90 PART 2 Examining Tools and Processes

the birth control pill as their method of choice. While descriptive study designs are

necessary in a practical sense, they are poor at developing evidence for causal

inference, so they are considered inferior to analytic study designs.

Getting analytical

Analytic designs include longitudinal cohort studies and case-control studies.

These are the strongest observational study designs for causal inference. Longitu-

dinal cohort studies are used to study causes of more common conditions, like

hypertension (HTN). It is called longitudinal because follow-up data are collected

over years to see which members of the sample, or cohort, eventually get the out-

come, and which members do not. (In a cohort study, none of the participants has

the condition, or outcome, when they enter the study.) The cohort study design is

described in more detail under the section, “Following a cohort over time.”

Case-control studies are used when the outcome is not that common, such as liver

cancer. In the case of rare conditions, first a group of individuals known to have

the rare condition (cases) is identified and enrolled in the study. Then, a compa-

rable group of individuals known to not have the rare condition is enrolled in the

study as controls. The case-control study design is described in greater detail

under the section “Going from case series to case-control.”

Going from observational to experimental

You may notice in Figure 7-1 that observational studies (which are either descrip-

tive or analytic) comprise most of the figure. Experimental studies — where par-

ticipants are assigned to engage in certain behaviors or interventions — are less

common than observational studies because they have ethical concerns, and are

often expensive and complex. However, experimental studies benefit from gener-

ating the highest level of evidence for causal inference  — much higher than

observational studies.

Climbing the Evidence Pyramid

Each of the study designs discussed in the previous sections generates a particular

level of evidence for causal inference. These levels of evidence may be arranged in

a pyramid. As shown in Figure 7-2, the study designs with the strongest evidence

for causal inference are at the top of the pyramid, and those with the weakest are

at the bottom.