No doubt you’ve seen something like this hierarchy of studies from Duke University’s Introduction to Evidence-Based Practice site. In this hierarchy, the studies are classified by the type of sample used.
Types of Samples
In a case study or case report, the sample is one or two individuals who experienced something unusual or noteworthy.
In a case-control study, the sample is a group of individuals with the characteristic of interest (usually diagnosed with something) called the cases and another set of individuals who do not have that characteristic called the controls. The controls are matched to the cases on as many demographic and health criteria as possible to allow researches to identify meaningful differences between the groups (the differences that lead to the diagnosis). Here is an example of a case-control study: Identifying risk factors for recurrent cesarean scar pregnancy: a case-control study.
In a cohort study, the sample is a group of individuals selected and followed for a time. If the study is prospective (meaning it starts with healthy individuals and watches to see who develops the diagnosis of interest), the group will include all people who do not have the diagnosis. Over time some will develop the diagnosis and researchers will try to determine what was different about those who did and those who did not. Retrospective cohort studies begin with a group of individuals with similar characteristics and some (or all) already have a diagnosis of interest. Researchers then ask about their history or watch what happens as time progresses to identify differences in disease progression. Here is an example of a cohort study: Risk of respiratory morbidity in term infants delivered by elective caesarean section: cohort study
In a randomized controlled trial, the sample is a group of individuals who meet very specific criteria. The sample is then randomly assigned to case and control. Here is an example of a randomized trial: A randomized controlled comparison between combined spinal-epidural and single-shot spinal techniques in morbidly obese parturients undergoing cesarean delivery: time for initiation of anesthesia.
In systematic reviews and meta analyses, the sample is a group of peer-reviewed studies meeting specific inclusion criteria. Here is an example of a meta analysis: Decision-to-incision time and neonatal outcomes: a systematic review and meta-analysis.
But this is not the only way to classify a study, and if you are just beginning to read scientific literature you may not understand all the built-in differences between the types of studies. So today we will talk about other ways to classify studies.
Observational or Experimental
In an observational study, the researchers watch what is happening to each participant without attempting to change behavior or health. Most birth research is observational.
In an experimental study, the researchers introduce an intervention to the study group. In the birth world, some experimental research is done comparing medication effectiveness and some is done comparing pregnancy behaviors like diet and exercise.
So why does observational vs experimental matter? The difference is due to the strength of evidence for causation. In an observational study the researchers are trying to identify associations between two variables — what characteristics are associated with giving birth by cesarean? However, by just watching the available women you might not be aware of confounding variables. This is common in birth because there are likely many differences between women who choose a birth center birth and women who choose a hospital birth. If we compared the rate of cesarean between hospital and birth center without controlling for those differences, we would end up with an over-estimation of the effect of a birth center on cesarean rate. So maybe in our study we adjust for risk factors, since some women are not eligible for birth center care. But that still leaves us with non-measurable differences
- women who choose a birth center do so for a reason (beliefs about the care they want), and that reason in itself may alter their risk for cesarean even if they gave birth in a hospital
- and women who choose a birth center may also make other health behavior decisions that alter their risk for cesarean
- and women who choose a birth center may have access to more resources than women who give birth in a hospital, both personal financial and/or community, which may alter their risk for cesarean
- and women who choose a birth center may participate differently in their health care decisions, which may alter their risk for cesarean
In an experimental study, the researchers are trying to isolate the effect of one particular characteristic — what effect does doula care have on the rate of cesarean delivery?
To be able to show causation with an experimental design, you need to have a randomized sample. This is not the same as a random sample. A random sample is achieved by randomly selecting members of a population to participate. Very little research actually has a random sample, because it is too difficult to achieve in health research. Instead, research relies on convenience samples. A convenience sample is achieved by including anyone who is willing to participate and meets the inclusion criteria. You can still randomize the sample for an experiment even if it is a convenience sample. To randomize, the participants will be assigned to either the experimental or control group based on a process of randomization (essentially flipping a coin, but more involved to ensure equal groups and allow blinding when required).
Cross-Sectional or Longitudinal
Survey research is generally cross-sectional, meaning the research team tries to gather information from people to represent the population at one point in time. A census is a special type of cross-sectional survey which attempts to gather data to represent the entire population at a given point in time. Vital statistics data is cross-sectional, and the Listening to Mothers series of studies are also cross-sectional surveys. They give us a view of what things look like right now, allow us to compare groups and measure trends over time.
In contrast, longitudinal studies follow the participants over a given and make multiple observations. Sometimes these studies last many years such as the Nurses Health Study or the Framingham Heart Study. Longitudinal studies let researchers identify changes over time and sequences of events.
Again, the difference between these is the weight given to the associations found. Why? A cross-sectional study helps you identify many possible associations but does little to tell you which came first. Here’s a hypothetical example not based on any evidence (just to illustrate). If you wanted to know if a daily 20 minute walk helps maintain a healthy blood pressure in pregnancy, you might survey all women to find out how often and how long they walked and then ask what their last blood pressure reading was. But doing a longitudinal study allows you to account for blood pressure changes throughout the pregnancy in women who walked and women who did not. You might find women with higher blood pressure are less likely to walk (maybe they have more discomfort or something else preventing them from walking). So even though in the cross-sectional study it looked like walking 20 minutes a day helps maintain healthy blood pressure, the longitudinal study may show women who have blood pressure issues don’t start walking, so the problem existed before the walking and was not prevented or fixed by it.
This is a long post today, I’m sorry. But I wanted to be sure you understood all the different ways to think about a study before we dive into how to interpret the statistics.