Longitudinal regression model for a continuous skewed outcome with zero inflated observations with application to oropharyngeal cancer cost

• Activity Number: 358 - SPEED: Statistics in Epidemiology
• Type: Contributed
• Date/Time: Tuesday, August 1, 2017 : 10:30 AM to 11:15 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ho-Lan Peng* and Wenyaw Chan and Chi-Fang Wu and Hongyu Miao and David R Lairson
• Companies: and University of Texas Health Science Center at Houston School of Public Health and University of Texas Health Science Center at Houston School of Public Health and University of Texas Health Science Center at Houston School of Public Health and University of Texas Health Science Center at Houston School of Public Health
• Keywords: Zero-Inflated Regression ; Gamma ; MarketScan ; Oropharyngeal Cancer ; Longitudinal ; Skewed outcome

Abstract

Zero-inflated regression has often been used in count outcome model such as Poisson or negative binomial outcome when excess zero count data are analyzed. For continuous outcome data, when unproportional zero values occur, very few zero-inflated regression models are available in the literature. In this study, we develop two models to analyze longitudinal health cost data. One uses the mixture of Bernoulli and Gamma distributions as the dependent variable and the other adopts the sum of Poissonly many gamma distributions as the dependent outcome. Random effects will be included in these regression models to handle the within-subject correlation and between-subject heterogeneity. Using traditional gamma regression, the theoretical probability that the monthly costs are less than 68 dollars is only 18% while the proportion of monthly costs that have a zero value is 30%. We will apply the proposed method to Truven MarketScan Commercial Claims and Encounter Database in measuring the differential cost due to oropharyngeal cancer. Simulation will also be conducted to assess the performance of the statistical model.