Table 1.
Latent Class Analysis estimates true prevalence of FFV to be 94.8%.
| Model | qPCR | ELISA | Prevalence (%) | DIC | ΔDIC | w | ||
|---|---|---|---|---|---|---|---|---|
| Se (%) | Sp (%) | Se (%) | Sp (%) | |||||
| UIP & Cov | 66.1 ± 4.2 | 42.3 ± 21.4 | 65.4 ± 4.3 | 45.9 ± 21.8 | 94.8 ± 5 | 241.2 | 0 | 1.0 |
| IP & Cov | 74.2 ± 3.3 | 94.6 ± 3.5 | 80.2 ± 5.1 | 95.0 ± 3.2 | 85.0 ± 4.5 | 261.2 | 20 | 0.0 |
| UIP | 83.0 ± 8.5 | 87.5 ± 8.4 | 95.3 ± 3.3 | 67.9 ± 16.5 | 69.0 ± 8.8 | 272.9 | 31.7 | 0.0 |
| IP | 74.5 ± 3.4 | 96.5 ± 2.7 | 94.3 ± 2.2 | 96.3 ± 3 | 77.6 ± 3.8 | 282.2 | 41 | 0.0 |
| Model Average | 66.1 ± 4.2 | 42.3 ± 21.4 | 65.4 ± 4.3 | 45.9 ± 21.8 | 94.8 ± 5 | |||
Average model coefficients ± standard deviation for the sensitivity (Se) and specificity (Sp) of qPCR and ELISA as well as true prevalence of FFV based on Bayesian Latent Class Analysis. Deviance Information Criterion (DIC) values and model weights (w) suggest uninformative priors and covariance between qPCR and ELISA to be the most parsimonious model. UIP = uninformative priors; IP = informative priors; Cov = covariance.