Comparison of two statistical models in predicting the outcome of female systemic lupus erythematosus patients in China

Abstract

Objectives: Comparing two different statistical models to predict female SLE patients’ outcome and analyze some related factors. Methods: 1072 female SLE patients were from the Provincial Hospital of Anhui Province and The First Ancillary Hospital of Anhui Medical University from 1990 to 2000. Two types of statistical models including loglinear and Cox proportional hazard model were performed according to this data. Results: Marriage situation, family place, admission situation, whether coming from a different division, nosocomial infection, first occurrent or not and number of drug types had significant effects on LOS after fitting of a loglinear model. Related factors from Cox proportional hazard model were little more than those selected from loglinear model. Based on the former model, a female SLE patient could be predicted that how long she would stay in hospital. But from the latter model, we could predict the ratio of the probability of improvement between different groups of female SLE patients with different individual or clinical characteristics. Conclusions: Factors affecting the length of stay of female SLE patients could be selected from either loglinear model or Cox model. But these two models would be used to do different predictions.