The incidence of depression among patients hospitalized for heart disease is quite high. Little is known about the course of depression, and even less is known about the trajectory of depression for different diagnoses of heart disease. We have two-year follow-up data from patients who were hospitalized for one of six different diagnoses of heart disease. These data are not complete, exhibiting a pattern of both intermittently missing data and drop-outs. We will present a new multiple imputation method for handling intermittently missing data. This method involves fitting a survival model to estimate the probability of remaining in the study. At each time point where intermittently missing values exist, patient survival probabilities are stratified into quintiles. A sample of scores is drawn with replacement from the patients who have depression scores at that time and within that quintile. This is used as a sample of potential depression scores and one value is drawn at random for each patient who has an intermittently missing score. After the intermittently missing values are imputed using our new method, two different existing methods for modeling monotone missing data will be compared. One approach is to use a sequential multiple imputation, where the imputations are generated using a Markov Chain Monte Carlo method. The second approach is to fit a drop-out model that combines a model to estimate the mean change over time and a logistic regression model to account for the drop-outs. These models will be applied to heart disease and depression longitudinal data.
Learning Objectives: At the conclusion of the session, the participant (learner) in this session will be able to: 1. Describe why ignoring missing data may lead to biased results. 2. Describe two methods for modeling longitudinal data in the presence of missing data
Keywords: Biostatistics, Depression
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: None
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.