The Study Of Hospitalization Costs And Case Mix In Pneumonia Patients Of Tianjin Medical Insurance

Objective: In the decades, the diagnosis related groups(DRGs) had become the world-recognized advancing payment. But there were few studies in China what were about pneumonia DRGs. Therefore, by through analyzing the cost of pneumonia patients with medical insurance, in 2013, Tianjin, we established a predict law of pneumonia patients with medical insurance. Then provided a reference to establishing DRGs prepaid way.Methods: Data was from 2013 Tianjin personnel database system for medical insurance. We used general statistical description to describe the hospitalization costs. We established forecast rules via the rough set theory. With the core idea of DRGs medical insurance payment methods as a guide, we used decision tree in data mining analysis methods for grouping the objects of this study. We used SAS 9.1 to accomplish data management, attribute selection and assignment process. Rosseta software was carried out on the rough set table attributes reduction and rules creation. When we evaluated the established rules, we used SPSS 13.0. And we used CART algorithm in SPSS Clementine 12.0 to build and prune the decision tree.Results: This study included 10963 patients who attended Tianjin medical insurance in 2013. There were more men(58.10%) than women(41.90%); the median age was 68, interquartile range was 24, in the majority with the elderly; the patients number in tertiary hospitals was the most(54.48%), followed by secondary-class hospitals(29.29%) and first-level hospitals(16.23%); the vast majority of patients were those who did not receive surgery(90.66%), a few patients received surgical treatment(9.34%); the median number of days in hospital was 11 days, interquartile range was 6 days. The median hospital total cost was 9483.47 yuan, interquartile rage was 9340.38 yuan. Among them, the median hospital total cost were that first-level hospitals(4147.08 yuan), secondary-class hospitals(8164.71 yuan) and tertiary hospitals(13013.50 yuan), respectively. Interquartile ranges were first-level hospitals(2724.89 yuan), secondary-class hospitals(5381.22 yuan) and tertiary hospitals(10992.65 yuan), respectively.(1) Rough set rules results: we chose 6 condition attributes such as gender and age. Hospitalization expense was the decision attribute. The decision table was established based on rough sets theory after reduction. At last we established 11 predict rules. The forecast accuracy was 83.90%. The area under the ROC curve(AUC) and 95% CI were 0.729(0.697, 0.760).(2) The decision tree results: We built a decision tree trough CART algorithm. It included 6 basic variables, such as hospitalization days and hospital grades. Their importance were hospitalization days(0.449), age(0.194), hospital grades(0.180), surgery or not(0.104), diagnosis(0.059) and gender(0.016). The forecast accuracy was 84.18%. The area under the ROC curve(AUC) and 95% CI were 0.770(0.744, 0.796).(3) The results of comparing the two models: Z inspection results of the two models' ROC curves was that Z value was 1.989 and P value was 0.047. According to the inspection level of ?=0.05, we could think that the two data mining methods' ability difference of classification was statistically significant. And the classification ability of the decision tree CART algorithm was better in classing hospital total cost of 10963 patients who attended Tianjin medical insurance in 2013.(4) The excess cost statistics results: In our total 10963 patients, there were 14 combinations had 435 excess cases with the constituent ratio 3.97%. Their total hospital cost was 14308991.51 yuan, accounting for 10.46% of the total cases' total cost(136825283 yuan). The excess hospitalization cost was 3291593.21 yuan, accounting for 23.00% of all excess cases' total cost, and accounting for 2.41% of total cases' total cost.Conclusion: For the patients who attended Tianjin medical insurance, the more hospitalization days, the higher hospital level, older the patients were, the higher total hospitalization expenses. Data mining is a useful method to extract information from huge amounts of data. Combined with the thought of the DRGs, it could show its unique advantages. The rough set theory is suitable for predicting the hospitalization costs of pneumonia patients. The rules established by decision tree could ensure the rational utilization of medical insurance fund and provide a theoretical basis for perfecting the social medical insurance system.