Clinical Study Of Conditional Logistic Regression And Data Mining On Risk Factors Of Nonsyndromic Cleft Lip And Palate

Objective: To discuss the main risk factors of nonsyndromic cleft lip and palate (NSCL/P),to evaluate the relative importance among these risk factors in occurrence of NSCL/P, to probe the protecting factors and to propose the reasonable predictive model for aristogenesis net.Methods: A hospital-based 1:1 matched case-control study design was applied to our epidemiological study.76 nonsyndromic cleft lip and palate children, the age in the 0～12 year-old were selected from some hospitals in Shandong province from Sep. 2006 to Sep. 2007 as collecting cases, and control cases were chosen from the same section or by same living area with the matched patients. A self-administered questionnaire prepared according to risk factors of NSCL/P was used to collect information, that is, the patients or controls of NSCL/P were asked some questions about these risk factors and the answers were filled in the questionnaire by investigators. After the investigation, these data were cleaned up by pair and input into computer to establish data base in Excel 2003. Firstly, to quantify all investigation items reasonably based on the test of the reliability and validity of the questionnaire. Secondly, to carry on the multivariate stepwise regression analysis in univariate condition Logistic regression analysis foundation, to discover the risk factors from multitudinous possible influencing factor and establish the 'best' model. Finally, to establish the classification tree and the LogitBoost predictive models separately according to the risk factors, to appraise the two models using the ROC curve, to establish the reasonable predictive model of NSCL/P. The entire statistical analysis process mainly used statistical software SAS 8.0, SPSS13.0 and WEKA which is developed by the New Zealand university.Results: 76 matched case-control groups were investigated and information was collected, a=0.1 to the test standard, by univariate conditional Logistic regression analysis, 23 risk factors were found with statistical significance. After the data relevant analysis, colinearity diagnosis and so on 14 factors entered multivariate conditional stepwise Logistic regression analysis according to sle=0.10, sls=0.15; sle=0.15, sls=0.15; sle=0.15, sls=0.20; sle=0.20, sls=0.20 respectively, finally we obtained 6 risk factors with statistical significance. In order to study the interaction of the 6 variables, we carried on partial correlation analysis and established regression model, we found that there was no statistical significance to integrate speciality. After the test of agreement, the optimal model was obtained according to sle=0.10, sls=0.15. There were 6 risk factors in the model which were related to NSCL/P, of the total, X8 (birth order, OR=4.136, P=0.004),X11 (genetic family history, OR=7.348, P=0.009), X12 (maternal disorder diet, OR=7.995, P=0.007),X19 (maternal stress during the first trimester of pregnancy, OR=6.371, P=0.000),X30 (maternal infectious history, OR=8.061, P=0.010) were significantly correlated with NSCL/P, 5 of them had significantly positive correlation to NSCL/P. X4 (fathers' education level, OR=0.040, P=0.000) was significantly negatively correlated with NSCL/P. After the evaluation with ROC curve, to establish the classification tree model which was used to predict the morbidity of NSCL/P.Conclusions: 5 factors including genetic family history, maternal disorder diet, maternal stress during the first trimester of pregnancy, maternal infectious history and birth order were promoting factors of NSCL/P, while fathers' education level was a protecting factor. This study emphasized on the need to educate the public about the importance of eugenics counseling and to popular the health care knowledge during pregnancy. Primary prevention should be done to the family which had family medical history and high onset risk. Furthermore, we should carry on tracking follow-up visit and early intervention in order to lower incidence rate of NSCL/P. According to the investigation, we established the classification tree model as the predictive model of the morbidity of NSCL/P.