PS2-09: Combining Clinical Databases With the EHR to Identify a Study Population: An Example Using Acute Myocardial Infarction.
ABSTRACT Background/Aims: Study populations are commonly identified using a single data source. However, the inclusion criteria may result in an unacceptable level of false positives/negatives. The utilization of multiple sources may allow for more complex inclusion criteria, therefore improving the accuracy of identification of cases and controls for use in research. For this study, Geisinger's electronic health record (EHR) was combined with clinical databases to identify a population of patients with suspected Acute Myocardial Infarction (AMI). Methods: The Geisinger Acute Myocardial Infarction Cohort (GAMIC) database was created to conduct research on AMI. Geisinger databases were queried to identify all patients who had elevated cardiac enzymes during an inpatient hospital admission at Geisinger Medical Center from Jan. 1, 2001 to Dec. 31, 2006. Data from the resulting patients were gathered and extensively reviewed for validity and coherence. When possible, overlapping data were collected from different sources to check for internal consistency. Discrepancies between sources were reviewed and resolved. The medical charts of patients who did not go to catheterization were manually reviewed to determine the reason for not going. Similarly, charts of patients with elevated cardiac enzymes who also had invasive surgery on the same visit were reviewed to determine the timing of the enzyme elevation. After applying all criteria, a cohort of AMI cases was derived and retained in the final GAMIC database. Results: There were 3,625 patients who had elevated cardiac enzymes during an inpatient encounter; 58% went for catheterization. Of the 3,625 suspected AMI cases, the most common reasons for exclusion were elevated enzymes occurring post open-heart surgery (n=275) and admissions with elevated enzymes but with a recent catheterization (n=61). Only 3,265 (90%) fulfilled all criteria applied from the diverse set of data sources. Interestingly, of the 360 that were excluded and the 3,265 AMI cases, 40% and 90% had an AMI discharge diagnosis for the inpatient admission, respectively. Conclusions: A more careful approach to identifying a study population and validating the data for internal consistency results in greater accuracy. In addition, more resources utilized in the beginning of the study to better capture the population of interest can lead to a reduction in resources needed later in the study if issues or questions arise from a poorly identified cohort.