IDEA/NEED: Automated ClinicalTrials.gov and Biomedical Literature Search & Summary

New medicines development needs a lot of initial and ongoing review of similar clinical trials as well as extraction of clinical trial characteristics from registries like clinicaltrials.gov, and from biomedical literature like the articles found on PubMed.gov search.

With the advent of large language models like GPT-4, search and summarization of clinical trial entries and biomedical literature may change for the better.

**** Finding and comparing multiple clinical trials for similar indications ****

An indication is a specific disease and often a specific sub-population of people with a certain disease. For instance, there is a common lung disease called COPD, often found in current or former smokers. The subpopulation of interest might be people with an absolute blood eosinophil count of at least 300 (based on a blood test). In this case, clinical trial designers often need to rapidly gather all of the similar clinical trials together, and then download and extract details from the clinical trial protocols. Often the protocols are contained in the Supplementary Material on the web pages hosting the full-text biomedical article (example: https://www.nejm.org/doi/full/10.1056/NEJMoa2303951 ). Other times the protocols are available on clinicaltrials.gov as a link at the very end of a clinical trial registry entry.

Comparing clinical trials can help study designers in a few ways
- Comparing the biostatistical design of clinical trials. For example, there is a type of design called an adaptive clinical trial.
- Identifying and comparing the types of outcomes measured in the clinical trial, or "endpoints"
- Evaluating how feasible it was to run a type of trial -- particularly if the clinicaltrials.gov entry or research publication made any remarks that the trial could not be completed or did not recruit to its target.
- Comparing the schedule of events (SoEs) or schedule activities (SoAs) within the protocols, to understand how often visits happen, what aspects are measured when, and opportunities to simplify the schedule to be less burdensome to patients or to the clinical research sites doing the study at each location

**** Finding safety-related publications on an automated and thorough basis ****

Some regulators like the FDA have different drug safety monitoring requirements, depending on the region and regulatory body. The FDA, for example, expects manufacturers to review relevant published studies and abstracts and submit reports for those that meet specific criteria, including identification of patient, drug, reporter, and adverse event. At the same time, manufacturers also are interested in publications that confirm the relative safety of different drug products, such as studies that may confirm no additional safety risks beyond what was already found in clinical trials, or even post-marketing data analyses that showed the drug was safer than expected in real-world use. In both cases, automated biomedical literature surveillance and summarization may be helpful to the Safety departments of therapeutic drug manufacturers.

**** Extracting quantitative data from prior clinical trials and clinical research publications ****
Often the tables and figures from prior clinical trials and biomedical publications are useful to bio-statisticians and others planning a clinical trial. Although the detailed data are sometimes not available, being able to extract an estimate of the detailed data from figures and other information in the papers can be helpful in design and planning.