Cardiology Coding Alert

Simple, stratified, cluster … learn the lingo of random sampling.

If you’ve been putting off internal audits or aren’t satisfied with your current methods, try the four steps below to get your team on the right track.

This real-world advice from Frank Cohen, MPA, MBB, director of Analytics and Business Intelligence for Doctors Management in Clearwater, Fla., will help you learn the best, most efficient ways to select a statistical sample of your practice’s charts.

Step 1: Recognize Why You Should Sample

The first question you should ask yourself before deciding what you will sample in your practice is, “Why do we sample?”

“Sampling has become a hugely important issue within our industry because of the surge of importance in things like analytics, the rise in interest in process improvement, and the fact that billing and coding audits, these recovery audits, are dominating our landscape,” Cohen explains.

Because it is often expensive and nearly impossible to collect 100 percent of the data from 100 percent of your data universe, sampling is an efficient and less expensive way to extrapolate data from a subset of that universe and infer the statistics.

Step 2: Determine Your Sample Size

Before you can start pulling charts that will comprise your data sample, you need to decide what your representative sample size is going to be. There are many formulas for how to select a sample size, Cohen says. He explains that whatever formula you use in your practice, you should make sure that the sample size is large enough to minimize sampling error and not so small that it no longer fairly represents the population in question.

“Too large a sample costs more money and resources without added benefits,” Cohen adds. “Too small a sample creates too much error and renders the results useless.” Your main goal when choosing your sample should be to ensure the sample is representative of the larger data universe you are trying to represent and that the sample is large enough to satisfy the assumptions of statistical analysis.

Step 3: Consider Random Sampling

Many auditing experts suggest that when you are performing chart audits on your physicians or coders, you should choose a random sample of the charts.

“Random sampling should eliminate bias in selecting units,” Cohen says. “So if I have a database of 10,000 claims and I randomly pull 30 of those claims, every one of those had an equal non-zero opportunity of being selected ... the sample I end up with should be non-biased and should be close to representative of the universe.”

If you have a random and representative sample, you can then infer the results also apply to the larger data universe. You have three methods to choose from for random sampling, according to Cohen, as follows:

1. Simple random sample: In this method, every claim has an equal chance of being selected for audit. An example of when you might use this method is when you want to look at and assess payer behavior.

2. Stratified sample: For this method, “you break the universe into homogenous sampling frames from which then a homogenous sample is drawn,” Cohen explains. You may do this by saying you are going to review all E/M claims or all cardiac catheterization claims, or if you are going to sample based on beneficiary type (age, gender, etc.) or by diagnosis code (i.e., all claims with a V code).

3. Cluster sample: “In cluster sampling, what we’re doing is organizing the units into smaller subsets,” Cohen says. You can do either two-stage (such as a random sample of beneficiaries and then random sample of claims for each) or multi-stage sampling (such as a random sample of beneficiaries from which you draw a random sample of claims from which you draw a random sample of claim lines).

Step 4: Make It Happen

Once you know why you are sampling, what you are sampling, and the method you’ll use to sample, you can go ahead and start the actual sampling process. Follow these steps for a simple, effective chart audit:

1. Define the population of interest — for example, you decide to review all claims from Physician A in your practice.
2. Create a sampling frame — continuing with the example, you decide to review all office visits for Physician A.
3. Determine the sampling method — you opt to do a simple random sample.
4. Calculate the sample size — you’ll pull 30 claims.
5. Sample the data — you go ahead and pull 30 office visit claims for Physician A at random.
6. Analyze the results — upon review, you make some determinations/analysis about those 30 claims.
7. Infer to the population of interest — finally, you extrapolate the data and apply it to the general data population.