Background and purpose: To analyze the data in which the correlation between observations are to be considered, a general method is using marginal model with repeated measures, yet there is another method called conditional model with random clusters. Âccording to the binary responses, the aim of the present study is to compare the efficiency of these two models in studying the risk factors affecting hypertension in Mazandaran province.
Materials and methods: This cross-sectional study is part of research studying the risk factors of the non-infectious diseases in Mazandaran province in 2007 with a sample size of 1000 cases. First, stratified sampling was employed and then random cluster sampling was used in each stratum. Ïn all cases, blood pressure was measured three times through standard methods. The obtained data were analyzed using test and fitting marginal and conditional logistic models.
Results: Ôf the total of 1000 (500 male and 500 female) cases, 38 percent had high blood pressure. Ïn both models four variables of age, physical activity, body-mass-index, and the consumption of fruits and vegetables were identified as factors affecting hypertension. Ïn addition, gender and fasting blood sugar were statistically significant only in marginal logistic model. Üsing goodness-of-fit criteria, values of all the statistics for conditional logistic model with random effects of individuals were lower in comparison with the marginal logistic model with repeated measures.
Çonclusion: Ït can be concluded that although one cannot decisively select a model as the most appropriate one, conditional logistic model fits were somewhat better than the marginal logistic model with repeated measures, so it can be used to analyze such data.

Keywords: Hypertension, marginal logistic model with repeated measures, conditional logistic model with random clusters, goodness-of-fit criteria

Full-Text [PDF 342 kb]   (4370 Downloads)