Mark Thomas, DVM, Dipl. ABVP, and Daryl Nydam, DVM, PhD, say there are three fundamental questions you have to know before setting out to do herd-based sampling:

1. Are you estimating a mean (e.g. the prevalence in a population or things where average is a good representation  —  like urine pH for DCAD cows; it can be too high or too low)?
2. Are you estimating a proportion (e.g. things where it is only bad if they are too low or too high  —  like rumen pH or BHBA, respectively)?
3. Are you estimating if a disease is present or not in the population?

Thomas and Nydam address question 2 in the context of sampling for ketosis. When selecting groups of cows for sampling, many veterinarians follow the suggestion to sample 12 cows. Is this an adequate number to make a management decision? A few questions must first be asked:

• What is the expected prevalence of disease? For ketosis, we might say 15 percent. If it is higher we will still find it. If it is lower, than it is not likely a significant concern.
• What level of confidence do we need to make the decision? Most scientific studies report a 95 percent confidence interval. We must understand that biological or clinical significance is different from statistical significance. The risk of missing a diagnosis of ketosis in a herd can be costly. Therefore a 75 percent confidence interval may be acceptable. Remember that to gain confidence we need to sample more cows.
• How much error will we accept? In other words, what level of biological or technical variation is acceptable? Epidemiologists generally agree on a 5–10 percent variation.
• What is the population of interest? For a disease like ketosis, the population at risk would be fresh cows.

Using a theoretical 600-cow herd with 35 cows less than 20 days in milk (population at risk), if the suspected level for ketosis in this herd is 15 percent of fresh cows (+/- 10 percent error) and the suggested 12 cows are sampled, we will be about 75 percent confident that our sample represents the entire population at risk. Research suggests that ketosis is a herd-based problem if  2–3 or more cows (3⁄12= 25 percent) test positive for the disease.