Difference analysis of results of disease risk assessment by using binary Logistic regression under different analysis strategies / 中华检验医学杂志

ABSTRACT

Objective To investigate the importance of conditional control on the results of binary Logistic regression analysis.Methods 664 male patients diagnosed with CHD and 400 healthy controls who visited the Department of Cardiology at People's Hospital of Yuxi City in Yunnan province from October 2010 to March 2013were enrolled consecutively in this case-control study, and 14 physiological and biochemical indexes[includingage,UA,TC,TG,HDL-C,LDL-C,APOA1,APOB100,Lp(a),HCY, TBIL,DBIL,IBIL,γ-GT]were collected.The correlations and differences in mathematical model results between physiological,biochemical indexes and CHD were compared by binary Logistic regression analysis under different statistical analysis strategies and then the model was analyzed by the ROC curve.Results (1)With no conditional control,all of the 14 physiological and biochemical indexes were directly inputted in the binary Logistic regression analysis,and after 11 steps regression analysis,11 indexes[age,UA,TG, HDL-C,LDL-C,APOA1,APOB100, Lp(a), HCY, DBIL, γ-GT]were retained.Because of the self-correcting ability of the binary Logistic regression analysis, relatively satisfactory results can be obtained. However,someof the resultsarehard to explain.(2)According to the application conditions of Logistic regression analysis,after performing the normality test, difference analysis and correlation analysis of these indexes, using condition analysis and control, considering the distribution characteristics of the indexes, excluding internal confounding factors among variables,9 more independent indexes[age,UA,TC,lnTG, lnLp(a),lnHCY,HDL-C,LDL-C and TBIL]were selected for Logistic regression analysis.After 7 steps regression analysis,7 indexes[age, UA, HDL-C, lnTG, lnLp(a), lnHCY and lnTBIL]were finally selected,and the area under the curve is 0.927.Conclusion When the binary Logistic regression analysis was used to confirm risk factors and establish risk assessment models for complex diseases, it is better to adopt strict conditional control to improve the reliability and validity of the analysis.