| Trichloroethylene(Trichloroethylene,TCE),as an important organic chlorine solvent,is widely used in electronic,hardware and other industries.Workers occupationally exposed to TCE can develop Occupational Medicamentos-like Dermatitis Induced by Trichloroethylene(OMDT),with the typical clinical manifestations of fever,generalized rash,liver dysfunction and superficial lymphadenopathy.OMDT patients are characterized by urgent incidence,great harm,high mortality,few control methods and previous studies.At present,glucocorticoid,such as methylprednisolone,is considered the most effective drug in the treatment of OMDT,but there is no unified treatment plan.For the rational usage of glucocorticoid in the treatment of patients with OMDT,the analysis on glucocorticoid usage,influencing factors and estimation of glucocorticoid usage is conducted.ObjectivesTo establish a database of occupational medicamentos-like dermatitis induced by trichloroethylene(OMDT)from patients'electronic medical records and provide reliable data for epidemiological study,etiological analysis,clinical treatment,and prognosis.To quantitatively analyze the factors influencing the level of glucocorticoid usage in patients with OMDT and carry out posterior probability reasoning based on logistic regression and Bayesian network for providing evidences and supports for glucocorticoid therapeutic regimen in OMDT patients.MethodsThe structure of the database was determined according to related literature and research interests in OMDT.Electronic medical records of all OMDT patients who were diagnosed and treated in Guangdong Province Hospital for Occupational Disease Prevention from 1988-2013 were collected and transferred into EpiData 3.1 database in five sections(baseline,clinical manifestation,auxiliary examination,usage of glucocorticoid,and hospital expense).The values of mean,standard deviation,proportion were calculated for data description.T test or t' test were performed with normal distribution data for statistical analysis,and ?2 test was carried out to evaluate the difference of rates(P<0.05).With binary logistic regression,the analysis of the factors influencing the level of glucocorticoid usage(the initial dose,the maximum dose,the total dose)was performed,and some significant variables were selected for Bayesian networks based on the results of regression and expert knowledge in SPSS25.0.Based on Tree Augmented Naive Bayes and the algorithm of expectation-maximization,the Bayesian network is constructed in Netica.Then Netica was used to carry out sensitivity analysis for the degree of influence of candidate variables on interest variables and posterior probability reasoning with CPTs(conditional probability tables).ResultsA database with a total of 196 OMDT patients was established,which consisted of five parts(baseline,clinical manifestation,auxiliary examination,usage of glucocorticoid,and hospital expense).Logistic regression and Bayesian network were used to predict the hormone usage during the treatment of the disease.The results were as follows:1.Results of binary logistic regression analysis:(1)the positive significant factors of initial glucocorticoid dose(more than 60mg)included the clinical symptoms of the mild and severe liver injury(OR=9.246,P=0.017;OR=23.258,P<0.001),exfoliative dermatitis(OR=2.703,P=0.047),fever(OR=4.492,P=0.007),lymphadenectasis(OR=3.058,P=0.045),swelling of tonsils(OR=4.530,P=0.007),abdominal tenderness(OR=16.088,P=0.002),and auxiliary examination index of hemoglobin and phosphocreatine kinase.(2)the positive significant factors of maximum glucocorticoid dose(more than 110mg)included the clinical symptoms of exfoliative dermatitis(OR=3.525,P=0.016)and toxic epidermal necrolysis(OR=17.599,P=0.011),and auxiliary examination indexes of platelet count,a-hydroxybutyrate dehydrogenase,mean corpuscular volume,mean platelet volume,ALT,IgA and C-reactive protein.The negative significant factor of maximum glucocorticoid dose(more than 110mg)was auxiliary examination index of Na.(3)the positive significant factors of total glucocorticoid usage(more than 2900mg)included the days in hospital,the clinical symptoms of jaundice(OR=3.552,P=0.006)and infection(OR=2.669,P=0.031),and auxiliary examination indexes of hemoglobin,IgA,mean platelet volume,a-hydroxybutyrate dehydrogenase.The negative significant factor of total glucocorticoid usage(more than 2900mg)was auxiliary examination index of cholinesterase.2.Construction and analysis of Bayesian network models:based on expert knowledge and the results of logistic regression,the Bayesian network of initial glucocorticoid dose consisted of 9 nodes;the network of maximum glucocorticoid dose consisted of 11 nodes;the network of total glucocorticoid usage consisted of 11 nodes.The results of sensitivity analyses with built Bayesian networks suggested that the top three variables that had the greatest impacts on the initial glucocorticoid dose were liver injury,rash type,abdominal tenderness;the top three variables that had the greatest impacts on the maximum glucocorticoid dose were Na,C-reactive protein,mean on platelet volume;the top three variables that had the greatest impacts on the total glucocorticoid usage were days in hospital,mean platelet volume,a-hydroxybutyrate dehydrogenase.3.Application of Bayesian network models:the initial glucocorticoid dose was taken as an example to illustrate the application of Bayesian network model.Without anything about OMDT patient,the probabilities of initial glucocorticoid dose of?60mg and>60mg were 24.6%and 75.4%respectively.If the mild liver injury of the patient was confirmed and other conditions were uncertain,the probability of initial glucocorticoid dose ?60mg increased from 24.6%to 33.3%,and the probability of initial glucocorticoid dose>60mg decreased from 75.4%to 66.7%.If lymphadenectasis was confirmed with OMDT patient,the probability of initial glucocorticoid dose ?60mg decreased from 24.6%to 21.1%,and the probability of initial glucocorticoid dose>60mg increased from 75.4%to 78.9%.If the mild liver injury and lymphadenectasis were confirmed with OMDT patient,the probability of initial glucocorticoid dose?60mg increased from 24.6%to 28.7%,and the probability of initial glucocorticoid dose>60mg decreased from 75.4%to 71.3%.Discussion1.The established comprehensive OMDT database can provide reliable data support for epidemiological investigation of the disease,improvement of therapeutic regimens,and prediction of interest outcomes based on machine learning.2.For the level of glucocorticoid with OMDT patients,the top three variables that had the greatest impacts on the initial glucocorticoid dose were liver injury,rash type,abdominal tenderness;the top three variables that had the greatest impacts on the maximum glucocorticoid dose were Na,C-reactive protein,mean on platelet volume;the top three variables that had the greatest impacts on the total glucocorticoid usage were days in hospital,mean platelet volume,a-hydroxybutyrate dehydrogenase.3.Based on developed Bayesian networks,the posterior probability of the level of glucocorticoid is estimated to provide evidences and insights for the rational use of glucocorticoid in the treatment of patients with OMDT.4.Bayesian network is an effective way to explore influencing factors and perform probability reasoning with missing data which characterized by complex,uncertain and interrelated variables. |
PDF Full Text Request