Use of Zero-Inflated Models in Analyzing Duration Data Involving Competing Events

• Activity Number: 292
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Alok Dwivedi*+ and Jiayang Liu and Patrick Tarwater and Juan B. Figueroa-Casas and Sada Nand Dwivedi and Rakesh Shukla
• Companies: Texas Tech University Health Sciences Center and Texas Tech University and Texas Tech University Health Sciences Center and Texas Tech University Health Sciences Center and All India Institute of Medical Sciences and University of Cincinnati
• Keywords: Zero inflated log normal ; Competing events ; Duration data ; Cox regression ; log normal regression ; competing-risks regression

Abstract

Clinical biostatisticians need to frequently deal with determination of factors associated with duration data (e.g., duration of mechanical ventilation (DMV), length of stay, treatment delay). Such records often involve competing events like mortality and are normally removed during analysis. However, this reduces sample size and statistical power. As an alternative choice, Cox regression has been used by treating such events as censored. However, such consideration may not be appropriate. This also causes a generalizability issue in related population. Competing regression by treating deaths as competing events may be a better choice. We propose an alternative approach to create a composite endpoint by replacing a duration involving death as "0" and using an appropriate zero inflated model. We compared proposed approach with alternate approaches in predicting the DMV. Of 155 patients, 29 died or withdrawn from ventilation and were coded as "0". Zero inflated log normal was found to be more appropriate for describing DMV with competing events. Zero inflated approach for duration outcome involving competing events provides richer inferences as compared to other alternatives.