Simple risk-based surveillance with differential sensitivity - calculation of sample size with two sensitivity groups
This page calculates the sample size for simple risk-based surveillance with one factor affecting test sensitivity. For example, a survey in which a high risk sub-population is preferentially targeted and where two levels of test sensitivity are possible.
This analysis assumes that there is no clustering of disease (for instance, we are working at the herd level), and that the effective specificity of the surveillance system is equal to one (all positives are followed up to ensure that they are not false positives):
One risk factor is considered, for which the following information is required:
- The relative risk: this measures the risk of animals (or herds) being infected in the high-risk group, relative to the risk of animals (herds) being infected in the low-risk group. For risk-based surveillance, this should usually be greater than 1;
- The population proportion: this is the proportion of herds/animals from the entire population that are in the high-risk group; and
- The surveillance proportion: this is the target proportion of herds/animals from the surveillance that are in the high-risk group.
In addition, the following parameters are required:
- The design prevalence: this is the assumed prevalence of disease, if the disease is present in the population. It is used as a standard by which the sensitivity of the surveillance can be evaluated;
- The individual unit (herd or animal) test sensitivity for both high-sensitivity and low-sensitivity groups;
- The target proportion of herds/animals in the high-risk group to be tested with the higher sensitivity test;
- The target proportion of herds/animals in the low-risk group to be tested with the higher sensitivity test; and
- The target surveillance sensitivity: the probability that the surveillance system would detect at least one infected animal if disease was present at the specified design prevalence.
- The required sample size for the surveillance system, for high-sensitivity and low-sensitivity groups in each risk group and totals;
- For comparison, the total sample size if representative sampling were used;
- The percentage reduction in sample size achieved by using risk-based sampling; and
- The effective probability of infection (EPI) for the high-risk and low-risk groups. EPI values approaching 100% suggest that, based on the values used for relative risk, population proportions and design prevalence, close to 100% of herds (or animals) in the high-risk group are expected to be infected. If this is unreasonable you may need to review the input values. Values over 100% mean that the model is invalid and processing will be stopped, with an error message. Input values must be changed to ensure EPI values are appropriate.