Latent class evaluation of three serological tests for the diagnosis of human brucellosis in Bangladesh

Data on the true prevalence of brucellosis, characteristics of three serological tests in livestock farmers, and PP patients from Bangladesh are provided.

A Bayesian latent class evaluation was used to estimate the true prevalence of brucellosis in livestock farmers, PP patients, and at the same time to evaluate three conditionally dependent serological tests. Bangladesh has to be considered to be endemic for brucellosis but with a very low prevalence in animals and humans [19]. In areas of low endemicity, the risk for human infection originates either from consumption of non-pasteurized dairy products or occupation threatening veterinarians, abattoir workers, farmers, and laboratory personnel. In this study, it was possible to estimate the true prevalence for livestock farmers. Sample sizes for other occupational groups were too small to do so, and the method of collection was also non-random. This is another limitation of this study. However, livestock farmers are a promising study group as almost 85 % of rural households own animals and 75 % of the population rely to some extent on livestock for their livelihood [20, 21]. The true prevalence for this group was estimated to be 1.1 %. Brucellosis is a pyrexic disease. As such, it was of interest to investigate also PP patients due to the assumption that brucellosis may be regularly ignored or misdiagnosed. If so, the number of pyrexic patients infected with brucellosis is considered to be valuable information not only for family physicians but also for policy makers. In this study, we focused on PP patients because these patients take antipyretic drugs and antibiotics inappropriate for brucellosis, and see doctors only if recovery does not occur. Among PP patients, 1.7 % were found to be positive for brucellosis which confirms our assumption that brucellosis is ignored or misdiagnosed by physicians in Bangladesh.

Both in livestock farmers and PP patients, the performance of all three serological tests was similar. RBT does not need sophisticated infrastructure or extensive training; it is amazingly cheap and fast. For the Bangladesh setting, RBT is the test of choice. For some endemic countries, authors reported specificity problems of the RBT [22, 23]. In order to overcome this specificity problem, Diaz et al. [24] recommended a modified protocol, i.e., predilution of serum 1:4. Interestingly, we found almost the same performance for the RBT as described by Diaz et al. [24] but without any modification. If the prevalence of a disease is very low as it is in Bangladesh, there will be lower positive and higher negative predictive values for the tests [25]. We have also observed lower positive predictive values of the serologic tests. The highest positive predictive value of RBT in PP patients was 42.7 % indicating that 42.7 % test positive patients truly have the disease and the remaining are falsely positive. The positive predictive value may be increased by applying a second test with high specificity and/or by testing patients having history of exposure with known risk factors like contact with animals, consumption of raw milk, and/or having some symptoms like pyrexia, arthralgia, backache, etc.

Anti-Brucella antibodies, especially IgG, can persist for a longer period of time, i.e., several months even after recovery from disease [26]. For that reason, the presence of anti-Brucella antibodies cannot reflect the true disease status as described above. Thus, diagnosis should be confirmed in a sero-positive patient by the presence of at least one of the clinical symptoms and signs suggestive of brucellosis like pyrexia, arthralgia, headache, backache, hepatomegally, splenomegally, etc. [23, 27]. Applying a more specific test genus or species-specific real time PCR may also be performed [28] to avoid unjustified costs, drug toxicity, and masking of other potentially dangerous diseases like tuberculosis, which are also endemic in Bangladesh.

For a quantitative test, the sensitivity or specificity depends largely on the cut-off value chosen and other factors like endemicity, status and duration of infection, persistence of antibody titres after treatment, presence of cross-reacting pathogens etc. [25]. The cut-off value of the iELISA (?20U/ml) used in our study seems to be appropriate to avoid false positives as its specificity was very high ranging from 99.3 to 99.6 %. WHO and OIE provide guidelines for STAT and RBT standardization, but not for the iELISAs. So, the iELISA test kits provided by different companies are not standardized and it is difficult to compare the results of different studies due to different cutoffs used. In general, a “new” cutoff should be determined under local conditions to avoid false positives.

Like many other authors, we have considered a STAT titre of 1:160 as positive [17, 22]. As already mentioned earlier, in regions where brucellosis is endemic, a large proportion of the population may have persistent Brucella-specific antibody titres. In this scenario, some authors recommend to use STAT titres of 1:320 or higher to avoid false positives [28, 29]. However, in our study, a STAT titre of 1:160 seems to be appropriate as this titre resulted in specificity ranging from 98.2 to 98.8 % indicating good fit for our setting.

The Bayesian latent class evaluation of diagnostic tests requires an assessment of variations in the prior information on the estimated parameters using a sensitivity analysis [30]. Our sensitivity analysis indicated that the use of diffused priors had no relevant influence on the estimated prevalence and test sensitivities and specificities.