Title : Application of principal component analysis and ordered logit model in diabetic kidney disease progression in people with type 2 diabetes
Abstract:
Diabetic kidney disease is one of the main microvascular complications caused by diabetes. Several clinical and biochemical variables are reported to be associated with diabetic kidney disease in people with type 2 diabetes. However, the interrelations among these variables could potentially distort the estimation of their effects on the progression of the disease. The objective of the study is to determine how the biochemical and clinical variables in people with type 2 diabetes are intercorrelated with each other and their effects on the progression of kidney diseases using advanced statistical methods. Principal component analysis, combined with ordered logit models, is used to explore the interrelationships between biochemical and clinical variables and their effect on the progression of kidney disease. This retrospective cross-sectional study retrieved data from 323 diabetic individuals in a polyclinic hospital at the University of Messina, Italy. The study identified three uncorrelated principal components. The first component, a linear combination of positively correlated glycosylated haemoglobin, glycemia, and creatinine, having a strong significant effect on kidney disease progression. Principal component two is a linear combination of positively correlated total cholesterol and low-density lipoprotein, while Principal component three is a linear combination of negatively correlated high-density lipoprotein and triglycerides. The cumulative odds, adjacent category, and continuation ratio models further revealed that age, sex, body mass index, and metformin treatment have significant effects on the progression of kidney disease. However, these effects are not proportional across the stage of kidney disease progression. Flexible ordered logit models, including partial, cumulative odds, adjacent category, and continuation ratio models, were used to address proportionality issues and enhance the accuracy of effect estimation. The study findings conclude that the partial, cumulative odds, adjacent category, and continuation ratio models are a robust technique for estimating effects, specifically when predictors have different effects at different stages of disease progression.
