LITFL has the dummie's guide to clinical trial design.
Lets say you have some fancy new idea, be it a new drug or some sort of new mechanical technique.
In vitro activity: Cell cultures or whole organs are subjected to the treatment, to determine the mechanism of action at a cellular level.
Animal model: The safety of the treatment is established, and dose-finding experiments are performed. Animal models are also used to determine such things as foetal toxicity of drugs.
Phase 1: healthy volunteers are subjected to the experimental treatment. The data is used to determine pharmacokinetics of the substance in humans.
Phase 2: Patients with the disease of interest are subjected to the experimental treatment. The magnitude of the treatment effect is thus estimated, and this helps determine an appropriate sample size for the next phase of trials (obviously if your treatment effect is very subtle, you will need a massive sample to detect it).
Phase 3: large scale trial on the patients with the disease, powered sufficiently to determine the effect of the treatment on the outcome measure of interest.
Phase 4: This is essentially the gathering of data which occurs after the drug is unleashed on the population at large. Date is collected about this "post-marketing experience". Rare side effects and long term outcomes are usually established by means of Phase 4 research.