The Application Of Generalize Additive Models In Medical Cost Control

Objective:Based on the data about the expense constitution collected from Cataract patients in Tianjin from 2003 to 2007, their medical expenditures will be compared between different Patients. The Generalize Additive Models (GAM) will be used to analyze the factors influencing hospitalization costs and then its results will be compared with the ones from the traditional multiple linear regressions. In this way, the advantages and feasibility of GAM can be further elaborated so that it can be used on various aspects such as a reference on related research, and also a guidance to improve the medical service efficiency as well as to reduce the burden of the Patients.Methods:The all information of Cataract patients were collected in Tianjin hospitals from January 1st,2003 to January 1st,2007. Selecting and sorting the information of the hospitalized patients. Analyze the factors influencing hospitalization costs by multiple linear regressions with SPSS 16.0. The data ware analyzed by using Professional statistical software S-PLUS6.2 and SAS9.2, and compared the results between with the traditional methods.Results:Materials cost took the highest proportion, more than 40%, the followings were the cost of treatment, inspection and drug. The costs of the above accounted for about 90%. Grey Relational Analysis found the correlation of materials cost was the first, followed by surgery. The drug cost, inspection cost, treatment cost and bed costs of the Patients with complications were more than without. The differences of the total costs, drug costs, inspection costs, treatment costs, operation costs, bed costs, materials and other cost were statistically significant (P＜0.05) in various hospitals levels. The 3 A hospitals were the highest. By multi-factor analysis, impact factors of hospitalization cost are disclosed as follows:days of hospitalization(β=0.726), rank of hospitalization(β=0.181), the insure year2005VS2003(β=0.055), the insure year2006VS2003(β=0.051), the insure year2007VS2003(β=0.056), the proportion of material costs (β=0.427). GAM found the factors influencing hospitalization costs were the same as traditional multiple linear regressions, apart from the age and occupation. The structure of a smooth map of GAM found independent variables (days of hospitalization, the proportion of material costs, age) with outcome variables (total hospitalization costs) was not a linear relationship. The model fit test found that two models are different. The AIC and SBC of the GAM is lower than that of multiple linear regressions. Clearly, GAM is more appropriate than traditional multiple linear regressions to handle this data.Conclusion:By using GAM analyze the impact factors of 5030 Cataract patients' hospitalization costs in hospitals, we not only found out the meaningful factors, but also the conclusion can provide a certain valuable references to the government departments who formulates relevant Policies of medicate and to the medical service units who should strengthen their management, improving economic efficiency and reduce the burden on Patients to achieve a multi-win-win situation in future. On the other hand, if the data was not normality, homogeneity of variance and the relationship of independent variables and outcome variables was not a linear; the result will deviate from the actual results, and even lead to results unreliable by using multiple linear regressions. But the GAM can be to solve such problems. This study found that screening the factors influencing hospitalization costs existed differences by GAM and multiple linear regressions. Not only it found the age was the influencing factor, but also the structure of a smooth map of GAM provided richer information. It prompts that in the future research we can try to apply this method to deal with similar problems.