Importance of Design Selection
There are tons of accepted and frequently applied study designs like parallel or cross-over, open or blinded, prospective, retrospective or case-study designs, etc. Each desgin provides a flexibility and sets up some limitations at the same time. The research/medical questions can answer which limitations are acceptable and what level of flexibility is required. The study statistician is able to form this picture with transforming the research questions into strict mathematical/statistical framework, An easy example to illustrate this creative process: the efficacy of a blood pressure decreasing intervention (let’s say, a drug) can be measured on a continuous measure (SBP/DBP decrease) and on a frequency distribution (normotensive vs. abonormal BP). In a perfect study design both opportunities are examined and the best fit to the specific study drug will be selected.
There are a numerous “classical” designs. They are well-known and the associated sample size determination (design and sample size determination are close relatives) is generally built in softwares like SAS, R or PASS. The ideal study design generally can be chosen from these traditional ones. But they also have some disadvantages compared to the so called modern designs: pooling of different phases is not really reported, flexibility in changing the number of treatment arms practically does not excist.
The term of modern designs is not an exact definition. Generally those desings are considered here which somehow extend the limitations of “classical (frequentist) designs”. The modern, sometimes labelled as Bayesian designs generally are more cost-effective compared to classical designs, as they allow to evaluate the outcomes after each patient and also allow the stopping of the study (with a confirmed positive outcome) before inclusion of the planned number of patients. But the price should be paid here, too.