We made three cardinal observations. First, for detection of precancerous lesions, there exists substantial intra-rater and inter-rater variation in the histopathological evaluation. Our results suggest that histopathology may be useful as a diagnostic test in demonstrably high degree of dysplasia or frank neoplasia but its value as a reference standard for diagnosis of low-risk precancerous lesions is questionable. Consequently, the use of histopathology as a reference standard against light-based assistance for diagnosis of high-risk lesions may lead to biased estimates of the diagnostic performance of these measures.
Second, we observed widely differing estimates of the sensitivity and specificity of the studied diagnostic protocols. However, caution needs to be exercised when reading and interpreting the results of latent class modeling [24–27]. A substantially different estimate of sensitivity (or specificity) for a test from that for the other tests can result from two scenarios: a) if the test is diagnostically inferior as compared to the rest; and b) if the test is using different criteria for classification of the disease state. In our case, the results do not necessarily imply that TBLU and CHTB are diagnostically superior to histopathology - rather it is possible that these tests use totally different criteria that do not compare with those used by histopathology. Nevertheless, our results clearly demonstrate (Table 2) that one of the main reasons for the controversial estimates of the diagnostic performance of light-based aids may be the classification method employed for the reference standard.
Third, a comparison of the diagnostic performance of TBLU and CHTB consistently indicated that use of CHEM may be somewhat redundant. From a primary health care perspective this finding is important since it will reduce the cost of diagnostic evaluation considerably by restricting the use of the more expensive component. Indeed the estimates of sensitivity and specificity of TBLU observed in this study are comparable with or better than those of other more expensive protocols like autofluorescence [28, 29], photodynamic diagnosis , and chemiluminescence . Our results are in agreement with the findings of Epstein et al which show that toluidine blue retention test holds promise as a screening tool for high-risk oral precancerous lesions since it can reduce a large number of unnecessary biopsies . Concurring with other studies [33, 34], our results encourage consideration of TBLU as a viable and feasible screening method in high-prevalence and low-resource scenarios like India.
There are important limitations of this study. First, as with the Mehrotra et al  study, our study recruited patients with a suspicion of a precancerous lesion for the reasons of feasibility as observed elsewhere . However, the protocol did preclude visually negative patients that could have been later detected by at least one of the diagnostic methods. Our estimates of high sensitivity may also partially reflect this spectrum bias thereby limiting a ready generalization of the results. Second, the study sample had an a priori high likelihood of a precancerous lesion. Therefore our study design does not permit a full evaluation of the screening performance of these tests but rather considers them in the more practical scenario of a tertiary care setting as a diagnostic aid.
In summary, our findings support those of Mehrotra et al  and demonstrate that improvements are needed for histopathological evaluation of precancerous lesions - especially, low risk lesions. Our findings also suggest that toluidine blue retention may be considered as a diagnostic strategy for oral cancers in countries like India. More robust and larger studies are required to assertively and definitively answer questions related to the screening use of these tools in high prevalence settings.