Background:
I wanted to do a longitudinal survey over (3 time points) to measure fatigue levels and some correlates on a group of patients.
I know our patient numbers based on the database. When I did an estimate of eligible patients based on diagnosis and age, we have 194 eligible patients.
From similar studies, 80% of patients are eligible for the study and an estimated 40% attrition rate during the longitudinal measures. We are not aiming to power any specific endpoints as there are still a lot of unknown wit regards to the thing we are measuring. Based on this my questions are:
1. Are the information presented enough to estimate the sample size I need?
2. If so, can you signpost or enlighten me on how to best estimate the needed sample size?
Hi Harry
Since you have previous data about eligibility, you may consider this information as a known proportion characterizing the sample. So, patients are 80% eligible and 20% are not.
Further, you decide on the level of confidence in your data say 95% (most use, unless you decide for higher confidence of 99%) any way for each case the corresponding Z-value is 1.96 for the first and 2.58 for the second. The last piece of information is defining the level of error that you should allow, say 5% (or if more accuracy is needed you lower the number.
The formula based on known proportion is:
sample size = (Z)(Z)(p)(q) / (E)(E) = (1.96)(1.96)(0.8)(0.2) / (0.05)(0.05) = 245.86 or 246 patients.
Then you say you have attrition rate of 40%, which we must account for:
246/(1-0.40) = Needed sample size = 410 patients.
Now you can perform a correction factor if your original population of patients is known (that is, the pool of patients from which you want to choose your sample for the research study). Then, I will guess here, the population defining all types of patients who are eligible (database), say 2,000 patients (you may correct this),
N = 2,000
original calc sample size = 410 patients
New modified size = n= (Calc siz) / { 1 + [Calc siz -1] / N } = 340 patients.
The other selection criteria you mentioned are not considered here, but when you apply it seems you get an acceptable number you had of 194 patients.
There could be other approaches.
Thank you so much for this help. Stats seems to elude me during stressful times. This is highly appreciated.
Hussin, thank you for this. I would like to use in as an example in a course..
- Badges
- Science topic
- Similar topics
- Engineering
- Mechanical Engineering