Dear colleagues,

I have a study data from a sample population that contains three different subsets of study subjects: 1. Healthy control with normal BMI, 2. Non-diabetic overweight/obese subjects and 3. Type 2 diabetic subjects.

I am trying to explore whether plasma lipid parameters, their ratios or indices could potentially predict the incident metabolic syndrome in the given sample population (N=142). For this, I want to develop a linear regression model using metabolic syndrome components (expressed in number as 0, 1, 2, 3, 4,and 5) as the dependent variable and all plasma lipids, their ratios and indices as the independent variables. My question is how should I develop this regression model? Should I develop the model using the data of whole study population that contains all three subsets of subjects mentioned above or first split the whole study data into two groups as " with metabolic syndrome" and " without metabolic syndrome" apply linear regression and report the beta coefficient values of only the subset of " with metabolic syndrome"?

Please also advise me whether there could another regression method that could best apply to my case. Thank you.