Table 2.
Prior distributions derived from previous literature for the disease prevalence and performance characteristics of the tests
| Models | Prior distribution | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Prevalence | Leukoconcentration | TBS-50/direct blood examination | |||||||
| Sensitivity | Specificity | PPV | NPV | Sensitivity | Specificity | PPV | NPV | ||
| I | Beta(4.61,48.21) | Beta(52.21,0.33) | Beta(202.31,30.25) | Beta(84.49,29.75) | Beta(122.37,0.33) | Beta(15.72,8.02) | Beta(63.65,0.33) | Beta(10.73,0.33) | Beta(80.46,6.36) |
| II | Beta(1,1) | Beta(1,1)T(1-sp[1])a | Beta(1,1) | Beta(1,1) | Beta(1,1) | Beta(1,1)T(1-sp[1]) | Beta(1,1) | Beta(1,1) | Beta(1,1) |
| III | Beta(4.61,48.21) | Beta(1,1)T(1-sp[1]) | Beta(1,1) | Beta(1,1) | Beta(1,1) | Beta(1,1)T(1-sp[1]) | Beta(1,1) | Beta(1,1) | Beta(1,1) |
PPV positive predictive value, NPV negative predictive value
aT(1-sp[1]) = technical term used in JAGS (Just Another Gibbs Sampler) software when defining the formula of the Hui-Walter BLCA model. This addition to the default formula is a safeguard that would have an effect in scenarios when JAGS estimates the Se/Sp to be 0%. This additional “T(1-sp[1])” component of the prior then forces JAGS to consider this Se/Sp as 100% instead, as 0% Se/Sp = 100% Se/Sp after simply switching the labeling of the test results. For the models used in this paper, inclusion/exclusion of “T(1-sp[1])” had no effect on the estimates of the test performance characteristics, meaning that this safeguard remained unused
To calculate priors for the thick smear test, data from Noireau and Apembe [28] were used. Se: 66.7% (49.8–80.9); Sp: 100.0% (97.7–100.0); PPV: 100.0% (86.8–100.0); NPV: 93.0% (87.6–96.0). For the leukoconcentration, data from Bouyou-Akotet et al. [36] were used: Se: 100% (97.2–100.0); Sp: 87.1% (83.2–90.5); PPV: 74.1% (67.0–80.5); NPV: 100.0% (98.8–100.0)