exposed and unexposed individuals. From this information, the risk of the event occurring in the exposed group relative to the risk for the unexposed group (the relative risk) can be estimated. Unlike well-conducted RCTs, results from cohort studies can be influenced by selection bias and confounding—if they exist. Selection bias occurs when there are unmeasured factors that are related to the outcome of safety and also affect selection for use or non-use of a supplement. Confounding can occur when use of a dietary supplement is strongly correlated to other characteristics of individuals that also affect the safety outcome. These two potential study design problems should be considered when adjusting the level of concern warranted by cohort studies.

Primarily due to their relative expense, greater need for dedicated personnel, and inability to detect rare, serious adverse events, cohort studies are likely to be used less frequently than case-control studies in a postmarketing environment. Expense of cohort studies decreases when computerized medical records can be used, and thus as more records include information about intake of dietary supplement ingredients, the availability of cohort studies for dietary supplements is likely to increase. Nonetheless, for identified rare, serious adverse events, case-control studies may be more common.

Case-control epidemiological studies are uniquely useful at estimating the likelihood that an ingested substance causes an adverse event when the occurrence of the event is rare or occurs following a long latency period. In a case-control study, cases (persons with the event of interest) and controls (persons who do not have the event of interest) are identified. The exposure rates among cases and among controls are then estimated.⁷ Using the estimated exposure (dose is rarely known) in the controls and in the cases, the odds of the event in the exposed group relative to that in the unexposed group (the odds ratio) is estimated. (The odds of an event are equal to the probability of the event divided by one minus this probability).

Case-control studies require information on fewer individuals than cohort studies (see next section). However, in case-control studies there is the potential for bias caused by inappropriate selection of the control group or