07/12/2024
Professor Harnett
What is the difference in risk of ischemic stroke for patients with a diagnosis of Atrial Fibrillation when prescribed Apixaban vs. Warfarin?
How can we do that?
Real World Evidence (RWE) in clinical research leverages data from routine clinical practice to provide valuable insights into the real-life impact of medical interventions over selected periods of time. This data comes from actual data indexed in electronic health records (EHRs) and sometimes that data is cascaded down to clinical registries. Either way, this data is referred to as “real world data” because this is retrospective data of events that actually occurred and were recorded in a structured format into a proper data schema.
Considered an augmentation to prospective (future-looking) clinical trials, RWE is a highly scalable and is a computational process, not a research subject process. The cost is fractional and can be done sometimes in hours.
There are national and global data networks that provide this capability. These networks include healthcare organizations who agree to supply de-identified data to the platform to create a uniform data repository based on clinical standards such as ICD-10, CPT, LOINC codes, RxNorm codification and others.
This de-identified, aggregate data is used by members of the ‘community’ who agree to share data in exchange to see the others’ data. A quid quo pro arrangement that allows fast analytics that can be done completely online.
This all said, RWE is somewhat criticized for an element of bias compared to randomized clinical trials where selection criteria are narrow, designed using a scientific balance between intervention and control groups, and some style of randomization, some including blinding. Like pragmatic trials where the criteria are somewhat looser, RWE makes up for some perceived bias with larger denominators.
For example, a health system that has one million patients total, an analysis can be subset into a specific phenotype and/or study design. This would reflect that organization, but not generalizable by the definition of research. But by using a large and diverse population that could increase the denominator 10-fold, or in this example, a total 100 million patients, this means a prevalent disease such as diabetes would yield hundreds of thousands of patients that can be used for RWE analysis. Propensity analysis can help to balance cohorts, and this too can be done (with some bias) using the analytics tools. Again, the numbers are significant.
For this quick demo, the denominator was ~32,000, the winner is irrelevant and results negligible, less than 1% difference, .654%.
Opmerkingen