Analysis And Development Of A Delirium Prediction Model For Intensive Care Patients

ObjectiveTo investigate ICU nurses delirium-related knowledge demands and the status of delirium prevention and management,identifying barriers for delirium risk assessment,prevention practice that should be addressed and other aspects,so as to provide reference for managers to develop training and learning programs.Then,develop a statistical model for delirium in adult intensive care patients,and validate this model based on data are readily available within 24 hours after intensive care admission.Expect to assist ICU professionals to assess the risk of delirium,providing a basis for the formulation of next treatment and care that may reduce the occurrence of ICU delirium.Methods1.We design a questionnaire including delirium knowledge,attitudes and practices three parts referring to the PAD guidelines,CAM-ICU training manuals and related literatures,and modify it according to the recommendation from nursing experts.We conduct a pilot survey to test the reliability and validity of this questionnaire before the formal investigation.After the correction,we applied to ICU nurses from three Grade A Class ? hospitals in Chongqing and Beijing during September 2016 and November 2016.2.The clinical variables were chosed refer to a number of domestic and foreign literature which searched in the databases such as China Science and Technology Journal Database(VIP),the Chinese Biomedical Literature Database(CBM),Pub Med and some other databases,with "delirium","intensive care unit","risk factors" as the search term.We collect patients information during the hospitalization period by searching electronic medical information system and paper recordsi n a general intensive care unit in a Grade A Class ? hospital in Chongqing from April 2016 to December 2016 by convenience sampling.The data was divided into the development queues(70% of the total)and the validation queues(30% of the total),and then model was established by the data from development queues.The factors that may affect occurrence of delirium were selected by single factor analysis for P<0.05,and then apply them to logistic regression analysis to identify the weight coefficient of each risk factor in the model,and develop the regression equation.The Hosmer-Lemeshow(H-L)chi-square was used to test the result of prediction model whether statistically significant different from the actual situation,and evaluate the distinguish capacity with the area under curve(AUC)which reflect the ability of distinguish delirium and non-delirious patients,the closer the value is to 1,the stronger the distinguishing ability of the model.ResultsOn delirium-related knowledge,155 nurses' score was 29.25±5.97,and the awareness of differences between delirium,dementia and depression only accounted for 3.2%;the score of attitude and behavior was 30.61±4.36 55.93±6.99 respectively,of which low satisfaction with physician-described delirium management and poor execution of delirium risk assessment were the most important barriers for implementation of a delirium guideline.The scores of nurses' delirium-related knowledge and attitude were statistically significantly different to the age,length of work,professional grade(P<0.05).Analyzing the overall level of nurses,knowledge and attitude,attitude and behavior,knowledge and behavior were positive correlation(P <0.01).Univariate analysis showed that urea nitrogen concentration,infection,GCS score,APACHE? score,emergency occupancy were associated with delirium in critically ill patients(P <0.05).Multiple stepwise regression showed that urea nitrogen was elevated,infection,disturbance of consciousness was independence predictors of delirium in ICU(P <0.05).No correlation was found between age,history of surgery,history of cognitive impairment,history of alcohol abuse,metabolic acidosis.ICU delirium prediction model: ea/(1+ea)× 100%,e is the exponential function,a = 0.865 × BUN level(= 1,2 or 3)+ 0.631 × GCS score(= 0,1,2 or 3)+ 0.873 × infection(= 0 or 1)-4.664.The results of the HL chi-square test of this model showed that there was no statistical significant difference between the predicted values of the delirium and the actual occurrence of a hypothesis test with significant differences(?2 = 9.104,P = 0.168).The area under the ROC curve was 0.749,P <0.05.When the threshold of the delirium model was 19.24%,the sensitivity was 0.834 and the specificity was 0.598.According to the distribution of the probability of delirium prediction and the best cut points,patients will be divided into three groups: low risk(? 20%),middle risk(20% ~ 40%),high risk(?40%),The incidence of delirium in the three groups was 10.4%,26/6%,45.5%,the incidence of delirium increased significantly with increased risk levels(P=0.022).The average number of days of ICU occupied in high risk group was higher than that in middle risk and low risk group(P = 0.029).ConclusionICU nurses generally deficiency of knowledge about delirium's differential diagnosis and clinical manifestations,while not familiar with some delirium risk factors and related clinical treatment guidelines,which in turn affects the pratice of risk assessment and the positivity to implement intervention targeted at prevent delirium.Therefore,managers should strengthen the education and training on the theme of delirious evidence-based knowledge,focusing on quality supervision of delirium prevention practice.In this study,infection,levels of urea nitrogen and disturbance of consciousness were independently associated with ICU delirium though multiple stepwise regression analysis,while we develop and validate a delirium prediction model for patients in general intensive care unit.According to the distribution of delirium probability of the verification cohort and the best cut point for risk stratification,which has certainly reference value for eliminating low risk groups of delirium.