Table 1.
Kosack et al. data calculated to illustrate the algorithms’ positive predictive valuea
| Site | HIV prevalence | Algorithm | Sensitivity | Algorithm PPV | |||||
|---|---|---|---|---|---|---|---|---|---|
| Lowest bound of performance based on low confidence interval (MSF data) | Highest bound of performance based on high confidence interval (MSF data) | Point estimate of confidence interval (MSF data) | Manufacturer data | Worst case | Best case | Point estimate of confidence interval (MSF data) | |||
| Guinea, Conakry | 2.7% | Determine | 98.30% | 100% | 100% | 99.9% | 98.89% | 99.98% | 99.89% |
| SD Bioline | 98.30% | 100% | 100% | 100% | |||||
| Uganda, Kitgum | 8.3% | Determine | 98.3% | 100% | 100% | 99.9% | 97.3% | 100% | 100.0% |
| HIV STAT‐PAK | 77.9% | 99.5% | 96.2% | 99.7% | |||||
| Uni‐Gold | 77.9% | 99.5% | 96.2% | 100% | |||||
| Uganda, Arua | 4.9% | Determine | 98.3% | 100% | 100% | 99.9% | 99.1% | 100.0% | 99.9% |
| HIV STAT‐PAK | 98.3% | 100% | 100% | 99.7% | |||||
| Uni‐Gold | 98.3% | 100% | 100% | 100% | |||||
| Kenya, Homa Bay | 26% | Determine | 98.3% | 100% | 100% | 99.9% | 95.1% | 98.9% | 97.7% |
| First Response | 98.3% | 100% | 100% | 99.4% | |||||
| Uni‐Gold | 96.8% | 99.9% | 99.6% | 100% | |||||
| DRC, Baraka | 0.8% | Determine | 98.3% | 100% | 100% | 99.9% | 93.6% | 99.8% | 98.9% |
| Uni‐Gold | 96.8% | 99.9% | 99.6% | 100% | |||||
| Vikia | 96.8% | 99.9% | 99.6% | 99.95% | |||||
a
Estimates for the algorithm assume that test results at each step are independent of those in the prior step; worst case and best case performance estimates were calculated using the lower and upper 95% bounds for each test respectively.