Introduction: The analysis of classified data in statistics and medical sciences is very important. If the binary response variable is misclassified, the results of fitting the model, will be skewed with false interpretation. The aim of this study was the application of logistic regression with misclassified variables in diabetes data.
Methods: In this descriptive study, data from 819 participants in the diabetes screening program at Zahedan Health Center in 2014 were used. Type 2 diabetes was studied in two ways. At first, by testing normal blood glucose (without fasting), the lack of correlation between type 2 diabetes and blood pressure was determined by logistic regression and odds ratios, and then a fasting blood glucose test was used for validation, which revealed significant results. False classification was considered based on the level of blood glucose due to low sensitivity and specificity of blood glucose test. To correct the classification error, the likelihood ratio method was used to estimate the coefficients. Data analysis was done using the software SAS version 9.1.3 and procedure NLMIXED with a significance level of 0.05.
Results: The correlation coefficient changed the odds ratio of diabetes in the blood pressure variable from 0.227 to 1.20, significantly (P < 0.001). In addition, other model variables were modified.
Conclusions: Logistic regression for data with error classification can be used as a suitable method for analyzing data with classification error. Validation using logistic regression for classification error data showed that high blood pressure has a significant effect on diabetes. It is suggested that the logistic regression method should be used in order to correct the odds ratio in view of the probability of classification error in the screening data.