Characterizing and managing missing structured data in electronic health records: Data analysis
The electronic health record (EHR) was originally built with improved patient outcomes and data portability in mind, not necessarily as a resource for clinical researchers. Beaulieu-Jones et al. point out that as a result, researchers examining EHR data often forget to take into account how missing data in the EHR records can lead to biased results. This “missingness” as they call it poses a challenge that must be corrected for. In this 2018 paper, the researchers describe a variety of correctional techniques applied to data from the Geisinger Health System’s EHR, offering recommendations based on data types. They conclude that while techniques such as multiple imputation can provide “confidence intervals for the results of downstream analyses,” care must be taken to assess uncertainty in any correctional technique.
