One way to see if a treatment works is to compare people who received the treatment with people who received no treatment or a different treatment. However, the ways in which people are different from each other can bias results. For example, if people who do not get treatment are older or sicker than are people who get treatment, results could suggest that the treatment works better than it really does. One way to avoid this type of bias is to compare people with themselves before and after treatment. Each individual’s risk of the outcome when treated is compared with that same individual’s risk of the outcome under no treatment or a different treatment, providing 1 stratum of data. The strata contributed by each individual are statistically combined to compare the risks of the outcome under different exposure levels. In these “highly stratified” approaches, which include case-only approaches, each individual contributes a stratum of information that may explain differences in exposure to the intervention and occurrence of the outcome during the study. The great strengths of case-only designs are that they only require information on cases, and they eliminate confounding by time-invariant characteristics, both measured and unmeasured. However, the results are typically reported as ratio measures of association rather than the absolute increase or decrement in risk or a measure of how many people need to receive treatment for one of them to experience benefit or harm. These measures may be preferable for communicating results for informed decision-making by patients, health care providers, and other stakeholders.
