Biostatistician Tamnil Alfred explains clinical trial design, from the four phases of drug development to why randomized controlled trials reduce confounding. He walks through randomization and blinding using a vitamin D study example, then introduces a dataset for hands-on R analysis.
Chapters
Hi, my name is Tamnil Alfred and I work as a biostatistician in the pharmaceutical industry. I am excited to be your instructor for this introductory course on the design and analysis of clinical trials. In this course I'm going to introduce important principles and some of the statistical analyses that we commonly perform for this course. I am assuming that you already have some experience using R and working with datasets. Let's begin with a definition. Clinical trials are scientific experiments used to evaluate the safety and efficacy of one or more treatments in humans. They are used by many organizations, including pharmaceutical companies, for drug development and to assess the effects of medical devices or procedures in the pharmaceutical industry. Clinical trials can be classified into four general phases. Phase 1 trials are typically conducted on a small group of healthy volunteers to observe what the body does to the drug and any side effects occurring at high doses. Phase 2 studies are conducted on a small group of patients with the disease using various doses of the new drug to identify which dose is optimal for efficacy and safety. Phase three is the most important phase when a selected dose of the new drug is evaluated in a larger group of patients for efficacy and safety against a control, the standard treatment or placebo and finally, once the drug is on the market, post marketing surveillance is done to identify any long term and rare side effects. In controlled trials, patients are typically randomized to one of at least two two treatment arms. Randomized controlled trials are considered the gold standard for assessing the effect of a new treatment and are extremely valuable in medical research. A well conducted trial can provide strong evidence for treatment effect by being able to reduce the impact of confounding that often affect observational epidemiological studies such as cohort studies. For example, suppose we are interested in assessing the effects of vitamin D supplementation on bone mineral density in older women. We may find that women who choose to take vitamin D have different social factors and habits to women who do not. They may smoke less, exercise more or be more likely to take calcium supplements. In an observational study, it may be difficult to exclude the effects of these and other potential factors on bone mineral density and accurately measure the effect of just the vitamin D supplementation. In a randomised controlled trial, we can assign women to receive either vitamin D or a placebo in a random fashion and therefore balance the two groups with respect to patient characteristics. We can check this by presenting summary statistics of the patient characteristics collected at baseline for each of the treatment groups. Another way to help us accurately estimate the treatment effect is by blinding. In a single blinded study, the patient will not be aware of which treatment they receive. In a double blinded study, neither the patient nor the study investigator is aware of which treatment is given. This can help reduce bias, for example, in the assessment of the treatment effect and how the study is conducted. Often this can be done by using placebos that resemble an active treatment. However, blinding may not be achievable, say in a trial comparing surgery with antibiotics for the treatment of appendicitis. Now let's look at an example of data collected in a trial by exploring one of the data sets we will be using in this.
