Abstract
BACKGROUND: Dichotomizing a continuous outcome, sleep quality score, is associated with loss of information, and bias. Here we compared the performance of dichotomizing sleep quality scores according to cut-point as outcome variable in multivariable logistic regression with using that as continuous outcome in an adjusted distributional method. METHODS: In this study, the data from the second phase of the Shahroud eye cohort study (ShECS) on 4710 adults were used. Sleep quality score using Pittsburg index was normalized using item response theory (IRT) method. Sleep quality score was used as a dichotomized variable in a logistic regression model and as a continuous variable in an adjusted distributional method. RESULTS: The overall prevalence of poor sleep quality was 44.9 % (95 %CI, 43.4-46.2). In the adjusted distributional model poor sleep quality was associated with female gender (OR = 2.08; 95 %CI: 1.9-2.2), old age (OR = 1.2; 95 %CI: 1.0-1.4), low economic status (OR = 1.3; 95 % CI: 1.1-1.4), Illiteracy (OR = 1.4; 95 %CI: 1.2-1.7), diabetes (OR = 1.1; 95 %CI: 1.0-1.2), hypertension (OR = 1.2; 95 %CI: 1.0-1.3) and tobacco smoking (OR = 1.2; 95 %CI: 1.1-1.4). There was no difference between the size and direction of the observed association between two competing models. The confidence interval of the ORs and the marginal differences in proportions of poor sleep quality for the diabetic and non-diabetic people using the distributional method was more precise (narrower confidence interval) than logistic regression. CONCLUSION: Using the adjusted distributional method based on linear regression instead of dichotomizing the continuous outcome in logistic regression leads to narrower and more precise CIs for ORs but size and direction of associations between two models are identical. Comparison between two models showed that statistical performance of two models is equals. In addition to increasing age, women have higher odds of poor sleep quality than men. Some other modifiable predictors such as smoking, diabetes, and hypertension can be investigated to improve sleep quality.