Related Experiment Video
Updated: Sep 21, 2025
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Analysis of the Cost and Case-mix of Post-acute Stroke Patients in China Using Quantile Regression and the
Mengjia Zhi1, Linlin Hu1, Fangli Geng2
1School of Health Policy and Management, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, 100710, People's Republic of China.
Purpose:
Post-acute care is fast developing in China, yet a payment system for post-acute care has not been established. As stroke is the leading cause of mortality and disability in China, patients constitute a large share of post-acute-care patients among all hospitalized patients. This study was to identify the cost determinants and establish a case-mix classification of the post-acute care system for stroke patients in China.
Patients And Methods:
A total of 5401 post-acute stroke patients in seven hospitals of Jinhua City from January 2018 to December 2020 were selected. Demographic characteristics, medical status, functional measures (eg, the Barthel Index, Mini-Mental State Examination, Gugging Swallowing Screen, Hamilton Depression Scale), and cost data were extracted. Generalized linear model (GLM) and quantile regression (QR) were conducted to determine the predictors of cost, and a case-mix classification model was established using the decision-tree analysis.
Results:
The GLM regression revealed that gender, tracheostomy, complication or comorbidity (CC), activities of daily living (ADL), and cognitive impairment were the main variables significantly affecting the hospitalization expenses of post-acute stroke patients. The QR model showed that the gender, tracheostomy and CC factors had a more significant impact on per diem costs on the upper quantiles. In contrast, cognitive impairment had a more substantial effect on the lower quantiles, and ADL significantly impacted the central quantile. Using tracheostomy, CC, and ADL as node variables of the regression tree, 12 classes were generated. The case-mix classification performed reliably and robustly, as measured by the reduction in the variation statistic (RIV=0.46) and class-specific coefficients of variation (CV less than 1.0; range: 0.18-0.81).
Conclusion:
QR has strengths in comprehensively identifying cost predictors across cost groups. Tracheostomy, CC, and ADL significantly can predict the expenses of post-acute care for stroke patients. The established case-mix classification system can inform the future payment policy of post-acute care in China.
