Subgroup analysis of large trials can guide further research: a case study of vitamin E and pneumonia