Association between Major Non-communicable Diseases Risk Factors and Fasting Blood Glucose in Iran: Comparison of Two Techniques, with and Without Dichotomizing the Response
Razieh Bidhendi Yarandi
Student Research Committee, Department of Biostatistics, University of Social Welfare and Rehabilitation Sciences, Tehran, Iran.
Mehdi Rahgozar
Department of Biostatistics, University of Social Welfare and Rehabilitation Sciences, Tehran, Iran.
Jalil Koohpayehzadeh
Head of NCD Surveillance Office, Center for Disease Control, Ministry of Health and Medical Education, Tehran, Iran.
Ali Rafei
Research assistant at Non-Communicable Diseases Risk Factors Surveillance office, Center for Diseases Control, Ministry of Health and Medical Education, Tehran, Iran.
Fereshteh Asgari
Center for Diseases Control, Ministry of Health and Medical Education, Tehran, Iran.
Enayatollah Bakhshi *
Department of Biostatistics, University of Social Welfare and Rehabilitation Sciences, Tehran, Iran.
*Author to whom correspondence should be addressed.
Abstract
Background: Dichotomizing a continuous outcome variable is a common approach to estimate the odds ratio (OR) as a measure of association. In the present study we aimed to compare a non-dichotomizing technique with logistic regression which exploits dichotomizing the response for estimating OR.
Method: Data including a total of 17,152 Iranian individuals aged 25–65 years were derived from the third national survey of non-communicable Diseases Risk Factors in Iran. To measure the associations between fasting blood glucose and attributed risk factors two distinct techniques were used. Using a non-dichotomizing technique, an approach proposed by B.K. Moser and L. Coombs (2004) was employed to estimate odds ratios and associated 95% confidence intervals (CIs); A binary logistic regression model was also applied to fit the data as a common dichotomizing approach. Finally the results of two methods were compared by use of relative efficiencies and relative length of CIs.
Results: The odds ratios provided by both approaches are approximately the same, but relative efficiencies and relative length of CIs are greater than 2 which reflected better results for the technique used a non-dichotomizing approach compared to Logistic Regression Model.
Conclusions: Dichotomizing continues outcome variable is not necessary to estimate ORs, especially when there is no pre-specified optimal cut-off point for the response variable.
Keywords: Non-dichotomizing, logistic regression model, odds ratio.