Coding Audit Sample Size Calculator
How many charts should you audit? Too few and the results prove nothing; too many and the audit eats the week. Enter the size of the population you are auditing, pick a confidence level and a margin of error, and get the sample size that gives your audit real statistical weight.
A coding audit is only as good as its sample. Audit ten charts out of a thousand and a clean result proves almost nothing; audit all thousand and the audit never finishes. The right number sits in between, and it comes from three inputs: how big the population is, how confident you want to be, and how much error you will accept. This calculator works it out with the standard formula — the Cochran sample size for a proportion, adjusted with the finite population correction — and shows both the unadjusted and the adjusted figures so you can see what the correction saved you.
Work out the sample size
Worked example
| Input | Value |
|---|---|
| Population | 1,000 claims |
| Confidence level | 95% (Z = 1.96) |
| Margin of error | ±5% |
| Expected error rate | 50% (conservative default) |
| Unadjusted sample size (n0) | 385 records |
| Required sample (with finite population correction) | 278 records — 27.8% of the population |
Without the correction you would audit 385 records; because the population is only 1,000, the correction saves 107 records with no loss of statistical weight.
How the numbers are worked out
| Step | Formula |
|---|---|
| Unadjusted sample size | n0 = Z2 × p × (1 − p) ÷ e2 |
| Finite population correction | n = n0 × N ÷ (n0 + N − 1) |
| Rounding | Always rounded up — rounding down would leave the audit just short of the required precision |
Z is 1.645 for 90% confidence, 1.96 for 95% and 2.576 for 99%. The proportion p is the expected error rate — 0.5 by default, the point of maximum variation. The margin of error e is entered as a percentage and used as a decimal (5% = 0.05).
What to know before you audit
- Small populations need a large share audited. For a population of 100 claims at 95% confidence and a 5% margin, the required sample is 80 records — 80% of the population. That is not a mistake: with so few records, only a large share gives a meaningful result, and for very small populations auditing everything is often the honest answer.
- A smaller margin of error costs records fast. Cutting the margin from 5% to 3% roughly triples the sample size. Pick the margin your audit actually needs — 5% is the usual choice for an internal coding audit — rather than chasing precision you will not use.
- The 50% default is deliberately cautious. If a previous audit found a 20% error rate, entering 20% shrinks the sample — for a 1,000-claim population at 95% and 5%, from 278 records to 198. Only use a lower rate when you have real evidence for it; guessing low just to save time undermines the audit.
- Random means random. Number every record in the population and pick the sample with a random-number generator. Auditing the most recent charts, the easiest charts, or the charts one coder handled is not a sample — it is a selection, and it will not support any conclusion about the population.
- This sizes the audit; it does not design it. Stratified audits (by coder, by specialty, by payer), multi-site audits and any audit whose results will be extrapolated to an overpayment amount need a design beyond this formula. Use this number as the starting point, and involve your compliance team before extrapolating.
The math on this page is the standard Cochran formula for estimating a proportion, computed entirely in your browser from the numbers you enter. Nothing you type leaves your browser.
Disclaimer: This tool is for educational and estimation purposes only and does not constitute billing, coding or medical advice. Coverage, coding and payment rules vary by payer and change over time — always verify with the current CMS files and the payer's own policy before submitting claims or making financial decisions.
Frequently asked questions
What is the finite population correction?
The standard sample-size formula assumes an effectively infinite population. When the population you are auditing is small — a few hundred claims rather than tens of thousands — that formula asks for more records than you actually need, because sampling a large share of a small population tells you more than the same number of records would from a huge one. The finite population correction adjusts the sample size down to match the population you really have, and it is what this calculator applies.
Why does the calculator assume a 50% error rate?
The sample size depends on how varied the population is, and the most conservative assumption is a 50% rate — the point of maximum variation. Using 50% gives the largest sample size, so the audit is protected no matter what the true error rate turns out to be. If you already know the error rate from a previous audit, you can enter it instead and the required sample will shrink.
Which confidence level should I pick for a coding audit?
95% confidence with a 5% margin of error is the conventional choice for a coding audit, and it is the default here. A 90% level needs a smaller sample but leaves more doubt; a 99% level is more certain but needs many more records. The choice is yours — the calculator shows the sample size for each.
How do I pick the records once I have the sample size?
Use a random method, not a convenient one. Assign every record in the population a number, then use a random-number generator to pick the sample — never hand-pick the records, and never audit only the records that look easy. A sample that is not random is not a sample, whatever its size.
Does this replace a statistician's advice?
No. This calculator gives the standard Cochran sample size for estimating a proportion, which is the right starting point for most internal coding audits. Stratified audits, multi-site audits, and audits that must support an extrapolation of overpayments to the whole population need a more detailed design — talk to your compliance team or a statistician before extrapolating results.
