Defining Depression Cohorts Using the Electronic Health Record: ICD-9 Codes Versus Medication Orders

Background: Electronic health records (EHR) allow health care researchers to conduct unprecedented large-scale studies on diseases, treatments and health care system utilization. EHR studies are limited by the quality of the data set available. Careful consideration must be given to how to define patient cohorts. One approach aimed at limiting the number of nonclinically relevant patients included in a cohort is rigid inclusion criteria. With rigid inclusion criteria, however, we run the risk of excluding those with clinical features who are receiving treatment but do not meet these criteria. They may not meet these criteria due to patient or provider bias against including certain features like ICD-9 codes in their health record, or perhaps there are administrative data sequestration protocols inherent in the system, barring researcher access to pertinent patient information. This may be the case with certain psychiatric conditions. Methods: We have compared two methods of defining a cohort of depressed patients using information in the EHR. Results: We show that either using ICD-9 codes for depression or medication orders for antidepressants results in exclusion of potentially clinically relevant patients in both cases. We also show that both of these methods result in cohorts with highly correlated clinical features such as emergency department usage and primary discharge diagnosis codes, outpatient clinic visitation frequency and inpatient discharge diagnosis codes. Conclusion: For the case of defining a cohort to study depression, less rigid electronic phenotypes may better capture patients who are receiving some sort of treatment for depression