I got a question from someone doing a quality review. She needs to pull a certain number of medical records out of 892 and see whether the doctors followed the clinical guidelines properly. The question is how to determine the proper number of charts to pull.
Since the outcome variable is binary (followed guidelines, deviated from guidelines), you would estimate a proportion and place 95% confidence limits around that proportion. The desired width of the confidence interval will tell you how many records you would need. So, for example, if you suspected that 80% of the doctors followed the guidelines, and you would be happy if your confidence interval around this estimate had a width of plus/minus 10% (that is, the interval would go from 70% to 90%), then you would need to sample 62 patients.
There’s a simple formula that you could use to get this value
height=“56”}
where H is the half width of the confidence interval and Z is the appropriate percentile from a normal distribution (1.96 for a 95% interval, for example).
You could also just find a program or a spreadsheet that does confidence interval calculations and use trial and error to find the right sample size. I have a simple spreadsheet that will work
or you could find something on the web. A nice general resource for pages that perform statistical calculations is
- Web Pages that Perform Statistical Calculations. John C. Pezzullo. Accessed on 2006-03-30. (Statistics, Software, Free) members.aol.com/johnp71/javastat.html
One wrinkle in the ointment is that only some of the charts that you pull will actually be relevant to your inquiry and there is no way of telling which charts will work without reviewing the entire chart. If you believe that only one out of every four charts meets your needs, then you should pull 248 charts to make sure that you will have around 62 that qualify when you are done.