This would mean that a reduction in BMI that is not associated with a reduction in calorie intake would not decrease the probability by as much as the model indicates, because calorie intake is maintained and has a separate impact on the probability. Omitted variable bias of this type is always a possibility, and the only way of burying the possibility is to include all feasible omitted variables in the model and see if they have a separate impact. Data limitations always preclude this perfectionist approach.