Table 3 Accuracy of new empirical algorithms based on data from rural Haiti to detect cervical infection.

Algorithm	Sensitivity	Specificity
Model 1, cut off >9.7
Training dataset (n = 517)	98.7%	10.1%
Model 1, cut off >9.7
Testing dataset (n = 351)	96.4%	11.5%
Model 1, cut off >12
Training dataset (n = 517)	93.3%	18.1%
Model 1, cut off >12
Testing dataset (n = 351)	89.9%	17.0%
Model 1, cut off >14.6
Training dataset (n = 517)	90.7%	46.4%
Model 1, cut off >14.6
Testing dataset (n = 351)	80.5%	31.9%
Model 2, cut off >2
Training dataset (n = 511)	97.4%	9.0%
Model 2, cut off >2
Testing dataset (n = 353)	95.9%	8.7%
Model 2, cut off >2.8
Training dataset (n = 511)	97.4%	20.5%
Model 2, cut off >2.8
Testing dataset (n = 353)	94.7%	15.8%
Model 2, cut off >5.4
Training dataset (n = 511)	85.5%	43.0%
Model 2, cut off >5.4
Testing dataset (n = 353)	84.7%	24.6%
Model 3, cut off > = 1
Training dataset (n = 541)	98.8%	16.6%
Model 3, cut off > = 1
Testing dataset (n = 383)	95.7%	12.2%
Model 3, cut off >2.3
Training dataset (n = 541)	94.0%	27.3%
Model 3, cut off >2.3
Testing dataset (n = 383)	89.8%	20.9%
Model 3, cut off >3.9
Training dataset (n = 541)	67.5%	60.5%
Model 3, cut off >3.9
Testing dataset (n = 383)	62.6%	54.1%
Model 4, cut off > = 1
Training dataset (n = 544)	94.0%	27.3%
Model 4, cut off > = 1
Testing dataset (n = 385)	89.9%	21.3%
Model 4, cut off >2.4
Training dataset (n = 544)	85.5%	40.6%
Model 4, cut off >2.4
Testing dataset (n = 385)	79.8%	35.0%
Model 4, cut off >3.6
Training dataset (n = 544)	39.8%	84.4%
Model 4, cut off >3.6
Testing dataset (n = 385)	34.6%	72.1%