NPs better than PAs in DM care

oldiebutgoodie, RN · Jan 19, 2008

Hi, all,

I am enjoying this thread regarding critique of the research. I have some questions for you statistics experts out there (I am still taking research class--maybe I'll be smarter in a few months!)

David, you were saying that 20 charts per practice is not enough to provide an appropriate sample size? Even though there were 20 charts per practice, for a total of 846 charts? In other words, each sample was too small, so the aggregate sample was too small?

Thanks,

Oldiebutgoodie

core0 · Jan 19, 2008

Hi, all,
I am enjoying this thread regarding critique of the research. I have some questions for you statistics experts out there (I am still taking research class--maybe I'll be smarter in a few months!)
David, you were saying that 20 charts per practice is not enough to provide an appropriate sample size? Even though there were 20 charts per practice, for a total of 846 charts? In other words, each sample was too small, so the aggregate sample was too small?
Thanks,
Oldiebutgoodie

Here is probably the best paper on sample size. It is math intensive but written in a way that makes sense. Also Lenth does a really good job of explaining things in a very clear manner:

http://www.son.wisc.edu/rdsu/library/tr303.pdf

Look at it this way. It is all about proportionality. Is the chart that you look at representative of the practice as a whole. If there are 1000 charts then what percentage of charts would represent the whole. You can usually make a decent representation above 5% depending on the rarity of the condition. This gives you about a 75% CI. Across a practice that will result in an acceptable proportion. I have never done a cross sectional analysis with less than 5% and I actually don't know how you would figure out what small percentage would give you a good answer.

Looking at it another way would you be comfortable with an insurance company rating your care of a patient population by looking at one chart. Or given the design of this study would you be happy with them rating your care by looking at two charts of another provider.

In the case of small practices they looked at 100% of the charts. You cannot tell what percentage of the charts were examined in the large practice, but if you figure that some of the practices had 10 providers (max physicians 8 max PAs 4) then that would equal a panel of about 20,000 patients. If you apply the standard 7% incidence of diabetes in the United States (having worked in that particular area it is closer to 11%) you end up with 1400 patients with diabetes. This equates to 1.4% of all the charts with diabetes that are looked at (charitably underestimating the incidence of diabetes and panel size). Given that they did not try to include all the providers there is a good chance that at least one provider was not included in the larger practices.

Actually the more I look at this the more that problems that I find. Another problem is the way that they looked for diabetes in the practice. They used the coding. However in six practices they found less than 20 charts with diabetes. Consider a solo practice with 2000 patients should have 140 patients with diabetes (the 95% CI is actually around 34 to 210). In all likelihood either the practice is either not diagnosing these patients or more likely not coding for diabetes. It does depend on the overall age of the practice and the underlying health of the population, but I have a hard time believing any practice is cannot find 20 diabetics in the practice. Most likely the practice is not coding diabetes in these patients. Usually this is a limitation of the coding environment (medicare only allows 4 ICD-9 codes per patient encounter). This probably means that patients with the most comorbidities are not in this study (the population who have the most sequelae of DM).

Finally the last issue I have with this paper is it is unclear which guidelines the paper used. They reference the 2001 guidelines but the publication does not seem to exist on the publishers web site. Looking at the 2003 guidelines several things are apparent.

1. The A1C guidelines. Per the 2003 guidelines A1C should be done every 3 months while initiating treatment. The recommendations also say that the A1c should be done twice a year (which is not the same as every six months) but that the frequency is dependent on the clinical situation and clinician judgment. The endocrinologist who I did my IM rotation with did q year A1C in this area because that was what Medicare would pay for in 2001.

2. Adults should be screened for lipids every year. In adults with low risk lipid values repeat lipid assessment every 2 years.

3. Microalbumin - screen for microalbumin in Type 1 patients with duration of greater than 5 years and all type 2 patients annually.

So the three areas of process that show a difference between NPs and PA (also NPs and Physicians) all require excess testing to reach the process goals. These are based on strict application of ADA recommendations (in some cases inappropriate application of the ADA recommendations) which are not necessarily used in clinical practice. That this is occurring is further suggested by the disconnect between outcomes and processes.

I would calculate the sample size at around 1900 (represents 5% of presumed diabetics assuming a panel size of 2000 and a 7% incidence rate). This assumes proper coding which I am not seeing evidence of.

The other thing that I am struck by is the lack of risk assessment by those practices that employ NPs and the lack of a diabetes registry.

Hopefully this helps.

David Carpenter, PA-C