The covariates for which one is able to control are necessarily limited, so we're left to speculate. I will speculate that if you could add a covariate such as "Used to drink lightly but stopped entirely because of a health problem" you will find the answer.
This study did create a variable which is the number of health problems and symptoms a person has. However, people can have one bad health problem/symptom (e.g., COPD on 3L home O2) which affects them more than 10 minor health problems might affect someone else.
As an aside, it is curious that this study did not replicate the repeatedly-observed finding of an association between obesity and mortality (OR = 1.00 in this study).
I will also add that this study is looking at older adults (aged 55-65 at the outset of a 20-year study, so they ended up dead or 75-85), which is different from the people discussed in the original thread (working age adults).
This study did create a variable which is the number of health problems and symptoms a person has. However, people can have one bad health problem/symptom (e.g., COPD on 3L home O2) which affects them more than 10 minor health problems might affect someone else.
As an aside, it is curious that this study did not replicate the repeatedly-observed finding of an association between obesity and mortality (OR = 1.00 in this study).
I will also add that this study is looking at older adults (aged 55-65 at the outset of a 20-year study, so they ended up dead or 75-85), which is different from the people discussed in the original thread (working age adults).