In ultra-rare disease research, our power is limited by very small numbers. How do you address the issue of low power due to small sample size, when increasing your N is not an option?
Hey, This is a problem for me too - and one that I have spent long and hard thinking about. If you can't alter N, then you may have to accept a lower power to your study. Altering power can be an issue for funding bodies. You therefore also need to ensure that you maximise uptake of N to the study (i.e. ensure that non-responders and recruitment are maximised - through good pilots and solid qualitative work if a clinical trial). Ensure your outcome is appropriate, and will similarly minimise N required - i.e. if using a binary outcome then the sample size required may be huge, but if a continuous/ ordinal outcome is equally plausible then this may lower sample size required. If N is still too small, then you need to think of collaborative means to increase it - be it local, national or international.
Thank you, Daniel. Most of our diseases are progressive, over time, and we don't even know the natural history of some of them. I prefer to pick my end points a priori, whenever possible.
Research in the real world is much more difficult than in grad school!
Dear Elizabeth, i appreciate your concern about the disease and the limitation by the numbers. However, you protect your interest by increasing your control group as suggested by Uppendra. Do not give up. lalitha kabilan
I agree you can increase the number of controls if it is case-control, but beyond 5-to-1 (five controls for every case), you get almost no increase in statistical power.
Also if you're looking for a comparison group, you can try to get access to a random sample of 'hospital controls' and look at their medical records. An example would be comparing cases of some rare disease (which you're studying) compared to hospital controls with, for example, some other disease like appendicitis.
Remember that when you're looking for a sample size when comparing two groups, keep in mind four things: 1) what is the level of significance (alpha level or p-value) that you're looking for---it doesn't have to be .05, it can be .10 for smaller studies; 2) how great should the chance be of detecting an actual difference if there really is a difference---in other words, the power (or 1 minus beta); 3)how large should the difference be between the two groups for the difference to be meaningful clinically; and 4) what is a good estimate of the standard deviations of the two groups. Also your chance of committing a Type I error or a Type II error.
See or buy the book: Basic and Clinical Biostatistics, by Dawson and Trapp.
If you are following your rare-disease subjects for a long time, then you'll likely have many data points. Or maybe your outcome is 'failure', e.g., death or relapse or some other endpoint. Survival analysis is probably what you'll be using in that case, either Kaplan-Meier or Cox proportional hazards models I don't know if Cox would give you more power than K-M. And check the literature for similar studies to yours and see what methods and results they've used.
I think the question should be more clearly formulated as it is unclear what kind of problem you want to investigate. In general there are 2 options for rare disease it is case-control study and experimental trial (like Nof 1 trial or quazi-experimental trial, cross-over types).
If the aim is to reduse sample size, then reducing sample size without losing power can be accomplished by one of the following ways: reducing variability and use of better statistics: use a continuous variables; find optimal cut-off value for variables; increase event rate (change definition of event, use composite endpoints); use surrogate endpoints; use same patient multiple times; stratify patients to reduce imbalance in groups and some other decisions can reduce required sample size.