A major challenge when researchers rely on non-randomized studies is the possibility that the groups being compared differ on some unmeasured factor. The authors developed a new research design that avoids this problem in settings where there is a prominent time trend in the exposure of interest—exposure to a new drug, for instance. Exposures with prominent time trends, often behavioral or environmental, are common.
The trend-in-trend design looks at changes in the frequency of an outcome as it relates to changes in an exposure such as a new drug, across subgroups that adopt the drug at different rates. For example, say there is a recently approved drug that we are concerned might increase the risk of heart attack. We would divide the population into subgroups based on the speed at which the subgroups adopt the drug. If the drug actually does increase heart attack risk, then subgroups with faster adoption will have larger increases in the number of heart attacks.
Randomized clinical trials are usually too small to identify rare adverse effects and are performed in a highly restricted group of patients—one that does not reflect the group that will actually take a drug once it is approved. Most follow-up studies require us to select a particular control group, and some drugs may not have a useful comparator drug. Follow-up studies also can give biased answers if the people who take the new drug are different, in unmeasured ways, than those who take the comparator drug. The trend-in-trend design has the potential to overcome these obstacles.