U.S. National Library of Medicine

Introduction to Health Services Research : A Self-Study Course

Module 5: Quality Filtering and Evidence-Based Medicine and Health (page 6 of 15)

Sampling

In most health services research questions, the investigators need data from huge groups of people, such as Medicare recipients, patients with myocardial infarctions (MI), or all post-menopausal women at risk for breast cancer. Their recommendations for reimbursement rates, post-MI surgical options, or breast cancer screening apply to millions of people. Yet, there is no easy way to include all the eligible people in one study to answer their questions.

Samples

Therefore, researchers choose a sample of people to study. This sample provides answers that are applied to the larger group in which everyone is really interested, the population. Any results from this sample will include differences from the "true" results if everyone was studied.

One goal of any study is to reduce this variation since large differences between the sample and the population make the study results less accurate and less applicable to groups outside the study.

Sample Size and Reducing Variation

Researchers have many ways to reduce this variation. One way is to guarantee an appropriate number of people in the sample, the sample size. There are statistical methods to determine this based on the results researchers hope to find and the amount of variation they expect in the population. For example, if they want to see a small difference in blood pressure between groups of people taking two drugs, they need a lot of people to study!

The selection of study participants also influences how representative the sample will be compared to the large group of people in the population. If researchers choose people based on chance (random sampling), it is more likely these people will be representative of the population.

For example, if they study the outcomes of surgery, they would randomly take people scheduled for surgery on different days rather than only taking a sample of patients scheduled on Monday mornings. It could be that surgeries on Monday mornings involve a different group of staff and residents than surgeries on other days. Randomly selecting surgical days from the whole week reduces the chance of skewed results in the study.

Determining the best sample size and composition is critical in health research and is difficult to evaluate. It may help to look for descriptions of random sampling in the Methods section of an article. The following describes how researchers investigating cigarette smoking in China randomly selected villages to survey:

We used a two-stage, stratified cluster method. First, 20 sample villages/urban units were randomly selected from the 280 villages in Minhang. Second, 52 sample neighborhoods were randomly selected from the 208 neighborhoods... (This) process gave every household and resident in Minhang an opportunity of being samples... (Gong, 1995)

Discussion Questions

1. From the reading above it sounds as if it is difficult to select a sample size. Which resources would you recommend to a student writing a Master's thesis in health services research involving a large population?
2. Do you routinely examine the Methods section of an article to determine the sample size? How important to quality filtering is the Methods section of an article? How do other researchers use the Methods section in their analysis of a paper?
3. Discuss why researchers strive for a representative sample. What happens to the quality of their data if their sample is too small?
4. What are the tradeoffs between cost and sample size?