OBJECTIVETo evaluate the value of a back propagation (BP) network in determining the risk factors of uterine myomas.

METHODSUsing stratified randomized sampling method, 1260 women were surveyed by questionnaire. 1:2 matched case-control study was conducted in 113 cases of uterine myomas. Neural network tools box of Software MATLAB 6.5 was used to train and simulate BP artificial network. The mean impact value (MIV) for each input variables was analyzed, and was compared with multiple logistic regression analysis and log-linear model for interaction between factors.

RESULTSBP artificial neural analysis showed that the leading risk factors for uterine myomas were delayed menstruation, family history of uterine myomas, cervicitis, menstrual disorder, induced abortion, pelvic inflammatory, oral contraceptive medication, and elytritis, with mean impact value -0.0405, 0.0361, 0.0162, 0.0143, 0.0135, 0.0117, 0.0094, 0.0087, respectively. Both BP artificial neural and logistic regression analysis showed that the sequence of leading risk factors were similar in the whole, but there were some differences observed, induced abortion was proved to be an important cooperation variable through logline model analysis respectively.

CONCLUSIONCompared to the conventional statistics method, BP artificial neural network could deal with the interaction between covariables preferably, thus provided a powerful method to risk factor analysis.