Table 2.
Predictors of assay Sn. and Sp. estimated with mixed-effect beta regression (N = 192).
| Fixed Effects | Sensitivity | Specificity | ||||||
|---|---|---|---|---|---|---|---|---|
| Difference in Performance against Manufacturer Value a | Absolute Performance Value b | p c | Difference in Performance against Manufacturer Value b | Absolute Performance Value a | p c | |||
| [95% CI] | [95% CI] | [95% CI] | [95% CI] | |||||
| Source of Evaluation | ||||||||
| Manufacturer | ref. | 93.6% | <0.001 | * | ref. | 98.5% | <0.001 | * |
| [90.6, 95.7%] | [97.8, 99.0%] | |||||||
| Independent | 3.3% | 90.3% | 0.001 | * | 0.2% | 98.3% | 0.247 | |
| [2.7, 3.4%] | [87.8, 92.3%] | [−0.1, 0.4%] | [97.8, 98.7%] | |||||
| Third Party’s Lab | 1.0% | 92.6% | 0.289 | 0.9% | 97.6% | <0.001 | * | |
| [0.1, 1.4%] | [90.5, 94.3%] | [0.9, 0.9%] | [96.9, 98.2%] | |||||
| NRL | −2.2% | 95.8% | 0.207 | * | 4.2% | 94.4% | <0.001 | * |
| [−2.3, −1.8%] | [92.9, 97.5%] | [2.7, 6.4%] | [91.3, 96.4%] | |||||
| US FDA | −2.2% | 95.8% | 0.038 | * | 0.4% | 98.1% | 0.047 | * |
| [−3.6, −1.3%] | [94.2, 97.0%] | [0.4, 0.4%] | [97.3, 98.6%] | |||||
| FIND Diagnostic | 18.6% | 75.0% | <0.001 | * | 0.9% | 97.6% | 0.008 | * |
| [14.6, 22.8%] | [67.8, 81.1%] | [0.6, 1.3%] | [96.4, 98.4%] | |||||
| Netherland CIDC | −0.2% | 93.8% | 0.825 | 0.5% | 98.0% | 0.060 | ||
| [−0.3, 0.0%] | [90.5, 96.0%] | [0.4, 0.7%] | [97.0, 98.7%] | |||||
| Doherty | 2.7% | 90.9% | 0.055 | 0.8% | 97.7% | 0.037 | * | |
| [1.5, 4.5%] | [86.1, 94.1%] | [0.4, 1.5%] | [96.2, 98.6%] | |||||
Note: Assay features of isotype, test type, antibody targets, multiplex detection, and time to result were fitted as covariates to adjust outputs. a Coefficients derived from the model were in log-odds formats, which were then converted to percentages for absolute performance value estimates. b Difference in performance matrix against manufacturer value was estimated using bootstrapping with 10,000 iterations. c corresponds to the significant level of p < 0.05; p attaches to absolute performance values. The mixed beta regression model bounded between 0–100%, clustering assay performance data in serological assays groups. * corresponds to the significant level of p < 0.05.