A randomized controlled trial (RCT) is a type of scientific study that tests whether a treatment, drug, or intervention actually works by randomly assigning participants to different groups and comparing the results. It sits at the top of the evidence hierarchy in medicine, ranking above observational studies, case reports, and expert opinion, because its design minimizes the biases that can distort results in other types of research.
How Randomization Works
The defining feature of an RCT is that participants are assigned to groups by chance, not by a doctor’s choice or a patient’s preference. This matters because it balances out differences between people, both the obvious ones (age, sex, severity of illness) and the hidden ones (genetics, lifestyle habits, unmeasured health factors). When those characteristics are distributed evenly across groups, any difference in outcomes can be attributed to the treatment itself rather than to something else about the people who received it.
The simplest version is like a coin flip: each person who enrolls has an equal chance of ending up in the treatment group or the control group. This is called simple randomization, and its strength is pure unpredictability. But coin flips can produce lopsided groups, especially in smaller studies. To prevent that, researchers often use block randomization, which ensures that after every set number of participants, the groups are roughly equal in size. In stratified randomization, participants are first sorted by important characteristics, like age or disease stage, and then randomized within those subgroups. This keeps the groups balanced on factors that could influence the outcome.
A critical safeguard is allocation concealment: the person enrolling a participant cannot know which group that person will end up in. In the landmark 1946 streptomycin trial for tuberculosis, the first modern RCT, the statistician Austin Bradford Hill sealed group assignments in numbered envelopes kept at a central office. The enrolling doctor only learned the assignment after the patient was already accepted into the study. This prevented anyone from steering sicker or healthier patients toward a particular group.
What the Control Group Does
Every RCT has at least two groups: one receives the treatment being tested, and the other, the control group, provides the baseline for comparison. Without a control group, there is no way to separate the effect of a treatment from the natural course of a disease, the placebo effect, or simple coincidence.
Control groups take different forms depending on the question being asked. A placebo control gives participants an inactive substance (a sugar pill, a saline injection) that looks identical to the real treatment. This isolates the biological effect of the drug from the psychological boost of simply receiving care. An active control gives participants the current standard treatment, which answers a different question: not “does this work better than nothing?” but “does this work better than what we already have?” Some trials use a no-treatment control, where the comparison group receives no intervention at all, though this approach makes blinding impossible.
Blinding: Reducing Bias After Randomization
Randomization handles bias at the start of a trial. Blinding handles it during the trial. If participants know they received the real drug, they may report feeling better simply because they expect to. If researchers know which group a patient is in, they may unconsciously interpret symptoms or lab results more favorably. Blinding prevents both.
In a single-blind trial, participants don’t know which group they’re in, but the researchers do. In a double-blind trial, neither participants nor the clinicians treating them know. In a triple-blind trial, the people analyzing the data are also kept in the dark. Double-blinding is the standard for drug trials because it guards against bias at every point where human judgment could skew results.
Blinding isn’t always possible. You can’t blind a surgical trial where one group gets an operation and the other doesn’t (though some studies have used sham surgeries, which raise ethical concerns). You can’t blind a study comparing physical therapy to medication. In these cases, researchers rely more heavily on objective outcome measures, like blood tests or imaging, that are less susceptible to interpretation bias.
Intention to Treat: Analyzing What Actually Happens
In a perfect trial, every participant would follow the protocol exactly. In reality, people skip doses, drop out, or switch treatments. The standard approach for handling this is called intention-to-treat analysis. It means every participant is analyzed in the group they were originally assigned to, regardless of what they actually did during the study.
This sounds counterintuitive. Why count someone in the drug group if they stopped taking the drug? Because intention-to-treat analysis preserves the balance that randomization created. The moment you start moving people between groups based on their behavior, you reintroduce all the biases randomization was designed to eliminate. It also reflects real-world conditions: when a drug is prescribed to thousands of people, some won’t take it as directed. The intention-to-treat result captures that reality.
Where RCTs Sit in the Evidence Hierarchy
Medical evidence is ranked by how likely a study design is to produce biased results. RCTs occupy the highest tier for individual studies. The only thing ranked above them is a systematic review of multiple RCTs, which pools data from several trials to reach a more reliable conclusion. Below RCTs sit cohort studies (which follow groups over time but don’t randomize), case-control studies, case series, and expert opinion.
RCTs earn this position because randomization directly addresses confounding, the problem where an unmeasured variable drives both the treatment choice and the outcome. Observational studies can try to control for confounders statistically, but they can never account for variables they didn’t think to measure. Randomization handles all confounders, known and unknown, in one step.
When RCTs Are Used
RCTs are most prominent in Phase III clinical trials, the stage where a new drug or treatment is tested in large groups of 1,000 to 3,000 people to confirm effectiveness, monitor side effects, and compare it against existing treatments. Phase III results are typically what regulatory agencies require before approving a new therapy. Some Phase II trials also use randomization, though on a smaller scale.
Beyond drug development, RCTs are used to test surgical techniques, public health interventions, educational programs, and policy changes. Any time you want to know whether doing something causes a specific result, an RCT is the strongest design available.
Limitations and Ethical Boundaries
RCTs are powerful, but they aren’t always possible or appropriate. The most fundamental constraint is ethical. You cannot randomize people to smoke cigarettes to study lung cancer. You cannot withhold a proven life-saving treatment to create a placebo group. The ethical standard known as equipoise requires that, at the start of a trial, there must be genuine uncertainty about which treatment is better. If strong evidence already favors one option, randomizing patients away from it is not justifiable.
Placebo-controlled trials raise particular concerns in psychiatry, where withholding effective treatment for conditions like depression or schizophrenia can cause serious harm. The capacity of some psychiatric patients to give fully informed consent adds another layer of complexity. Sham surgery trials, where patients undergo anesthesia and incisions but receive no actual procedure, face similar criticism: the risks of anesthesia and surgical wounds are real even if the operation is not.
There are practical limitations too. RCTs are expensive and time-consuming. They often enroll narrowly defined populations (excluding people with multiple health conditions, older adults, or pregnant women) to reduce variability, which can make results less applicable to the broader population. A drug that works well in a tightly controlled trial with carefully selected participants may perform differently in everyday clinical practice.
Trials conducted in developing countries have drawn scrutiny when placebo controls are used in situations where effective treatments exist but are unavailable locally. Critics argue this exploits the lack of healthcare infrastructure to run studies that would be considered unethical in wealthier countries.
The Five Types of Bias RCTs Address
The Cochrane Collaboration, which sets the standard for evaluating trial quality, identifies five domains of bias that can affect RCT results. Bias from the randomization process occurs when group assignment is predictable or improperly concealed. Performance bias arises when participants or clinicians behave differently based on knowledge of group assignment. Bias from missing data happens when people drop out at different rates between groups. Measurement bias occurs when outcomes are assessed differently depending on the group. And reporting bias emerges when researchers selectively present results that favor a particular conclusion.
A well-designed RCT has safeguards against all five: proper randomization with concealed allocation, blinding of participants and researchers, protocols for handling dropouts, blinded outcome assessment, and pre-registered primary outcomes published in a clinical trials database before results are known. Pre-registration is especially important because it prevents researchers from testing dozens of outcomes and only reporting the ones that showed a positive result.