For a patient, participating in a clinical trial is complicated to organize, in addition, the drugs tested are very rarely effective in the field of neurodegenerative diseases. However, it seems that there are unexpected benefits to volunteering for a clinical trial.
Clinical trials in neurodegenerative diseases are often disappointing, there are probably many reasons for that situation. A core aspect of clinical trials is that the population who received the treatment should be representative of the general population. For logistic reasons, biotechs that have few employees have to subcontract clinical trials. Principal investigators and subcontractors have every reason to select patients who present a textbook-like disease.
Suspecting that the clinical trial population is not representative of real-life patients, an international team wanted to characterize the progression of Parkinson's disease using real-world data to guide the design of clinical trials and identify subpopulations.
The increasing availability of real-world data, and recent advances in natural language processing, particularly large language models, allow for an easier and more granular comparison of populations than before.
This study includes two research populations and two populations derived from real-world data.
The research populations are the Harvard Biomarkers Study (935 patients), which is a longitudinal biomarker cohort study with structured in-person study visits, and finally Fox Insights (36,660 patients), a research study based on the Michael J. Fox Foundation online self-survey.
The real-world cohorts are Optum Integrated Claims electronic health records (157,475 patients), representing large-scale linked medical and claims data and de-identified data from Mass General Brigham (Mass General Brigham, 22,949 patients), a University Hospital.
Structured, anonymized data from Mass General Brigham's electronic health records is augmented using natural language processing with a large language model to extract measures of Parkinson's disease progression. This extraction process is manually validated to verify accuracy.
Motor and cognitive progression scores change more rapidly in the Mass General Brigham than in the Harvard Biomarkers Study (median survival to H&Y scale: 5.6 years versus more than 10 years); median decline to mini-exam of mental status 0.28 versus 0.11. In real-world populations, patients are diagnosed more than eleven years later! After diagnosis, in real-world cohorts, treatment with Parkinson's drugs is initiated 2.3 years later on average than for patients in clinical trials.
This study provides a detailed characterization of Parkinson's disease progression in various populations. It delineates systemic discrepancies between patient populations enrolled in research settings and real-world patients.
The study shows systematic differences and potential directional biases between research and real-world datasets. Patients in research populations are diagnosed much earlier, start levodopa and other Parkinson's medications earlier, and show slower changes in clinical scales of motor and cognitive progression. Real-world-based populations are diagnosed at older ages, start medications later than research cohorts, and experience more rapid changes in clinical scales.
These discrepancies are likely due to a combination of selection bias, but exact attribution of causes is difficult using existing data. This study emphasizes the need to diligently consider potential biases when planning a clinical trial.