Comparison logistic regression and discriminant analysis in identifying the determinants of type 2 diabetes among prediabetes of Kermanshah rural areas

authors:

avatar Eghbal Zandkarimi 1 , avatar Alireza Afshari Safavi 1 , * , avatar Mansour Rezaei 1 , avatar Ghazban Rajabi 2

Dept. of Biostatistics and Epidemiology, School of Health, Kermanshah University of Medical Sciences, Kermanshah, Iran
Disease Prevention and Control Unit, Kermanshah Health Center, Kermanshah University of Medical Sciences, Kermanshah, Iran
Journal of Kermanshah University of Medical Sciences: Vol.17, issue 5; e77059
published online: August 29, 2013
article type: Research Article
received: January 14, 2013
accepted: June 11, 2013

how to cite: Zandkarimi E, Afshari Safavi A, Rezaei M, Rajabi G. Comparison logistic regression and discriminant analysis in identifying the determinants of type 2 diabetes among prediabetes of Kermanshah rural areas. J Kermanshah Univ Med Sci. 2013;17(5):e77059. 

Abstract

Background: Failure to control diabetes on time leads to irreparable complications in other parts of body such as heart, kidneys, eyes, and etc. As those most susceptible to the disease are pre-diabetic individuals, this study aims to identify the Determinants for diabetes in people with pre-diabetes employing two advanced statistical methods of logistic regression and discriminant analysis.
Methods: Data were collected from 17 health centers in the rural area of Kermanshah city. 100 diabetic patients and 100 pre-diabetes persons (controls) were enrolled. Demographic data, body mass index, fasting blood sugar (FBS), oral glucose tolerance test (OGTT), blood pressure, blood lipid levels and daily activity in two separate forms by the staff of Disease control from health records were collected. Logistic regression and discriminant analysis to identify Determinants and ROC curve were used to compare the predictive power of the models.
Results: The predictive power of logistic regression and discriminant analysis were 0.884 and 0.80 respectively. Sex (P= 0.027) and FBS (P<0.001) in logistic regression and age (P=0.014), FBS (P<0.001) and OGTT (P<0.001) were significant in the discriminant analysis. Logistic regression model was more sensitive (79%).
Conclusion: In this study, logistic regression was more powerful in the separation of patients from pre-diabetic. In communities that have high affinity between case and control groups, identifying differences needs stronger methods. Thus, using these methods recommended in medical studies.

Fulltext

References

  • 1.

    The references of this article is available on PDF.