The expanding landscape of digital health and real-world evidence in pharmacoepidemiology: reflections from ISPE 2024 Annual Meeting
The International Society of Pharmacoepidemiology (ISPE) is an organization focusing on pharmacoepidemiologic research – the use of epidemiological methods to evaluate the use, benefits and risks of medicines and interventions. Key themes of annual ISPE meetings include pharmacovigilance, drug utilization research, comparative effectiveness research and therapeutic risk management. Attendees usually include individuals from academic institutions, the pharmaceutical industry, non-profit and for-profit organizations and government agencies. This year’s ISPE Annual Meeting continues to highlight the importance of real-world evidence (RWE) in regulatory science, echoing one of last year’s key themes. Additionally, the growing attention to digital health, particularly in the development and assessment of digital tools, brings exciting advancements. These innovative technologies are being increasingly applied to pharmacoepidemiological research, improving data collection, analysis, and ultimately, decision-making in health care.
In this Guest Column, Florence Ma (Associate Consultant, Lane Clark & Peacock LLP [LCP]), Saniya Deshpande (Analyst, LCP) and Ben Bray (Partner, Evidence Generation Lead, LCP) reflect on the key takeaways from the ISPE 2024 Annual Meeting, focusing on the evolving role of RWE, the integration of digital health technologies, and their implications for pharmacoepidemiologic research.
Bridging the gap between the concept of digital health and real-world applications
One of the major themes in this year’s conference was “The future of pharmacoepidemiology, powered by technology,” with a plenary talk dedicated to this topic. This included a focus on the roles of person-generated health data, artificial intelligence (AI) and machine learning (ML). The use of generative AI, particularly GPT-4, in health care was a particular focus along with key considerations to bear in mind about its accuracy and potential for biases. Martin Seneviratne (Phare Health) took listeners through the history and origins of generative AI in pharmacoepidemiology in his talk “From sparks to ignition – GenAI in healthcare & epidemiology”, ending with an emphasis on the importance of safety infrastructures around the use of generative AI. Kathryn Rough further emphasized the responsibility of pharmacoepidemiologists in her talk “Looking forward: The role of pharmacoepidemiolgists in ensuring safe, ethical, and effective use of AI/ML”. In this presentation, pharmacoepidemiologists were urged to leverage their expertise for the robust evaluation of AI and ML models in health care. Listeners were guided through the steps involved in evaluating such models, with detailed examples of how this has already been implemented.
Whilst the plenary session focused on clinical applications of AI, a separate symposium (led by students from the University of North Carolina at Chapel Hill) titled “Using generative AI to more efficiently conduct systematic reviews, solve coding problems, and communicate research findings” zoomed in on the research applications of AI in pharmacoepidemiology. These included:
- Literature reviews – sourcing relevant research papers
- Generating, explaining and debugging code
- Proof-reading and/or summarizing large documents
Both the plenary session and symposium urged researchers and other users of generative AI to be aware of the extent of its abilities, stressing that human involvement and subject expertise are integral its safe and effective use.
LCP presented another use case of generative AI in pharmacoepidemiology research – the use of GPT-4 to create a structured database of medications along with their licensed and off-license indications – with two posters, one of which was a Spotlight poster presented by Dr Ben Bray on the first day of the conference. One poster outlined the development and clinical validation of the knowledge base, whilst the second (spotlight) poster gave an overview of its external validation. Although GPT-4 was able to identify many of the licensed and off-license indications, clinical validation was essential to remove the inaccuracies and hallucinations in the output. The results of external validation suggested that the knowledge base accurately captures real-world prescribing, and may be useful in RWE research in fields including computational drug repurposing and adverse event detection. Another poster in the same session described and evaluated the use of LLMs for automated abstract generation, presented by Fanny Raguideau from HEVA (France). This highlighted the use of LLMs in a separate aspect of pharmacoepidemiological research.
Evolving data and methodological considerations in the use RWE in decision-making
The plenary session on the second day, titled “Role of real-world data for regulatory science when randomized clinical trials are not feasible: lessons learned from perinatal pharmacoepidemiology”, built on the years of discussions around real-world data (RWD) studies. It expanded on the both the limitations and potential ways forward for in using observational studies to inform regulatory decisions, particularly in the therapeutic area of perinatal pharmacoepidemiology where RCTs are often not feasible.
