A Latent-Class Mixture Model for Incomplete Data with Applications to a Trauma Transfusion Study

• Activity Number: 559
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: International Chinese Statistical Association
• Author(s): Mohammad Hossein Rahbar*+ and Jing Ning and Sangbum Choi and Jin Piao and Chuan Hong and Hanwen Huang and Deborah J. del Junco and Erin E. Fox and John B. Holcomb
• Companies: University of Texas Health Science Center at Houston and MD Anderson Cancer Center and University of Texas at Houston and University of Texas at Houston and University of Texas School of Public Health and University of Georgia and University of Texas Health Science Center at Houston and University of Texas Health Science Center at Houston and University of Texas Health Science Center at Houston
• Keywords: Latent Class Models ; Mixture Model ; Classification ; Incomplete Data ; Posterior Probability ; Transfusion

Abstract

We discuss development of an extended finite mixture model to classify a latent class membership based on the joint distribution of an incomplete Gaussian and a binary outcome. The proposed research is motivated by a trauma transfusion study that aims to identify trauma patients who suffer from severe hemorrhage (SH) and may require activation of a massive transfusion protocol. Traditionally, massive transfusion has been defined as transfusion of at least 10 units of red blood cells (RBCs) within 24 hours of emergency department admission. This definition has never been validated as a surrogate for severe hemorrhage and is subject to misclassification. We define SH as a latent variable in a logistic regression that is linked to a linear regression that models the log of the number of RBCs used and accounts for potential induced censoring and survival status through another logistic regression model. We develop a generalized EM algorithm to estimate the posterior probabilities of membership as SH. We evaluate the performance of our method using a simulation study with applications to data from a retrospective trauma transfusion study.