The limitations of using simple definitions of glucocorticoid exposure to predict fracture risk: A cohort study