Cross-sectional studies are commonly used in veterinary epidemiological research—particularly to measure infection prevalence and compare infection prevalence between certain subpopulations (for example, between animals under different management conditions).
Calculating the sample size required for statistically robust results is an important part of designing such a study. A variety of free sample size calculators, including those in Ausvet’s Epitools, are available to do all the maths—you just need to choose the right calculators and guide them with a few inputs!
Key considerations in sample size calculation, with links to appropriate calculators, are summarised in this flowchart:
Calculating the required sample size is an iterative process, and should always involve repeated calculations with inputs varied to their plausible limits. This will leave you with a range of required sample sizes to consider. The optimal sample size for your study will be informed by considering sample size calculations for statistically robust results, alongside logistics and budgetary considerations. This is discussed further at the end of the blog.
Example of the use of one of the Epitools sample size calculators:
Let’s explore the use of one of the Epitools listed in the flow chart above: “Sample size to estimate a true prevalence with an imperfect test”.
This tool requires you to enter five inputs:
This Epitools calculator will then provide a sample size requirement for the given input values, and two tables that indicate how the sample size requirement varies with a degree of variation in those inputs:
There’s no need to panic that the sample size requirement is high! There are options to reduce this requirement, as indicated by the tables provided in the output below the sample size estimate. These include:
- using a more accurate diagnostic test in the study (higher sensitivity and/or specificity)
- for example, we can see from the output tables (below)- if we can use a diagnostic test that is 90% sensitive and 95% specific, the sample size requirement drops to 249
- accepting less precision and/or a lower level of confidence in the prevalence estimate
- for example, as per the tables (below), if we are willing to accept a level of 0.1 for precision in the prevalence estimate, the sample size requirement drops to 144
Sample size requirements: it’s about optimising sample size to make your study work
The conservative approach to choosing the sample size for your study may be to use the largest sample size estimate obtained from the calculations.
However, in the real world, we need to consider logistical and budgetary factors such as:
- the practical ability to collect a certain number of samples
- the cost of obtaining each sample, and
- the cost of available diagnostic tests per sample.
For example, a pragmatic approach where samples are exceptionally difficult or expensive to obtain may be to use a sample size at the lower end of the range, identifying through the sample size calculation process how that may affect precision in prevalence estimates and statistical power in comparisons (if relevant).
Alternatively, where samples are relatively cheap and easy to obtain, it may be more efficient to use a cheaper diagnostic test of lower sensitivity and/or specificity, and in doing so accept the relative increase in the required sample size, rather than use a highly accurate but substantially more expensive diagnostic test on fewer samples.
But my study isn’t a cross-sectional study of infection prevalence!
Do not fear! There are other Epitools to address sample size calculation for different types of observational studies (such as cohort and case-control studies) and different types of outcomes (such as numeric outcomes). They are available online (http://epitools.ausvet.com.au/content.php?page=SampleSize) and may be the subject of future blog posts!