Visual inspection with acetic acid as a cervical cancer test: accuracy validated using latent class analysis

Table 4 Comparative values: conventional versus LCA model results

Reference standard	Disease prevalence (± SE)	Test positive	Sensitivity (± SE) (*)	Specificity (± SE)
1. Colposcopy/Biopsy LGSIL+	0.233 (0.009)	VIA Abnormal, CA	0.640 (0.022)	0.671 (0.012)
		Pap LGSIL +	0.298 (0.021)	0.923 (0.007)
		HPV >= 1.0 RLU	0.649 (0.022)	0.638 (0.012)
2. Colposcopy/Biopsy HGSIL+	0.096 (0.006)	VIA Abnormal, CA	0.775 (0.030)	0.639 (0.011)
		Pap LGSIL +	0.445 (0.035)	0.905 (0.007)
		HPV >= 1.0 RLU	0.800 (0.028)	0.613 (0.011)
3. LCA disease derived from Trichotomous ^(†) VIA, Pap, HPV, including colposcopy/biopsy: 2 class solution	0.302 (0.028)	VIA Abnormal, CA	0.637	0.701
		Pap LGSIL +	0.417	0.997
		HPV >= 1.0 RLU	0.973	0.807
		Colposcopy/Biopsy LGSIL+ ^(‡)	0.442	0.857
		Colposcopy/Biopsy HGSIL+	0.251	0.970
4. LCA disease derived from Trichotomous ^(†) VIA, Pap, HPV, including colposcopy/biopsy: 3 class solution	0.118 ^(§) (0.022)	VIA Abnormal, CA	0.744	0.568
		Pap LGSIL+	0.560	0.820
		HPV >= 1.0 RLU	0.972	0.568
		Colposcopy/Biopsy LGSIL+	0.926	0.758
		Colposcopy/Biopsy HGSIL+	0.632	0.860

^(*) Standard errors (SE) are not given for the LCA-derived sensitivity and specificity estimates because some of these estimates are themselves calculated from combinations of conditional probabilities, which have individual maximum-likelihood estimated (MLE) standard errors. The MLE standard errors for these component conditional probabilities are given in Figures 1 and 2, in parentheses.
^(†) Trichotomous response codes are used to define the latent classes, but results are re-interpreted dichotomously in order to calculate sensitivity and specificity estimates.
^(‡) Results are given for Colposcopy/Biopsy with a low threshold (LGSIL+ is test positive, as it would be for Pap smears), and also with a high threshold (HGSIL+) as the latter is the threshold at which the gold standard for VIA testing is usually set when using the traditional formula.
^(§) Prevalence of class 3 (disease) in a 3-class model; all other LCA prevalence estimates are from 2-class models.

ISSN: 1471-2288