Discrepancy of Obesity Prevalence and the Effect Sizes on Chronic Diseases depending on assessing tool: Self-reported vs. Measured Weight and Height

Abstract

While there are strong correlations between self-reported and directly measured anthropometric data, the discrepancy and systematic errors associated with these, particularly among middle-aged and older persons residing in South Korea, remain a contentious issue. All participants were selected from the Korean Longitudinal Study of Ageing (KLoSA), a panel study conducted by the Korea Labor Institute

data from 510 participants (290 females

56.9%) were analyzed. We considered general characteristics, including sex, age, education, marital status, employment, income, and residential region, and used self-rated health as a generic indicator of health status. On the first subject, accuracy of obesity classification from self-reported data, one-way ANOVA, t-test, and Scheffes test (α = 0.1) were employed to explore the difference between directly measured and self-reported values. Sensitivity and i specificity values were used to assess the validity of obesity diagnoses based on self-reported BMI. The means of BMI differences were 1.3 (±1.2) kg/m2 among men and 1.8 (±1.5) kg/m2 among women. In men, the difference could be attributed to measured BMI and residential region

among women, age and education level influenced the discrepancy in BMI. Scheffes test (α = 0.1) for multiple comparisons of group means revealed that women over the age of 65 years, with relatively low middle-school education, who lived in rural areas, and had a measured BMI of 25 kg/m2 or more, were significantly more likely to have BMI discrepancies. In contrast, for men, significant predictors were living in rural areas and being obese. Although adequate correlations were seen in self-reported BMI, they indicated low sensitivity, with 46.5% and 60.1% among males and females, respectively. However, specificities were very high, at 97.8% and 98.0% for males and females, respectively. On the second subject, different measure of obesity leading to relevance change, χ² was used for differences in obesity prevalence diagnosed by BMI calculated from self-reported height and weight by categories in each variable. Multiple logistic regression for obesity in the case of using reported BMI and measured BMI was conducted to define what variables have correlation with obesity prevalence by sex. Categories that showed higher prevalence of obesity in employment, income, and region groups were men who are employed (23.7%), receive top-quarter income (30.9%), and live in smaller cities (18.6%), respectively in men. The overall prevalence of obesity had risen in each variables, indicating the incidence of the ii under-reporting. The measured obesity showed no difference in highest prevalent category, however, overall prevalence of obesity was significantly higher than reported obesity in women. In men, age was a significant factor identifying the discrepancies between reported and measured obesity. Reported obesity of men aged 45-54 was 29.8% whereas measured obesity of those was 57.5%. In contrast, age of the women showed no significant trend or differences in reported and measured obesity. In men, level of education reported no differences on report and measured obesity, however, women showed significant difference between level of education and obesity prevalence. Regarding relevancy, after adjusting age, sex, and other socioeconomic status, self-rated health group presented statistically significance in moderate health status group than bad health status group in all. Analyzing after age and socioeconomic status were adjusted with sexes, in men, each variable was difficult to identify the statistical trend. However, in women, obesity prevalence was reported higher as the age decreased, reported obesity was not significant but measured obesity was. For the study of the change in association from using different measurements, we made model with obesity and chronic conditions

diabetes, hypercholesterolemia and hypertension. The prevalence of diabetes was 15.9% in both genders. There were significant trend between diabetes prevalence in self-rated health, working status, residence area, and measured obesity in all population. The prevalence of hypercholesterolemia was just 12.6%, 64 cases. There were no significant variables for chai square test. The prevalence of hypertension were 51.8% in men and 51.0% iii in women. The level of obesity from both method and residence region were significant in chai square test. As results of multiple logistic regression to assess the association between obesity and hypertension, compared obesity from reported with measured BMI, odds ratios were considerably changed by BMI measuring method and sex. For all, both men and women, odds ratio of measured obesity was higher than reported. Considered separately by sex, odds ratio of reported obesity was relatively high in both sexes. Finally, the evaluation of effect of obesity from self-reported BMI on chronic conditions, is relevant for estimation of health risk itself and its trend depending upon obese level, compared with effect of obesity from measured BMI. But, for estimation to the magnitude of the risks, we have to be concerned about not only under-estimation but also over-estimation. This study population was just 510 persons. Even though self-reporting error comes from multi-directional and diverse elements, all possible covariates could not be included in the analyzing model because of sample size. Therefore, the results of this study has potential instability. A larger study based in community is needed to explain a more precise estimate of effect of reporting error in obesity classification.