Study Of Tools For Diagnosing Smear Negative Pulmonary Tuberculosis

PART ISTUDY OF THE DEMOGRAPHIC AND CLINICAL FEATURES OF SMEAR NEGATIVE PULMONARY TUBERCULOSISObjective To explore the demographic and clinical features of smear-negative pulmonary tuberculosis(SNPT) and seek from major predictors identifying SNPT,in order to give some scientific foundation for diagnosing SNPT.Methods Using a case-control study,272 SNPT and 288 non-tuberculosis(non-TB) patients were investigated by means of a questionnaire on general conditions, anamnesis,history of present illness,findings of chest radiography and laboratory markers,which sum to 5 sects and 41 items.All data were analyzed by univariate statistic method. Significant variables were analyzed further by multiple logistic regression model to identify major predictors of SNPT.Results Through single factor analysis,twenty-three variables including alcoho -lism,married,non-scar of BCG,expectoration,night sweats,low fever,low fever aternoon, lack of strength,loss of appetite,weight loss,chest pain,cough chronicity,expectoration chronicity,night sweats chronicity,chest tight chronicity, abnormal HCT,abnormal ESR, abnormal sites of chest radiograph,the amounts of abnormal lung field,pleural adhesions, pleural effusion and pulmonary cavity were closely related to SNPT.By entering these factors into stepwise logistic regression analysis,six variables including weight loss,night sweats,lack of strength,chest tight,abnormal sites of chest radiograph and pleural adhesions entered into the final multiple logistic regression model.The six factors were major predictors of SNPT.Conclusion There are more male patients than female patients in SNPT.The proportion for rural patients is higher than that of urban patients.Typical clinical manifestation of pulmonary tuberculosis is absent in SNPT and their appearances on chest radiography have poor specificity.Patients with SNPT are less likely to have cavities on the chest radiograph.Night sweats,weight loss,lack of strength,pleural adhesions.abnormal manifestation of upper lung field and upper or middle lung field are important predictors of SNPT. PART IISTUDY OF A SCORING SYSTEM FOR DIAGNOSING SMEAR NEGATIVE PULMONARY TUBERCULOSISObjective To develop a scoring system for diagnosing SNPT,and to evaluate its performance in the diagnosis of SNPT,which to help doctors identify TB from the patients of sputum smear negative and control TB transmission.Methods 560 subjects were randomly divided into modeling sample and testing sample on the propration of 4:1.Using the modeling sample and based on the β-coefficients derived from the major predictors in our logistic regression model of our first part,we developed a scoring system.The testing sample was used to evaluate the diagnostic performances of the scoring system.Results The scoring system ranged from -3 to 13.The patient under 4 scores was more likely non-SNPT than SNPT,while the patient with 4 scores and over had very high probability of SNPT. When the system was applied to the modeling sample and testing sample,accuracy,sensitivity,specificity and the area under the receiver operator character -istic curve were 88.4% and 86.6%,91.3% and 84.8%,86.4% and 89.1%,0.888±0.017 and 0.870 ±0.037,respectively.Conclusion The scoring system used in diagnosing SNPT has higher sensitivity and specificity.And it is easy to be operated.So the scoring system can be used as a tool for the diagnosis of SNPT and deserve further investigation. PART III STUDY ON THE MODEL OF THE DIAGNOSIS OF SMEAR NEGATIVE PULMONARY TUBERCULOSIS BASED ON ARTIFICAL NEURAL NETWORKObjective To establish a model based on artificial neural network(ANN) in the diagnosis of SNPT and to evaluate its performance.Methods The modeling sample and the testing sample of the second part were used as training sample and checking sample of the third part.The training sample was used to screen out significant single parameters and to develop the diagnostic model of SNPT based on ANN.Training sample used the study system with the supervision of a tutor.After the ANN model obtained discriminant function by training or studying.it judged what category the unknown sample belonged to.The checking sample was used to check the reliable degree of the discriminant function so that we selected the structure of ANN and decided the end point.The area under the receiver operator characteristic curve, sensitivity, and specificity were used to evaluate the model's performance in both training and testing samples.Results The structure of artificial neural network is (29-9-1)-BP. When the model was applied to the training sample and the checking sample,the area under the receiver operator characteristic curve,accuracy, sensitivity and specificity were 0.984±0.004 and 0.955±0.018,91.07% and 93.10%,90.91% and 88.89%,91.30% and 100%,respectively.Conclusion Artificial neural network model used in diagnosing smear negative pulmonary tuberculosis has better diagnostic capability,which can be used as a tool for the diagnosis of smear negative pulmonary tuberculosis and deserve further investigation. PART IV STUDY ON EVALUATING VALIDITY OF SCORING SYSTEM AND ANN MODELObjective The validating sample was used to evaluate and compare with their validities of the scoring system and the ANN model in diagnosis of SNPT.Methods Using the scoring system and the ANN model.the validating sample was diagnosed.The sensitivity,specitivity and some related indexes were computered respectively.We developed ROCs of the scoring system and the ANN model and compared the areas under their ROCs by paired t test.Results When the scoring system and the ANN model were applied to the validating sample,sensitivity and specificity were 84.21% and 88.89%,80%and 100%,respectively;Youden's index,positive likelihood ratio and agreement rate were 0.64 and 0.89,4.21 and ∝,82.76% and 93.10%,respectively;Negative likelihood ratio, misdiagnosis rate and missing diagnosis rate were 0.197 and 0.105,17.24% and 6.90%,15.79% and 10.53%,respectively.The areas under the receiver operator characteristic curve were 0.926±0.049 and 0.989±0.015,respectively.By paired t test,the areas between scoring system and ANN model were unsignificant(.P>0.05).Conclusion In term of the patients for suspicion of TB,both scoring system and ANN model have higher diagnostic value.Compared with traditional methods,both scoring system and ANN model are easy to operate and non-invasive.So they are likely to become new methods in diagnosis of SNPT.