In the future, electronic medical records will hopefully make it easy to “mine” for quality data on all patients. Currently available computer systems often make queries of this type very difficult and time-consuming. This leads many people to the mistaken impression that measurement is not possible.
One technique for dealing with problematic data sources is to use sampling -- measure a limited number and extrapolate to the whole population. The nice thing about QI is that, unlike research, you don’t necessarily have to do this in a randomized, controlled fashion. You can use a convenience sample -- whatever sample is relatively easy for you to get your hands on.
For instance, a convenience sample can be:
- A shelf in the file room
- A day’s schedule
- A single payer group (e.g., Medicaid), especially if data is more readily available
When using a convenience sample, you always have to keep in mind that it is NOT a random sample, and may or may not reflect the whole group. In the above list, for example, the characteristics of Medicaid patients may be different enough from patients with commercial insurance that your measures may be significantly different for some things.
Suppose you wanted to assess the timeliness of medication administration on a hospital ward, but the Medication Administration Record (MAR) system can’t handle a query of “what percentage of med's were administered within 30 minutes of scheduled time?” There are too many patients and too many medications to manually review them all. What might you do? What might be the limitations of these samples?