You know those types of questions during the Progress Test where you start reading them and you think: “Hmm… Case-control… Randomised control… I kind of think I recognise some of these words,” but then when you finish reading the full question you are like: “Wait, what?” Then there is a good chance that you have been confronted with a question about study design. Those of you who have not spent an ungodly amount of time in the Junior Scientific Masterclass programme (for that good resume juice) will probably not really have a good idea of what that entails.
Therefore, as a veteran of the JSM extracurricular programme, I will enlighten you with the science, nay, the art of study design (and help get a few extra points on your next Progress Test). Afterwards, I will attempt to navigate some of the most common biases in research which are sometimes asked about on these tests. This article is best read in conjunction with the Professor Progress Test about statistical tests in a previous edition of the PanEssay.
Descriptive studies
The name ‘descriptive study’ is pretty descriptive. Namely, they describe something and do not prove anything. An example of a descriptive study is the case report/series in which a single interesting or a series of interesting patients are reported. For example, a drug which is traditionally not given is given experimentally and it is surprisingly effective. If we ignore the obvious ethical issues with this it could be the basis of a larger study, but in itself proves nothing scientifically. Therefore as you can surmise, there is very little here about which questions could be based on for the Progress Test.
Analytical studies
Again we have a pretty descriptive title. Analytical studies, quite fittingly, analyse. In other words, you can make a conclusion based on this type of study. You usually start with an idea, for example, if you smoke then you have a higher chance of developing a lung tumour. Or, for example, if you eat apples then you have a lower chance of crashing a chopper into a cliff wall. The earlier example is a bit easier to work with, so for the examples of analytical study designs I will keep referring to this concept. This is the type of study which produces the famous ‘risk factors.’
Cross-sectional studies
This is the most simple analytical study (and I personally think it could also be classified as descriptive). In this type of study you record a group of people and ask if they smoke and figure out if they have lung cancer. Then you put that in a table. Now you know that this many smokers have lung cancer and that many do not have lung cancer. However, you do not know if the lung cancer is caused by smoking, maybe it is actually because more people who work in the radioactive waste industry smoke. Therefore, you are not allowed to say that smoking causes lung cancer, you ARE allowed to say that more smokers have lung cancer (in the studied population).
Case-control studies
The case-control study design is the bread and butter of articles about how peanut butter decreases the risk of heart disease and how drinking a glass of wine each day is somehow healthy or unhealthy depending on the weather. In this design, you take a group of people with lung cancer and a group of people without lung cancer. Then you start figuring out if they have been exposed to, for example, smoking or working radioactive waste management. This is usually done by looking in a database going through past records of patients. You are allowed to say that smoking and/or a profession in radioactive waste management could increase the risk of lung cancer.
Cohort studies
The cohort study is the best way of figuring out if smoking causes lung cancer but it is also massively expensive and time-consuming. You take a big group of people and ask them if they smoke. Then you wait and keep asking them if they develop lung cancer. Then at the end, you can tell who developed lung cancer and who did not. This way you can say pretty accurately that smoking makes you that many times more likely to get lung cancer. But again, this takes many years and many euros to find out. Usually, only governments can really afford to conduct this type of study on a big scale.
Controlled studies
Do you think eating an apple a day keeps the herpes away? Well, prove it! If you can you will make a lot of money but proving it also takes a lot of money. The most logical way to prove this is by giving one group of people an apple a day and give none to another group. Then you compare the number of people who still have herpes after this apple a day therapy and the number of people with herpes who did not have this ground-breaking new treatment. Generally, results are quite disappointing, which is especially frustrating for the researchers. The best way to ensure that the results are most reliable is by randomising the group that gets the apples and the group that does not get the apples. This makes sure that people with worse herpes are not getting more apples than those with better herpes. This makes sure the researchers cannot influence the results by selecting good patients for the treatment which makes the treatment look better.
Biases
There are many types of biases and many of them are pretty common sense. But a few of the less intuitive ones keep showing up on Progress Tests so I will quickly go over some of them.
Lead time bias
Lead time bias is actually not really a bias of study design, it is actually a bias caused by screening programmes. These programmes usually also serve as epidemiologic studies themselves so I feel justified including this type of bias. Imagine you want to screen people for lung cancer and you use an MRI to do this (actually a terrible idea but do not worry about it). You find A LOT more people than you expected based on previous epidemiological studies. After a while, no more people are actually dying of lung cancer than before. What happened? Well, you just found a lot of cancer before it became symptomatic, which is when most people go to the doctor. So these people were found ahead of time but still have the same cancer as before. Therefore it seems there are a lot more people with lung cancer, but it is actually the exact same number as before. So it gives them bad stats.
Length time bias
In the same sphere of bias as the lead time bias the length time bias is mostly a problem in screening. Your screening programme only measures the existence of cancer at one timepoint. If every type of cancer was the same this would be great, however, some are very slow, some are very aggressive. This means that many aggressive forms of cancer are missed in this programme because they either have already died or the cancer has not developed yet.
