Randomized Controlled Trials

What kind of questions can we answer with randomized controlled trials?

Treatment: 1). Drug A vs drug B vs placebo. 2). Medication vs surgery for appendicitis. 3). Counseling vs drug therapy for depression. 3). Herbal therapy vs drug therapy for depression.

Prevention: 1). Aspirin vs placebo to prevent heart disease. 3). Folate vs placebo prevent birth defects.

Diagnostic strategy: 1). CT or chest x-ray or MRI to diagnose blood clots in lung (pulmonary embolism)? 2). Ultrasound or CT to diagnose appendicitis?

Screening: 1). Digital mammography vs film mammography. 2). Self-breast exam education vs usual care. 3). Fecal occult blood test or colonoscopy?

1. Do you compare individuals or groups?: 1). Randomize Individuals. 2). Randomize groups.

2. Do you compare between groups or within groups?: 1). Parallel groups design – comparing one group to another (Strongest and most common type of RCT design. CONSORT diagram). 2). Crossover Trial: Comparing patients in the same group at different times (Need fewer patients than parallel groups design, but has greater potential for bias/confounding. May be more difficult to mask patients to treatment vs control.)

3. Superiority or non-inferiority trial?: Superiority trial: is my intervention better than nothing or than standard therapy? Non-inferiority trial: is my intervention no worse than the standard intervention?

Designing a good randomized controlled trial:

1. Control or comparison group
2. Randomization method
3. Masking
4. Allocation concealment
5. Description of dropouts/withdrawals
6. Adequate duration of study
7. Analysis: per protocol or ITT

Understanding the Results of a Clinical Trial: Statistical vs clinical significance

Example

Study randomizes patients to rosuvastatin (Crestor) and two older drugs (atorvastatin and simvastatin). All three are statins that lower cholesterol. Researchers follow patients for 5 years. Rosuvastatin: 36% relative reduction in the likelihood of myocardial infarction (MI); Atorvastatin: 1.4% decrease in the likelihood of MI (2.5% vs 3.9%); Simvastatin: you have to treat 71 patients to prevent one additional MI.