The Cross-Sectional Study

In the previous parts we met studies that move through time to find the link between a factor and an outcome. A cohort starts today and follows people toward the future to see in whom the outcome occurs. A case-control study starts from today’s patients and goes back into the past to see who was exposed to the factor. In this part we reach a study that neither moves toward the future nor goes back into the past, but measures everything at one and the same moment.

How this study works

First, let us recall two terms. “Exposure” means that a person has had contact with a factor we think may affect their health; for example, they have taken a drug or worked in a particular environment. “Outcome” is the event we want to know is or is not related to that factor; for example, a disease.

In a cross-sectional study, the researcher examines a group of people at a specific point in time and, in that same examination, asks both whether they were exposed to the factor and whether they have the outcome. Both pieces of information are collected together, in a single round. The researcher then looks at whether those who were exposed have the outcome more often than the rest. If so, we say that an association has been observed between this factor and this outcome.

The difference between this and the two previous studies lies in time. In a cohort, we first record exposure and then wait years for the outcome to occur. In a case-control study, we first find the patients and then ask about their past. In a cross-sectional study there is no gap between the two; we ask about both, right now and together.

An example

Suppose we want to know whether taking a particular drug during pregnancy is related to cleft palate in the newborn. The researcher interviews a large group of women and asks each of them two things: have you given birth to a child with a cleft palate, and did you take that drug during pregnancy? Both questions are answered in a single interview. The researcher then divides the women into two groups, those whose child has a cleft palate and those whose child does not, and compares in which group drug use was more common. If it is markedly more common among the mothers of affected children, an association has been observed between the drug and cleft palate. Cleft palate, however, is a rare outcome, and such a study needs a very large population to reach enough affected children; this is exactly the situation in which, as we said in the previous part, case-control is the best option.

Why this design is attractive

This study is simple and inexpensive. It needs neither years of follow-up nor a search for rare patients. A single round of data collection is enough. Nor does it raise ethical problems, because the researcher gives nothing to anyone and withholds nothing from anyone; they only ask people about their present situation and their past. For this reason, the cross-sectional study is one of the most common designs seen in published papers.

First limitation: association, not cause and effect

The most important thing to remember when reading a cross-sectional study is that it can only show that two things are seen together. This is called “association”. What it cannot show is that one caused the other; this is called “causation”. The difference between the two is simple but fundamental. That drug use was more common among the mothers of children with cleft palate means the two go together. But going together does not, by itself, prove that the drug causes cleft palate.

The reason is that when factor and outcome are measured at the same time, their temporal order is often not known from the study itself; in our example that order is clear from outside the study, because the drug was taken before birth, but for many questions it is not. More importantly, the possibility that a third factor is involved always exists. For example, a particular underlying condition in the mother may both have led her to take that drug and, independently, have raised the risk of cleft palate. In that case the drug and cleft palate are seen together without the drug being the cause. A cross-sectional study has no way of separating this possibility from true causation. So the result of such a study should be read as “these two were seen together”, not as “this one causes that one”.

Second limitation: reliance on memory

In many cross-sectional studies, information is collected by asking, which means that both exposure and outcome depend on people’s recall. This is the same recall bias we saw in the previous part. A mother whose child has a cleft palate may remember the drugs she took during pregnancy more carefully, while a mother with a healthy child has never thought about that period in this way. This difference in the accuracy of recall can by itself create an association that does not exist in reality.

Where this same design is right

So far we have examined the cross-sectional study for causal questions and seen that it is a weak design. But there is another kind of question for which this very design is the first choice: the diagnostic question.

A diagnostic question means wanting to know how well a test works; for example, how reliable bitewing radiography is in detecting interproximal caries. Answering this question does not require exposed and unexposed groups. Instead, each patient is examined by two methods: the test we want to evaluate, and a method we trust more, called the reference standard. The results of the two methods are then compared for each patient to find out how many times the test was right and how many times it was wrong. This is done at a single point in time and needs neither waiting nor going back into the past. For this reason, the appropriate design for a diagnostic question is a cross-sectional study. We will see this point again in the part on question types.

Summary

A cross-sectional study measures exposure and outcome at one moment and together. It is simple and inexpensive and raises no ethical problems, but for causal questions it can only show that two things are seen together, not that one causes the other, and because it usually relies on people’s recall, it is also vulnerable to recall bias. The same design is a suitable choice for the diagnostic question, in which a test is compared with a reference standard.

The next part reaches the lowest rung of the evidence ladder: the case report and case series, observations a clinician has seen and recorded in a few patients.