Online Program

Improving the Quality of the DAWN Data through Imputation
Victoria Albright, RTI International 
Rong Cai, Center for Behavioral Health Statistics and Quality, SAMHSA 
*Victoria McNerney Scott, RTI International 

Keywords: Imputation, Hot-Deck, Race, Emergency Department

The Drug Awareness Warning Network (DAWN) is a survey of non-Federal, short-stay, general medical and surgical hospitals in the United States that have one or more emergency departments (EDs) open 24 hours a day, 7 days a week. DAWN collects data on drug-related visits to these EDs, including the number and type of drugs involved, the case type (e.g. suicide, accidental ingestion, etc.), as well as demographic information of the patient. For most variables, there is very little missing data, and the current DAWN methodology does not include any imputation. For example, less than one percent of the DAWN records have missing values for age or gender. However, approximately 16 percent of DAWN records in 2009 had a missing value for race/ethnicity. This missingness is not equally distributed across ED facilities. Almost 64 percent of the missing race/ethnicity data came from only 24 of the 242 facilities participating in the 2009 DAWN. To account for this variability in missingness across facilities, we propose using a hot-deck imputation method where donors are chosen within facility when overall missingness is low and across facility when missingness is high.