Using Statistical Approaches to Stratify Hospital-Readmission Risk After Hip Fracture

• Activity Number: 310 - Topics of Variable Selection
• Type: Contributed
• Date/Time: Tuesday, July 31, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Qingqing Dai* and Zhuqi Miao and Lan Zhu
• Companies: Oklahoma State University and Oklahoma State University and Oklahoma State University
• Keywords: readmission; logistic regression; group lasso; risk stratification

Abstract

The hip fracture readmission has brought tremendous economic burden to both patients and hospitals. Previously published readmission risk scoring systems seem promising in stratifying general patient population into different risk groups. But they were not derived specifically for hip fracture readmission, therefore, the reimplementation results from these scoring systems are not satisfactory for hip fraction patients. In this paper, we focus on hip fracture patients' records and develop a risk stratification tool for this population, using logistic regression with group lasso. Unlike lasso, whose solution varies with the way we encode categorical variables, group lasso ensures all the dummy variables that encode each categorical feature to be either included or excluded from the model together. Furthermore, group lasso estimators have been shown to be statistically consistent. 39,513 elective index admissions of hip fracture patients aged 50 years or older, with 192 features from Cerner Health Facts database, are included in this study. The regularization parameter is tuned through the cross-validation process. The final scoring system will help in medical decision making.