Quality control using a multilevel logistic model for the Danish pig Salmonella surveillance antibody-ELISA programme

In Denmark, the level of Salmonella infection in pig herds is monitored with a surveillance programme using an indirect antibody ELISA. Our purpose with the present study was to determine whether sample results from the programme were useful in the quality control of this ELISA. Test results from the year 2003, in which the laboratory experienced a technical problem with an automatic microtitre-plate washing machine, were examined statistically. We chose 3 months for the analysis: January, where the problem was moderate, June with the problem more serious, and November, where the problem had been solved. A logistic analysis was carried out with outcome 0 for a negative test result and 1 for a positive test result. Row and column on the microtitre plates, multiprobe robot, and their interactions were included as fixed effects, and date, plate, and slaughterhouse were included as random effects. Backward elimination was carried out using alpha=0.05 to achieve a final model for each month. The row and the column were significant in January and June, and a robot effect was also included in the model for January. In June, an interaction between row and column was identified. In November, none of the fixed effects was significant. Breaking the months January and June into shorter time intervals showed that the row and column effects were significant also when data were from only 1 week, whereas the robot main effect was not significant in most periods and the interaction effects were not significant throughout. Analysis of the test results from the wells with test samples gave good information on systematic errors across the microtitre plates, and severe errors appeared significant even when data from short time periods were used.