Accounting for imperfect diagnostic test results on data from an entire population

[Giovanni Piumatti]

Hello,

I have data from a seroprevalence study which collected samples from an entire population (about 95%) of health workers within the same hospital. I thought about using the logitem package to account for test sensitivity and specificity in a logistic model and obtain adjusted prevalence. Would it still make sense to do it considering that I have the entire population?

Thank you in advance for your help,

Giovanni

[Phil Bromiley]

The entire area of using population data for statistical analysis has been somewhat controversial. Most of our statistics assume sampling, but it seems pointless to reject data you have in order to have a sample. And it is not clear how you think about estimating relationships with population data. It is fine to say that the mean of the population is the true mean, but we normally want to estimate relationships among population characteristics. You also have to decide what the population really is. If you are using this hospital as a sample from the general population of hospitals, then except for sample selection issues there is no problem.

[Giovanni Piumatti]

Dear Phil,

thank you for your answer. Our goal is not to extend the findings from our current population to other hospitals. Would you then say that we can not test any relationship between population characteristics a part from looking at descriptive statistics? If I would still correct prevalence for sensitivity and specificity of the adopted test, for example in a logistic model, would it be correct to include other factors in such a model (e.g. age and gender)?

Giovanni