Data mining-based detection of rapid growth in length of stay on COPD patients

Background: Due to the rapid development in industrialization and urbanization, air pollution has aggravated in the past few years, chronic obstructive pulmonary disease (COPD) has become the world's fourth leading cause of death. Through the study of factors affecting medical cost, we find that there is a strong correlation between medical cost and length of stay (LOS). Therefore, it is necessary to analyze and predict the LOS for patients with COPD. Objective: To identify COPD patients with rapidly increased LOS next year by the method of data mining. Methods: Based on the data from healthcare survey data in one city of China, we extracted the 36985 patients with COPD who had been admitted to hospital, from January 2011 to December 2012. On the basis of LASSO Logistic regression and random forests (RF), we established the prediction models. Results: The precision and accuracy of the method was proved correct and reliable (the AUC of LASSO: 0.62, the AUC of RF: 0.63) when employing 29 factors. Conclusion: This study provides an accurate and applicable method to detection the potential patients who are most likely to rapid growth in LOS.