While there was an increasing trend in the number of studies on drug use and safety in pregnancy or perinatal period, post-exposure studies requested by FDA and EMA most commonly used data from pregnancy registries (Margulis et al., 2019), followed by electronic health records. This trend has sustained in recent years. In the presentation by Sonia Hernández-Díaz, investigation of pregnancy exposures and outcomes were framed as explicit causal questions. Methods like target trial emulation with well-defined study criteria and timeframes could be used to address the common issues in observational studies such as the lack of randomization.
These talks underlined the importance of robust data sources and methodologies in studies using real-world data. For example, the knowledge gap on medication safety in breast-feeding remain due to a lack of high-quality real-world data – an ongoing challenge that exists beyond perinatal pharmacoepidemiology.
“Understanding of the clinical and public health significance of study findings is also critical in decision-making. Safety studies in pregnancy are often rare, meaning that a high relative risk does not imply a large absolute risk. Reporting baseline rates of risk would provide context and aid decision-making.”
On day 3, the lighting session covered the current challenges and need for RWE in regulatory approval and health technology assessment (HTA). Janet Sluggett (University of South Australia) presented on the validity of claims data in long-term care facilities in Australia which had good accuracy in recording prescription medication exposure. Xinyue Liu (MSD, US) discussed the assessment framework developed for laboratory data and the subsequent assessment of five Observational Medical Outcomes Partnership (OMOP) common data model (CDM) real-world databases. Data quality was only moderate, contributed by factors such as CDM mapping failure and high number of null values. Suzanne N Landi (GSK, US) presented a framework for using RWE in safety signal assessments. Key considerations included availability and robustness of published studies, lifecycle of the product, and reliability and timeliness of real-world data sources. In the UK, National Institute for Health and Care Excellence (NICE) published the NICE real-world evidence framework in 2022. An analysis by LCP found that while RWE was used in an increasing amount of single technology appraisals (STAs) from June 2022 – January 2024, the RWE framework was only referenced sparingly (8 out of 115 submissions) and all of which used RWD to estimate treatment effectiveness.
A recurring theme was that the reliability of RWD is highly dependent on the study parameters and even the therapeutic area. For example, a data source could be useful for safety studies of psychotropics but not appropriate for effectiveness studies of over-the-counter medications. Conceptual frameworks could facilitate the development of use cases, the assessment of data quality and the reporting of study results. In the case of the NICE RWE framework, further improving transparency of its use could provide learnings for all stakeholders involved in HTA.
Integrating RWE to shape the future of pharmacoepidemiology
The ISPE 2024 Annual Meeting showcased the latest research and methods in pharmacoepidemiology, and more importantly fostered insightful discussions around the rapidly changing landscape.
“The ever-growing volume of RWE needs to be coupled with involvement of multiple stakeholders starting from study design for it to contribute to regulatory and downstream decisions in a timely fashion. This is especially crucial in nascent technologies such as products or studies involving AI/ML, where regulatory guidance is still very much being developed.”
Advanced technologies offer great promise for streamlining data analysis and uncovering patterns that may not be visible using traditional methods, but their application in pharmacoepidemiology requires careful oversight. As regulatory frameworks evolve to keep pace with these innovations, there is an increasing need for transparent methodologies, ethical considerations in data use, and continuous dialogue between stakeholders to ensure that RWE is used effectively to improve patient outcomes.
Authors
Florence Ma
Associate Consultant, LCP
Florence is a pharmacotherapy specialist and health data scientist with experience in real-world data analysis. She has a background in pharmacy practice and has contributed to various informatics and healthcare technology initiatives.
Saniya Deshpande
Analyst, LCP
Saniya is an analyst with a background in Psychology, Applied Statistics and Health Informatics and is a proficient analyst in R and Python. She has experience implementing machine learning and NLP models using real-world data.
Ben Bray
Partner, Evidence Generation Lead, LCP
Ben is a medical doctor and epidemiologist with over 10 years experience in RWE and health data science. He has extensive experiences in advanced epidemiology methods (e.g., causal inference) and machine learning analytics. He has supported many biotech and pharma clients with early phase RWE, including innovative uses of clinicogenomic RWD.
