Heterogeneity in Assessing Self-Reports of Caffeine Exposure: Implications for Studies of Health Effects