The main problem in using this concept is that there is no universal number below which the risk is reduced, since it depends on the individual weight history, which is also subject to recall bias. Other factor that could bias this analysis is the already discussed reverse causation: weight loss could be a proxy of a subclinical disease, in which the overall prognosis would be poor. So, ideally, we should exclude data on those who unintentionally lost weight in the past.