Study Of Case Mix For Inpatients In Shenzhen City

Objects:The main objective of this study is to fix the case mix model which will be applicable to inpatients in Shenzhen city by using 400 thousand abstracts of the inpatient medical records during 2003 to 2005 at 12 city-owned hospitals coming from "Shenzhen information system for abstracts of the inpatient medical records" in order to provide references in controlling cost and exploring the transformation of payment method.Methods:Identify the grouping axes and variables to classify by using statistical description, rank-sum test and muti-linear regression as well as knowledge from relative literature. 80% data in every subdatabase are sampled by randomization will be designed as train data, and the rest data in every subdatabase will be designed as test data, in the course of data mining , fix case mix model by using decision tree. Select lgsumcost as target variable and age, first intensive care, multi-operation codes, co-morbility, rescuing, inter-hospital infection, operation, discharge status and incharge statue as classified variables. Fix decision tree model by variance analyse method and pre-pruning. Compute the case mix index according to the case mix model, explore the relations between hospital's index and hospital's total cost.as well as the relations between hospital's index and hospital's per-inpatient cost. Results:The outcome include 664 groups, of which 270groups classified from 50 frequent diseases, accounting for 40.66%, 394 groups classified from 63 non-frequent diseases, accounting for 59.34%. Operation, first intensive care, age become the main nodes. Assess the mode by train data and test data with RIV, nonparameter test, ROC. All indicate good results. At the same time, the standard cost decided by model output account for 81.82% of hospital actual cost, which demonstrates the rationality of the model. Compute the case mix index according to the case mix model, result show that they have positive correlation.Conclusion: The standard cost decided by model output account for most of hospital actual cost, which demonstrates the rationality and accessiblity of the model.At the same time, assess the mode by train data and test data with RIV. nonparameter test, ROC. The model have homogeneity in same group and heterogeneity in different groups.All indicate good results, achieve the study aim. In subsequent study, need to enhance the work of record managing, cost measuing, clinical pathway and so on.