Multilevel Models Applications To The Analysis Of Hospitalization Expense For Cholelithiasis

Objective:Through the collation of health care costs, to analyze the characters of the hierarchical structure data,and explore the use of multilevel modeling analysis of Medicare data, and evaluate the effect of the analysis of the multilevel model in health care costs. Then explore the factors that affect the cost of health insurance, and find better explanation of each factor caused by variations in the cost of health insurance combined with the application of multi-level model, and provide the reference and scientific basis for the research and evaluation of health care costs.Methods:The data came from a random sample of hospitalized patients with medical insurance from2003to2007in Tianjin city. Excluding the hospitalization days<1day, the total cost exceeds x+3s, and transferred patients. We collected valid cases4038cases included in the analysis. Apply the univariate analysis for the impact factors of the total cost of hospitalization, and use SAS9.2to establish a multilevel model for statistical modeling of health care costs and its influencing factors to evaluate multilevel model fitting results.Results:(1) In the4038gallstone patients, the number of male patients were1432, the number of female patients were2606, the mean age of patients was (57.94±12.44),the number of cases in the groups of age (18-45,46-60,61-94)were558,1739and1741,and the number of cases in the first grade, second grade and third grade hospitals were33,444and3561. There were75hospitals in the data, and16hospitals were first grade,39hospitals were second grade and20hospitals were third grade. The hospitalization cost of cholelithiasis patients presented growth trend from2003to2007(F=11.818,P=0.001); the result of single-factor analysis showed that age,admission days,the rank of hospital,frequency of admission,comorbidity,position of calculus.position of inflammation and operation are all the influencing factors of cost(P<0.01).(2) The Medicare cost data of gallstone patients presented the hierarchy, we calculated the Intro-Class Correlation Coefficient(ICC) was0.17,it means that there approximately were17%of the total variance in the outcome measures by patients hospitalized due to the difference. This shows the existence of a certain degree of homogeneity of the group data.(3)After adding the level-2explanatory variables into the empty model, The model estimated will converge quickly by six iterations, it means the model fits well. The-2LL significantly reducing(χ2=15.0,P=0.0001)means that the model with explanatory variables fits the data better than the null model. The information criteria measures(AIC,AICC and BIC) also reduced.(4) After adding the level-2explanatory variables into the model, the-2LL reduced from845.9to-2839.4(P<0.05), and the information criteria measures also reduced. This shows that adding the level-1of the explanatory variables into the model can significantly improve the model fit. σ2reduced from0.07129to0.02861and σ02reduced from0.007264to0.004205. This shows that the level-1explanatory variables could affect the outcome both at the individual level and at the group level-2variation.(5) After testing level1random slopes, the variance covariance of random effects overall test was statistically significant (χ2=319.28,df=3,P<0.05); σu02=0.004317(Z=2.86,P=0.0021) and σu12=0.000032(Z=2.4,P=0.0083) mean both the β0j and β1j were random coefficients.(6)After testing cross-level interactions, the interaction between yylb and zyts_C was non-significant(P<0.05).This means that the effect of zyts_C did not affected by yylb. Using ML method to estimate the model shows the model did not fit better after adding the interactions.(7) Finally, we determine the final model was a mix model with one level1random slope and non-cross-layer interactions.Conclusion:Medicare cost data has a hierarchical structure, and multilevel model can be flexible enough to handle the Medicare cost data in a hierarchical structure. It can effectively analyze the influencing factors of insured hospital costs of gallstone disease. Age, admission day, rank of hospital, frequency of admission, comorbidity, position of calculus, position of inflammation and operation are all the influencing factors of cost. Controlling the hospital costs could thought from the hospital level, length of stay and evaluation of hospital treatment capacity. The multilevel model can be applied to the identification, evaluation and control of health care costs.