Faisal Zain has navigated the complex intersection of medical device innovation and digital health for years, witnessing firsthand how the industry struggles to turn fragmented data into life-saving treatments. As Big Tech players like Amazon and Google move deeper into the life sciences, the challenge is no longer just generating data—it is connecting it to the human experience. In this discussion, we explore how bridging the gap between molecular AI and real-world clinical context is the key to unlocking the next generation of patient-centric therapies. We dive into the technical hurdles of harmonizing “dirty” clinical data, the power of longitudinal records in identifying hidden comorbidities, and the shift toward a continuous chain of custody for patient information that could accelerate drug development timelines.
Big Tech is rapidly advancing AI for antibody design and molecular hypotheses. How do you bridge the gap between these computational models and real-world clinical context, and what specific data types are most critical for validating AI-generated candidates in human populations?
The missing layer in today’s digital health landscape is the patient-level real-world data that provides a heartbeat to abstract molecular models. While platforms like AWS BioDiscovery or Google DeepMind’s AlphaFold 3 are incredibly powerful at a computational level, drug discovery cannot happen in a vacuum; it requires a bridge to the actual human experience to prioritize which candidates will truly perform. To bridge this gap, we must integrate longitudinal health records that include a diverse mix of electronic health records, lab results, claims, and genomics. At SEQSTER, we have focused on a massive scale, drawing from 158 million de-identified patients and 211,000 clinicians to ensure that AI-generated hypotheses are tested against the messy reality of human biology. This high-fidelity context—which includes everything from physician notes to imaging—allows researchers to see if a molecular lead actually aligns with the phenotypic realities of a target population.
Health systems often operate in silos where data remains fragmented between different providers and labs. What are the primary technical hurdles in harmonizing “dirty” clinical data, and what step-by-step process ensures that disparate ontologies become usable for enterprise-level life sciences applications?
Interoperability is the true disease of modern healthcare, where one system like Sutter Health often cannot communicate with Kaiser, and even regional branches of the same system remain isolated. The primary hurdle is that clinical data is notoriously “dirty,” characterized by inconsistent naming conventions and fragmented entries that make it nearly impossible for AI to process at scale. To solve this, we employ a rigorous back-end process: we first connect to the disparate sources, then cleanse the raw inputs to remove noise and errors. Following this, we harmonize the data by standardizing the nomenclature and ontology, ensuring that a lab result from a small clinic in San Diego matches the format of a major research hospital in Northern California. This transformation of the Electronic Health Record (EHR) into a Longitudinal Health Record (LHR) creates a single, clean source of truth that enterprise life sciences can finally trust for high-stakes research.
Clinical trial recruitment often focuses on narrow criteria, yet longitudinal records can reveal hidden patterns, such as significant comorbidity overlaps between migraine and endometriosis patients. How can researchers better utilize integrated data streams to spot these signals, and what impact does this have on recruitment metrics?
By moving away from static snapshots and toward integrated data streams, researchers can uncover clinical signals that were previously invisible, such as the fascinating link between neurological and reproductive health. In a recent collaboration with AbbVie, they initially sought just 50 migraine patients for a study, but by utilizing our longitudinal platform, we identified over 5,400 qualified candidates in just three months. Most strikingly, the data revealed that 22% of the female migraine patients on specific medications also had endometriosis documented in their electronic health records. This kind of real-time visibility transforms recruitment from a guessing game into a precision exercise, significantly reducing the time and cost required to fill trials. When you can see the complete picture of a patient’s journey, you don’t just find participants; you find the right participants who represent the complex comorbidities of the real world.
Traditional research often relies on scheduled site visits, potentially missing significant events like emergency room visits or new prescriptions. How does maintaining a continuous “chain-of-custody” for patient data change the R&D timeline, and what advantages does this real-time visibility offer over standard clinical registries?
Maintaining a continuous chain of custody for patient data fundamentally shifts the R&D timeline from a series of disjointed check-ins to a living, breathing stream of intelligence. We are currently powering initiatives like the MS iN Network with the Multiple Sclerosis Association of America and Novartis, where we track significant encounters in real time, such as a patient visiting the emergency room or starting a new medication between official study visits. This real-time visibility ensures that no critical event is missed, which is a massive advantage over standard clinical registries that rely on retrospective updates or patient memory. By capturing these interactions as they happen, researchers can identify safety signals or efficacy trends months earlier than traditional methods would allow. Ultimately, this approach saves millions of dollars and, more importantly, accelerates the delivery of life-saving medicines to the people who need them most.
Large language models are increasingly being used to help patients and researchers navigate complex medical histories. Why is provider-verified data superior to self-reported information for these conversational AI tools, and how do you determine which specific model architecture is best suited for different healthcare use cases?
Provider-verified data is the gold standard because it maintains a rigorous chain of custody directly from the electronic health record systems, whereas self-reported information is often plagued by recall bias or missing clinical nuances. When using conversational AI tools to parse through tens of thousands of patient histories, the accuracy of the output is entirely dependent on the quality of the “fuel” you feed into the model. We have integrated several major models, but we have found that Anthropic’s Claude architecture is currently the most effective for our computer software and specific healthcare use cases due to its precision and reasoning capabilities. Determining the right architecture involves matching the model’s strengths—whether it be linguistic fluidity or structured data extraction—to the specific needs of the researcher or patient. It is vital that we don’t just rely on off-the-shelf tools; we must pair verified, structured inputs with the right LLM to ensure the medical insights generated are both safe and actionable.
What is your forecast for the future of AI-driven longitudinal health records?
My forecast is that AI-driven longitudinal records will soon become as invisible and essential as the internet itself, moving from a specialized technology to a fundamental part of daily life for every patient and physician. Just as people in 1999 spoke about “searching on the internet” as a novel activity, we are entering an era where AI-powered medical insights will be the default background for every clinical decision and drug development workflow. We will see a shift where every individual has a legal, consented, and complete record of their health that they can share as easily as music was shared in the early digital era, but with the security and precision required for modern medicine. As these models become more integrated, the “interoperability disease” will finally be cured, allowing for a future where healthcare is truly predictive, personalized, and powered by a continuous flow of high-quality, real-world evidence.
