Data-quality issues and alternative variable-screening methods in a questionnaire-based study on subclinical Salmonella enterica infection in Danish pig herds.

Our aim was to determine risk factors for subclinical Salmonella enterica infection in Danish finishing-pig herds. In this paper, the evaluation, combining and initial reduction of variables is presented, along with assessment of the hypotheses in the preliminary statistical testing. The first group of herds was selected at random with no former knowledge of S. enterica infection. Both the herd prevalence and the within-herd prevalence among these herds turned out to be low; hence, some additional herds were selected from The Danish Salmonella Control program, based on their high seroprevalence. This resulted in a hybrid case-"control" design of the study and therefore, five different methods of categorising the data were used to ensure that variables were not wrongfully excluded as a result of using an improper design. Our questionnaire focused on management, infection-limiting precautions and feed and feeding procedures. To establish the prevalence of S. enterica infection within herds at the time of the visit, 50 blood samples from each herd were collected and serologically examined. The reliability of each variable from the questionnaire was assessed and it was decided which variables should be selected, disregarded, combined with other variables and/or recoded. In the simple statistical testing (2x2 tables, cut-off: P=0.25) herds were defined as subclinically S. enterica infected if the within-herd proportion of individual pigs with OD%>10 was more than 20%. The results included questionnaires from 96 randomly selected and 39 high-seroprevalence herds and 6814 blood samples. The initial 95 variables originally included in the questionnaires were reduced to 21 by critical check, combination, recoding and preliminary screening. We failed to demonstrate "herd size" as a risk factor for subclinical S. enterica infection in pig herds.