public health relevance.

The consortium includes other eminent principal investigators such as Ronald Petersen, MD, PhD at Mayo Clinic; Zoe Arvanitakis, MD, MS at Rush University; and Yong Chen, PhD at the University of Pennsylvania. Co-investigators from partner sites bring complementary expertise spanning neurology, data science, and epidemiology, ensuring a comprehensive multidisciplinary framework supporting the initiative’s ambitious goals.

UTHealth Houston is also contributing a robust team of co-investigators from its Department of Neurology, McWilliams School of Biomedical Informatics, Cizik School of Nursing, and School of Public Health. This breadth of expertise facilitates a holistic approach to studying Alzheimer’s disease, encompassing molecular mechanisms, patient outcomes, caregiver impacts, and population health dynamics.

As the ReCARDO project advances, it is poised to create a paradigm shift in how Alzheimer’s disease and related dementias are studied, enabling real-time synthesis of evidence from diverse real-world sources and accelerating the translation from data to therapeutic discovery. The integration of ontological innovation and AI-driven analytics heralds a new era of precision medicine strategies tailored to the complexities of neurodegenerative disorders.

This landmark initiative underscores UTHealth Houston’s commitment to addressing one of the most pressing challenges in contemporary medicine through collaborative, data-intensive science. By leveraging its deep expertise in informatics and neurotechnology, UTHealth Houston aspires to contribute decisively to global efforts aimed at mitigating the devastating toll of Alzheimer’s disease.

Subject of Research: Alzheimer’s Disease and Related Dementias, Real-World Data, Ontological Innovation, Artificial Intelligence in Biomedical Informatics

Article Title: UTHealth Houston Leads $27 Million National Initiative to Harness Real-World Data for Transformative Alzheimer’s Disease Research

News Publication Date: Not specified in the content

Web References:
https://www.uth.edu/data/contact-us
https://sbmi.uth.edu/faculty-and-staff/hongfang-liu.htm
https://sbmi.uth.edu/faculty-and-staff/licong-cui.htm
https://www.alz.org/getmedia/c05f7ba4-9aea-4cb0-8898-5e8bff3f0930/executive-summary-2025-alzheimers-disease-facts-and-figures.pdf

Image Credits: UTHealth Houston

Keywords: Neurodegenerative diseases, Dementia

The study collected a dataset that included 62 text-based questions and 8 image-based inquiries from reputed medical institutions, further enriched by insights from the U.S. Centers for Disease Control and Prevention (CDC). By focusing on text-based queries, researchers evaluated several prominent AI models, including Gemini Pro, Claude-3, Claude-2, GPT-4, and GPT-3.5. Each model’s responses were measured, granting a comparative insight into their efficiency and effectiveness across various metrics. In contrast, the image-based questions were tackled by deploying the capabilities of Gemini Pro Vision and GPT-4V. This bifurcation allowed for a comprehensive assessment of their performance across diverse formats.

The findings of the study revealed impressive performance across the board when it came to text-based questions. Gemini Pro stood out with a Reference Answer Accuracy Rate (RAAR) of 79.4%, followed closely by Claude-3 with 78.7%. This was complemented by Claude-2, GPT-4, and GPT-3.5, which exhibited RAARs of 76.5%, 75.7%, and 70.3%, respectively. Such accuracy rates underline the capability of these models to comprehend and process complex information inherent in medical training and biosafety protocols. Their ability to generate benchmark responses not only assists in learning but also promotes better understanding of essential biosafety concepts among future researchers and practitioners.

Simultaneously, the investigation into the image-based queries highlighted that GPT-4V was the frontrunner, outperforming its counterpart, Gemini Pro, with RAARs of 78.7% and 76.5%, respectively. This indicates that while both models have robust performance metrics, GPT-4V has a slight edge when processing visual data, which is particularly critical in a laboratory setting where accurate interpretation of images can influence safety outcomes. The multidimensional capabilities of these AI systems could pave the way for enhanced training modules, where real-time visual monitoring and diagnostics could become integral elements of biosafety education.

Despite these promising advancements, the study and broader discussions around generative AI in biosafety reveal a series of limitations that cannot be overlooked. Bias within AI models remains a significant concern, as it can lead to erroneous outputs, especially in sensitive fields like medicine. Additionally, the quality of training data directly influences model effectiveness, where the lack of high-quality datasets pertaining to rare events poses challenges to reliable AI performance. These limitations are compounded by the dynamics of real-time processing, which is essential in fast-paced laboratory environments.

The ethical implications surrounding AI usage in medical fields also call for critical scrutiny. Issues of privacy, lack of transparency, and potential ethical breaches necessitate careful navigation. As these technologies gain traction, the call for implementing precautionary measures becomes increasingly urgent. Establishing uncertainty markers, automating bias detection mechanisms, and promoting human-AI collaboration are pivotal steps in addressing inherent challenges posed by these advanced models.

A balanced approach involving robust verification systems and accountability mechanisms is essential for ensuring responsible AI deployment in healthcare and biosafety domains. Developing standardized datasets and engaging in federated learning could contribute to refining AI systems’ learning processes while minimizing disadvantages rooted in traditional hierarchical structures. Furthermore, the research community is urged to focus on explainable AI, which would augment trust between researchers and AI technology by demystifying the operational processes behind AI-driven decisions.

As the landscape of biosafety training and laboratory research evolves, the importance of models such as ChatGPT and Gemini cannot be overstated. Their capacity to provide personalized learning experiences, automate material generation, and support course design creates a transformative opportunity for medical education. Leveraging the strengths of AI can lead to a new paradigm of knowledge acquisition, wherein future researchers are better equipped to face challenges in biosafety through enhanced training modules.

The implications of integrating AI within biosafety laboratories extend beyond mere educational tools; they encompass the potential for real-time monitoring, predictive maintenance, and anomaly detection. Such applications not only promise to improve educational frameworks but also significantly bolster laboratory safety protocols. By ensuring a stable operational environment, researchers can focus on innovation while being assured of their safety protocols.

A future-oriented vision for LLMs reveals their potential not just in addressing historical queries but also anticipating future biosafety challenges. By training these models with extensive datasets drawn from realistic scenarios, researchers can utilize them to simulate responses and training procedures, preparing for a diverse array of situations. This proactive approach could drastically reduce risks associated with laboratory operations, ultimately enhancing public health outcomes.

In conclusion, while the potential of generative AI in biosafety is promising, enhancing accuracy and mitigating risks must be at the forefront of future endeavors. The continuous evolution of these technologies calls for a collaborative effort within the research community to tackle inherent challenges while maximizing the benefits. The ongoing dialogue surrounding AI’s role in healthcare will likely lead to groundbreaking developments, with biosafety standing as a pivotal area for innovation.

As researchers and institutions continue to explore the potential of AI, the collaboration between human intelligence and machine learning is set to revolutionize the field of biosafety. The key to overcoming current limitations lies in a commitment to responsible development, informed by empirical research and ethical rigor. Indeed, the future of biosafety in the context of AI is a journey that promises to redefine educational landscapes and enhance public safety in unprecedented ways.

Subject of Research: Evaluation of AI models in biosafety laboratory settings
Article Title: Performance of Large Language Models in Biosafety Laboratories
News Publication Date: October 2023
Web References: Chinese Medical Journal
References: Chang Qi, Anqi Lin, Anghua Li, Peng Luo, Shuofeng Yuan
Image Credits: Chang Qi, Anqi Lin, Anghua Li, Peng Luo, Shuofeng Yuan

Keywords

Applied sciences, AI in healthcare, biosafety training, large language models, medical education