Response by Ahmad et al to Letter Regarding Article, “Validity of Cardiovascular Data From Electronic Sources: The Multi-Ethnic Study of Atherosclerosis and HealthLNK”

In Response:We appreciate the interest of Mewton et al in our work. 1We agree with the assertion that fundamental differences exist in the data collection and verification processes in traditional clinical studies (either observational cohort studies or clinical trials) and those from large healthcare databases.However, the issues with data quality exist across all research methodologies.For example, the time-and resource-intensive methods used in traditional cohort studies often rely primarily on participant self-report, which can result in underreporting.Our work highlights the overlap, discordance, strengths, and limitations of a traditional community cohort compared with an electronic data research network.Furthermore, our findings suggest that multiple methods of outcomes ascertainment may result in more complete event ascertainment.Depending on the study question and resources, these methods may include traditional medical records review and adjudication, electronic health records (EHRs), claims records, patient-reported events, and novel methods such as geofencing from smartphones. 2 As researchers continue to utilize existing and novel electronic sources for event ascertainment, accompanying methodological work on the accuracy, limitations, and efficiency of these approaches, or more likely combination of approaches, will be needed.We would also like to note that for large healthcare databases, the challenges of data quality vary by data source, which most often include data from EHRs, insurance claims, and registries, which can have multiple data feeds.Our work focuses on data quality issues in an EHR-derived database within the fragmented, US healthcare system, which raises unique issues compared with those in SWEDE-HEART (The Swedish Web-system for Enhancement and Development of Evidencebased care in Heart disease Evaluated According to Recommended Therapies), a national cardiac registry in Sweden. 3Thus, we cannot recommend a single set of recommendations to improve data quality in large healthcare databases.With the use of EHR data becoming more common, we, as a research community, need to continue to adopt standards for publication to ensure reproducibility and high quality research, which may include wider adoption and possibly expansion of existing international guidelines, such as RECORD (Reporting of Studies Conducted Using Observational Routinely-Collected Data) and STROBE (Strengthening the Reporting of Observational Studies in Epidemiology ). 4,5Key steps in need of further development for analyses of large healthcare databases, especially for EHR-derived databases, include: (1) Standards for reporting of data quality and the analytic pipeline, including preprocessing, cleaning, visualizations, and statistical analysis; (2) Standards for the publication of data dictionaries and documentation specifying data quality issues and recommended analytic approaches for large healthcare datasets; and (3) The development and testing of methods to account for biases introduced when using EHR data.