RCT (Randomized Controlled Trial)

So far we have seen that a clinical study measuring the effect of a treatment compares two groups of patients: one that received the treatment and one that did not. If the researcher decides which group each patient goes into, the study is experimental; if the groups formed on their own in real life, it is observational. Observational studies come in several kinds, depending on whether they see patients at a single moment or follow them over time (cross-sectional or longitudinal), and whether they collect data from today onward or take it from old records (prospective or retrospective). Now we reach the most credible kind of experimental study: the RCT.

The problem the RCT was built to solve

Comparing two groups only means something when the two groups are alike in every respect other than the treatment itself. Suppose (a hypothetical example) we want to compare two surgical techniques, and the surgeon, without realising it, picks the easier patients for the new technique and the more complex ones for the old technique. At the end of the study the new technique shows a better result. But that result is not because of the new technique; it is because better patients ended up receiving it. A study built this way does not give the right answer, however carefully it is carried out.

The RCT was built to solve exactly this problem. The name stands for randomized controlled trial, and each word marks one feature. Trial means a study in which a treatment is tested on patients. Controlled means it has a control group, the group that does not receive the treatment and is set beside the treatment group for comparison. And randomized means that which group each patient falls into is decided by a random method: not by the patient, not by the doctor, and not by any rule that could be guessed in advance. It is this third word that separates the RCT from other experimental studies.

How randomization makes the two groups alike

Every patient has many characteristics that can affect the outcome of treatment: age, comorbidities, hygiene habits, smoking, genetic background, and many other things the researcher does not even know to think about. When patients are divided between two groups by hand, there is always a risk that one of these characteristics will pile up in one group without anyone intending it, which is exactly what happened in the surgical example.

Randomization means handing this division over to chance. When allocation is completely random, every characteristic, whether or not the researcher knows about it, has an equal chance of landing in either group. The researcher does not need to know every influential factor in advance and plan for each one. It is enough to bring chance in, and all of them, known and unknown, end up in the two groups in roughly equal measure. That is why randomization counts at once as the simplest and the most powerful tool of clinical research.

But this tool only works when the division is truly random. The right method is a coin toss or, as is usually done, a table of random numbers or a computer-generated sequence. If patients are divided by whether their birth date is odd or even, by record number, or by alternate attendance, the division is not random, because anyone involved in the study can guess which group the next patient will fall into. As soon as it can be guessed, the door opens to interference; for example, a complex patient can be enrolled a day later so that they land in the other group. These studies are sometimes called pseudo-randomized or quasi-randomized, but in fact they are not randomized and should not be counted as RCTs.

Bias

Anything that moves a study's result away from the truth is called bias. The unconscious selection of easier patients for the new technique, for example, was a bias. By definition, bias can work in either direction, making the treatment effect look larger or smaller than it really is. But when researchers have examined published studies to see which way errors go in practice, they have found a clear pattern: weak studies most often show a treatment as better than it is. That is, they either present a treatment as effective that may not be effective at all, or they make a genuinely effective treatment look more effective than it is. This matters when reading a paper: when a study is weak and reports a very good result, it is more likely that the result is an exaggeration than that the treatment is really that good.

Randomization is the most important tool for reducing bias, but on its own it is not enough. Randomization makes the two groups alike at the starting point, but after the study begins, people's behaviour can once again create differences between the groups. The next tool is for that stage.

Blinding and placebo

A patient who knows they received the new drug may report less pain than they really have, because they expect to get better. A doctor who knows a patient is in the treatment group may, without realising it, follow them up more closely or record the result a little more optimistically. Neither of them intends to cheat, but knowing the group affects their behaviour. The way to prevent this effect is blinding: the people involved in the study do not know which group each patient is in. When neither the patient nor the researcher knows, the study is called double-blind.

For the patient not to know which group they are in, we need a placebo. A placebo is something that looks exactly like the real treatment from the outside but contains no active substance. For example, if the drug under study is a small white capsule, the placebo is also a small white capsule of the same shape and size, only with an inert substance inside. If the control group receives nothing at all, the patient immediately knows they are in the control group and blinding is lost. With a placebo, patients in both groups receive something that looks the same, and their expectations of treatment stay equal in both groups.

The problem is that in many important dental studies a placebo is not possible. The surgical equivalent of a placebo is a sham intervention, meaning the patient is taken to the operating room and the appearance of surgery is carried out without any real surgery being done. No patient would agree to be in such a group in a study of orthognathic surgery or TMJ surgery. For this reason surgical studies are necessarily open, meaning the surgeon knows what they did and the patient knows what was done.

But being open does not mean blinding is lost entirely. In every study, besides the patient and the surgeon, there are three other people who can still be kept unaware. The first is the person who assesses the outcome of treatment. This person should not be the surgeon and should, as far as possible, not know which treatment the patient received. To help with this, patients are also asked not to give the assessor clues during the examination. Of course the surgical scar usually gives it away, so an outcome measure should be chosen from the start that can live with this limitation. The second is the statistician who analyses the data, and the third is the person who writes the final report of the study. For these two, the data are handed over with the groups identified only by a code (for example group A and group B), and until the analysis and writing are finished, it is not revealed which code was the treatment group. Keeping the statistician unaware has gradually become more common, but keeping the report writer unaware is still rarely done.

Two special designs in dentistry

In dental research, especially in periodontal studies, two special forms of RCT are seen often. In a cross-over design, each patient receives both treatments in two separate time periods and their own result is compared with itself. In a split-mouth design, each patient's mouth is divided into two halves and each half receives one treatment. The advantage of these two designs is that the effect of a treatment can be shown with fewer patients. But both are valid only when certain conditions are met; otherwise the result becomes misleading. Explaining these conditions is beyond the scope of this series. It is enough to know that when you see such a design, it is a study that should be read with extra care.

Summary

An RCT has three components: a treatment is tested on patients, there is a control group for comparison, and patients are divided between the groups at random. Randomization, if it is truly random, makes the two groups alike at the start of the study in all characteristics, known and unknown. Blinding, which a placebo makes possible, prevents knowledge of the group from affecting the behaviour of patient and researcher after the study begins. Where a placebo is not possible, the outcome assessor, the statistician and the report writer can still be kept unaware. All of these tools exist to reduce bias, an error that in practice most often shows a treatment as better than it is.

Not every experimental study has these components. Some have no control group at all, and some build their control group from records of previous years. These studies, and why they count as weak evidence, are the subject of the next part, together with the first kind of observational study we will reach: the cohort.