Study On Models And Prediction Of Nosocomial Infection In Neurosurgery Operation Patients Of Large General Hospital

Nosocomial infections are prominent in today’s medical safety issues. In the event ofnosocomial infection, controlling it is relatively difficult. It affects prognosis and outcomeand increases medical costs and brings enormous economic burden to patients and society.What’s more, it can lead to serious disability or death of the patient. Therefore, it’sessential to predict hospital infection and to strengthen the prevention and control ofnosocomial infection. In this paper, it builds mathematical models to predict hospitalinfection, in the sense that, each patient initiative to detect and prevent infection fromoccurring, rather than be passive hospital infection after treatment given to patients. Sothe "gateway" of hospital infection prevention and control will be advanced, and it willreduce the incidence of nosocomial infections fundamentally. It is important to preventand diagnosis nosocomial infection, and prevents outbreak of nosocomial infection.Based on real time-nosocomial infection service system (RT-NISS), Relationalinformation about patients in neurosurgery is collected during period from January in2010to August in2013. Patients as case group, Non-infected patient as control group. Theanalysis of risk factors for these patients from two aspects: on the one hand, the number ofoperations is considered. On the other hand, the patients are divided into only once surgicalpatients, twice surgical patients and three times surgical patients according to the effectiveoperation. First, it obtains statistically significant indicator variables from single-factoranalysis. Then these variables are put into multivariate analysis. And the potential riskfactors about nosocomial infections with patients in neurosurgical ultimately are obtained.Logistic regression model is constructed in software of SAS9.1. The process of rough setattribute reduction is achieved in Software of Rosetta Johnson reduction algorithm. SVMmodel and BP neural network model are built in Xi’an Jiao tong University Merrill Lynchdata mining platform. The effect of predict of the three models are evaluated with the areaunder the ROC curve and the pros and cons with it is compared. It collects operation patient’s information during the period from January in2010toAugust in2013in neurosurgery. A total of cases is10546, including666cases ofnosocomial infection,831cases of times. The rate of nosocomial infection is6.32%. Thenosocomial infection case prevalence rate is7.88%. The site of infection in the centralnervous system ranked first (49.22%), followed by respiratory system (23.83%). Thehighest of pathogen is53.44%(Gram-negative bacteria), followed by23.99%(Gram-positive bacteria), and the proportion of fungi is22.57%. The top three surgicaldiseases by number of patients in neurosurgery resident are cerebral vascular disease,3244cases (30.76%), intracranial tumors,2803cases (26.58%), canal tumor,785cases(7.44%). The top four infection ratio diseases are brain injury disease (12.15%),hydrocephalus (11.00%), congenital brain (9.02%), intracranial tumors (8.49%).The correlation factors are classified as patient-self factors and iatrogenic factors. Andthe iatrogenic factors are divided into two parts according to infection time. The indicatorsin patients with nosocomial infection before infection with uninfected patients docase-control study. The risk factors of nosocomial infection are analyzed. Multivariatelogistic regression analysis shows that, considering the number of operation, the importantinfluence factors included gender, tracranial tumor, congenital disease of brain,hydrocephalus, pain and functional diseases in neurosurgery, in ICU and in ICU seven days,urinary catheter, ventilator, make use of tube feeding. To the patients with one operation,the important influence factors include gender, congenital disease of brain, hydrocephalus,in ICU and in ICU seven days, the use of urinary catheters, ventilator, nasal feeding,craniotomy operation, anesthesia operation, class II incision, emergency operation,operation when the length of not less than three hours. To the patients with two operation,the important influence factors include age greater than59years old, congenital disease ofbrain, length of hospital stay more than7days, central venous catheterization, urinarycatheter, ventilator, the first operation for interventional diagnosis, the first operation whenthe length of not less than three hours, second times of craniotomy operation class IIincision operation, second times, second of emergency operation.The samples are divided into training and testing ones by the time. Data about patientsfrom January in2010to December in2012as the training samples while data about patients January to August in2013as test samples. Logistic regression model, rough setattribute reduction and SVM model and BP neural network model are built to train thesamples.Considering the number of operations, results of three predicted models are as follows:the number of0.480is selected as the threshold in logistic regression model. Theaccuracy rate of model test is74.40%while the accuracy rate of prediction is71.48%.The sensitivity of it is0.9567and specificity is0.6769. The area of ROC is0.834. Inmodel of rough set attribute reduction and SVM, the accuracy rate of testing is85.99%while the accuracy rate of prediction is73.71%. The sensitivity of it is0.7829andspecificity is0.7333. The area of ROC is0.914. In model of BP neural network, theaccuracy rate of testing is86.80%. While the accuracy rate of prediction is84.15%. Thesensitivity of it is0.6667and specificity is0.8559. The area of ROC is0.968.For one operation patients, the accuracy testing of three models are74.30%,81.06%,88.42%. The accuracy rate of prediction test samples are75.68%,66.92%,82.40%,respectively. The sensitivity is0.8381,0.8476,0.7048, respectively. The specificity is0.7506,0.8317,0.8332, respectively. The areas of ROC are0.832,0.906,0.989,respectively.For two times surgery patients, the accuracy testing of logistic regression model, roughset attribute reduction and SVM model, BP neural network model are89.00%,84.89%,94.78%, respectively. The accuracy rate of prediction test samples are88.34%,82.51%,90.58%, respectively. The sensitivity are0.7619,0.7619,0.381, respectively. Thespecificity are0.896,0.8317,0.9604, respectively. The areas of ROC are0.954,0.934,1.000, respectively.The ability of the three models to distinguish is good. The effect of BP neural networkmodel is superior to the other two models. The sensitivity of logistic regression model isthe highest. Considering sensitivity and specificity, rough set attribute reduction and SVMmodel to solve such problems is relatively the best.In this paper, a variety of mathematical models are used to study factors influencingnosocomial infection in neurosurgery. It is the first time to use hierarchical analysis method(according to the number of degrees surgical patient classification) in the country. And it is the first time to use the method of rough set attribute reduction SVM model to predictnosocomial infection. The effect of Prediction of rough set attribute reduction SVM modelis better.