Comparing linear regression and quantile regression to analyze the associated factors of length of hospitalization in patients with gastrointestinal tract cancers

Abstract

Objectives: The aim of this study is to compare linear regression and quantile regression to analyze the predictors of duration of staying in hospital for patients with GI cancers. Methods: This study was designed as a retrospective cross-sectional survey included all consecutive GI cancer patients admitted over one year period in a random selected hospital group located in Tehran metropolitan in 2006. Residence, age, sex and type of cancer were analyzed respectively using linear and quantile regression. Results: A total of 2,674 GI tract cancer patients were included in the study. There were 1,616 men (60.43%) and 1,058 women (39.57%). Results of the linear regression analyses showed that only type of cancer was significant. The diagnostic criteria showed that only 7% of variation has been predicted by linear regression. In spite of linear regression sex and type of hospital were significance in quantile regression analysis. Conclusions: The results have demonstrated that if the duration data showed major skewness, using quantileregression leads to better interpretation and richer inference.