Sameena Salvucci, PhD¹, Hannah Kyeyune, MS¹, Sameer DeSale, MA¹, Lev S. Sverdlov, MD, PhD¹, and Thomas M. Brady, PhD². (1) Synectics for Management Decisions, Inc., 1901 N. Moore Street, Suite 900, Arlington, VA 22209, 703-807-2309, sams@smdi.com, (2) Office of Applied Studies, Substance Abuse and Mental Health Services Administration, Parklawn Building, 5600 Fishers Lane, Suite 16-105, Rockville, MD 20857

This paper describes an imputation strategy employed in a study of addiction treatment. The Alcohol and Drug Services Study (ADSS), a nationally representative sample survey of substance abuse treatment facilities and clients, was used to examine the association between retention and level of methadone dosage. One of the key variables in this study, methadone dose had a high level of nonresponse (28 percent). The dependent variable, retention, as measured by the average length of stay (ALOS) was lower among the cases where methadone dose was missing as compared to those where methadone dose was reported (461 days and 591 days, respectively). Thus, omitting all cases where the methadone dose was missing could introduce significant bias. We present two types of methods used to impute missing methadone dose values: (1) PROC IMPUTE, a regression-based distributional estimation procedure that is believed to be more general and to produce more accurate results than a standard “hot deck” procedure, and (2) a logical imputation. Simulation study results compare analyses based on imputed data using these two methods to analyses based only on cases where methadone dosage was reported. Differences in point estimates and their standard errors will be presented. Results showed PROC IMPUTE as the most appropriate procedure to use to impute methadone dosage. Imputation has become one of the most popular tools to deal with missing value problems in survey data analyses because it prevents the loss of information due to deletion of incomplete records and can reduce nonresponse bias.

Learning Objectives: Learning Objectives

At the conclusion of the session, the participant will be able to
• Describe missing methadone dosage data patterns in the Alcohol and Drug Services Study.
• Articulate the features of the PROC IMPUTE imputation procedure.
• Describe the differences between estimates obtained using PROC IMPUTE, logical imputation, and deletion of those who did not report methadone dosage.

Keywords: Substance Abuse Treatment, Statistics

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.

Statistical and Data Analysis Issues for Health Services Research

The 132nd Annual Meeting (November 6-10, 2004) of APHA