Background: Many observational real-world studies have historically been designed to utilize the patients and resources at a single site. In recent years, this concept has expanded to leverage patients and data at health network levels. However, the healthcare experiences of patients in the United States is often fragmented over multiple sites and networks of care. This study compares completeness of health journeys in a single network of care with an approach that leverages data from all sites of care that a patient has visited. Objectives To assess and compare the duration and completeness of health journeys of multiple sclerosis (MS) patients within a single network of care, which mimics a traditional site-based study, with a curated dataset from all facilities where patients had received care, the PicnicHealth approach.
Methods: Using PicnicHealth’s MS patient registry we simulated a dataset that would be available if patient information was restricted to one of five care networks across the U.S. (the single-network or “SN” dataset), and compared with what was observed for the same patients from all facilities where they had received care as captured by PicnicHealth (the “PH” dataset). To be eligible for analyses, patients must have had at least 10% of their inpatient/outpatient visits and at least one visit post-MS diagnosis in one of the systems of interest, to ensure that the SN dataset would “know” about their diagnosis. Patients were further restricted to those who have visits in only one of the target systems of interest.
Results: 370 qualifying MS patients residing in 37 U.S. states were identified. Compared with the SN dataset, the median patient in the PH dataset had more years of visits (SN: 4, PH: 9), neurology encounters (SN: 4, PH: 13), providers (SN: 6, PH: 15), hospitalizations (SN: 1, PH: 2), and hospital days (SN: 1, PH: 4). Using the PH dataset as a “gold standard”, the SN dataset would have observed only 47% of MS relapses, 81% of newly-initiated MS treatments, 70% of Expanded Disability Status Scale (EDSS) measurements, and 60% of neurology magnetic resonance imaging. Results from an ongoing quantitative bias analysis will also be presented at the conference.
Conclusions: Our analyses demonstrate that missing information is more likely in traditional site-based methods or analyses limited to single networks than in the PicnicHealth methodology. This can lead to misclassification and selection bias, which may result in biased insights.
