Risk Adjustment Model On Nosocomial Infection Ratio Of Methicillin-Resistant Staphylococcus Aureus For Public Reporting

[Background]Methicillin-resistant staphylococcus aureus （MRSA） infection is one of the main nosocomial infections. It has become the important challenge of nosocomial infection prevention and control, and it is listed as the leading one of the three most harmful, expansive infectious diseases hard to controlled in the world. MRSA nosocomial infection has many influencing factors, and it is difficult to treat the infection because of wide drug-resistant spectrum.It not only increases the length of hospital stay of patients, and also increases the patients’ medical expenses burden. Nowadays, China has become one of the world’s worst affected countries by MRSA infection.Transparent regulation is a frontier management innovation, and is recognized and promoted widely as a means to improve the performance of hospital infection by many developed countries. In these countries, indexes of nosocomial infection （such as CLABSI and CAUTI） have been reported publicly. However, in order to avoid misleading conclusions, these indexes need to make risk adjustment before reporting publicly. Meanwhile, risk-adjusted indexes are also more conducive to track and reflect the progress of nosocomial infection prevention and control over time. As the benchmarking of prevention and control of nosocomial infection, the United States has mature practices of risk-adjustment and transparency. Although China has tried on the regulation of hospital infection, but always stays within the scope of monitoring in the hospital, and has not carried out transparency to compare its performance among hospitals.Therefore, the study will apply the method of evidence-based practice to build risk adjustment model for MRSA nosocomial infection according to the experience of American CDC. It aims to provide for China to implement the transparency of MRSA nosocomial infection in future. It is will be more reasonable and fair to reflect the performance of prevention and control of nosocomial infection in different hospitals in different time. Finally, it would help implementing the transparent regulation for MRSA nosocomial infection in China.[Methods]The study will incorporate in the literature research of high level and high quality by quality evaluation of literature, then define the relevant concepts, such as defining MRSA risk factors suitable for reporting publicly, and conclude a set of MRSA risk factors of nosocomial infection conforming to hospital information disclosure. On the basis of related requirements of risk adjustment factors and accessibility and integrity of information, the data acquisition scale will be designed and finished. Then the empirical study will be carried out by random sampling. Statistical methods of univariate analysis, logistic regression, neural network model （ANN） and decision tree models are applied to find key risk factors, then draw ROC curve and calculate C index to evaluate these results. Finally, the study will build the risk adjustment model for MRSA nosocomial infection.[Results]（1） Study on finding risk factors suitable for transparency1) Defining the scope of the risk factors suitable for transparency:demographic characteristic, medical utilization in a year, health information based on diagnosis, drug use prior to admission and self-reported health status.2) Quality evaluation of literature and the set of risk factors:31 literature researches are incorporated in the study, and their average quality score is 7.1. There are 63 influencing factors totally of MRSA nosocomial infection found in these researches. According to the scope of the risk factors suitable for transparency,24 factors are got into the study.（2）Build the risk adjustment model of MRSA nosocomial infection1) Nosocomial infection rate and detection rate of MRSA:503 inpatients are included in the study in total. Nosocomial infection rate of MRSA is 9.34%（47/503）, detection rate is 22.1% when they are admitted and is 65.4%（329/503） during their hospitalization. 2) Logistic regression for MRSA nosocomial infection:indwelling catheter （P=0.001）, WBC （P<0.001）, MRSA colonization （P<0.001） and hypertension （P=0.004） are get into the model, P value in Hosmer-Lemeshow test of goodness of fit is 0.229, so the model fits the experiment data quite well. R square value of Nagelkerke is o.59, and sensitivity, specificity and the overall accuracy are 5.3%,98.5%and 94.4% respectively. C index is 0.932. And the risk adjustment model of MRSA nosocomial infection is: Y is Predicted Probability, and X1, X2, X3 and X4 are indwelling catheter, WBC, MRSA colonization and hypertension respectively. The result of risk adjustment is:MRSA SIR= （O₁+O₂+...+On）/（Y₁+Y₂+...+Yn）（O:whether inpatients get MRSA nosocomial infection, and 1 is YES and 0 is NO） 3) ANN model for MRSA nosocomial infection:the model involves indwelling catheter, WBC, MRSA colonization, previous hospitalization and hypertension in analysis according to importance ranking. Its sensitivity, specificity and the overall accuracy are 46.8% ,97.8% and 93.0%, and C index is 0.931.4) Decision tree model for MRSA nosocomial infection:the model involves indwelling catheter, WBC and MRSA colonization in analysis according to correlation. Its sensitivity, specificity and the overall accuracy are 2.3%,96.5% and 94.2%, and C index is 0.917.5) Discussion on building the risk adjustment model of MRSA nosocomial infection: when building the ANN model, definition of number of network hidden layer and its units is short of theoretical direction and mainly relies on experience. The model cannot explain signification of its variables well and does not also have memory for weights and threshold of the model. So if learning sample is changed, the model should be rebuilt. And decision tree model has similar problems with ANN model. But logistic regression model can be built easily and is highly accessible. It can explain variables and results best and can be promoted to use profitably because of excellent stability.[Conclusion]The study finds that MRSA colonization, indwelling catheter, WBC, hypertension and previous hospitalization （the factor is found only in ANN model） are key risk factors of which are suitable for reporting publicly. By comparing the three models, logistic regression is the best choice to building the risk adjustment model of MRSA nosocomial infection.