I often joke with colleagues that “statistics rule the world,” but I have to admit I’m only half joking. In polite society I usually receive a few chuckles when I make this statement, but within clinical research this mantra is an absolute. If your goal is to achieve a definitive conclusion, you must start with sample size planning. Any clinical study designed to investigate efficacy of a new treatment MUST know how many subjects are needed to arrive confidently at a definitive answer.
While this post is not intended to be an exhaustive guide, it is a reminder that clinical research is all about gathering data to draw conclusions about new treatments, confidently. Without confidence what have we really achieved?
There are three possible answers to evaluating the clinical effectiveness of a new treatment:
- The study has sufficient information to confidently conclude the new treatment is more effective than its comparator,
- The study has sufficient information to confidently conclude the new treatment is not more effective than its comparator, or
- There is insufficient information to conclude with confidence whether the new treatment is more effective than its comparator.
Conclusions 1) and 2) represent definitive decisions that can be made with an acceptably high level of confidence. Conclusion 3) often results when sample size planning is not done effectively and, at study end, the researcher finds out not enough subjects were studied to allow for a definitive conclusion.
A common misconception is that 30 subjects per treatment arm are adequate for many studies. The number 30 is important in the statistics world but not for sample size decisions. Many studies need hundred, even thousands, of study subjects to reach their goal, whereas some studies (such as crossover or tightly controlled lab studies) may need fewer subjects per treatment arm.
The needed sample size for each study is uniquely determined by:
- The primary research objective,
- Primary study endpoint,
- Expected relative efficacy of the investigational treatment relative to its comparator, and
- Measures of confidence desired for the resulting conclusion.
To better understand statistical concepts in clinical research I’ve compiled a couple of articles that are worth a read: Truth, Lies and Statistical Tests: Powering the Study Clinical trial design — for beginners.