# Statistics: Sample size for research

Watch
Announcements
Thread starter 5 years ago
#1
I am doing a study which is testing a new diagnostic tool for detecting breast cancer. I am comparing it to a gold standard so get sensitivity/specificity values.

Is there any sample size calculations I can do to find out how many patients I need to test?

My stats knowledge is terrible so any help will be appreciated!
0
reply
5 years ago
#2
It's usual to do enough samples to have a 95% confidence interval, and on this website
http://www.sciencebuddies.org/scienc...icipants.shtml
you can see what the approximate margins of error, given the 95% confidence interval.

It's up to your judgement to work out the payoff between sample size and margin of error, but if you would like more information about sampling this looks like a good resource:
http://stattrek.com/sample-size/simp...om-sample.aspx
0
reply
5 years ago
#3
(Original post by elephantalkali)
I am doing a study which is testing a new diagnostic tool for detecting breast cancer. I am comparing it to a gold standard so get sensitivity/specificity values.

Is there any sample size calculations I can do to find out how many patients I need to test?

My stats knowledge is terrible so any help will be appreciated!
So presumably you are looking for a sample size that gives you a specified precision for your estimates of sensitivity and specificity. The usual way of doing this is via 95% confidence intervals. Specificity and sensitivity are both expressed as proportions - and therefore a good way of modelling them statistically is as binomial proportions. This wikipedia article guides you through the ways of calculating confidence intervals for binomial proportions, but basically your workflow is to:

(1) Specify how wide you want your confidence interval to be.
(2) Estimate (from previous studies) what the values of sensitivity and specificity are likely to be. (You need this as the confidence interval estimates are functions of the actual values of the proportions; if you can't, assume that they are both 0.5 and this gives you a "worst case" estimate).
(3) Solve for n in the formula for the confidence interval. I suggest that you use the normal approximation interval in the first instance. If the sensitivity and/or specificity get close to one or zero, then this becomes increasingly inaccurate and you have to be more sophisticated - follow on down the page to see how!
0
reply
X

Write a reply...
Reply
new posts
Back
to top
Latest
My Feed

### Oops, nobody has postedin the last few hours.

Why not re-start the conversation?

see more

### See more of what you like onThe Student Room

You can personalise what you see on TSR. Tell us a little about yourself to get started.

### Poll

Join the discussion

Yes (37)
29.13%
No (90)
70.87%

View All
Latest
My Feed

### Oops, nobody has postedin the last few hours.

Why not re-start the conversation?

### See more of what you like onThe Student Room

You can personalise what you see on TSR. Tell us a little about yourself to get started.