We found that a Widal titer of ≥1:80 was the optimal indicator of typhoid fever in our study population. The PPV, NPV and specificity in the primary analysis was more-or-less unchanged from the cut-off titers of ≥1:80 to ≥1:320, whereas the sensitivity was highest at a cut-off titer of ≥1:80. Although the Widal test at this cut-off titer performed relatively well in terms of sensitivity, specificity and NPV, its PPV was low. It has been argued that PPV is the most important measure of a clinical diagnostic method since it represents the proportion of patients with positive test results that are correctly diagnosed . The PPV is not intrinsic to the test; it is affected by prevalence of the disease. In our setting, where 16 (1%) out of 1,680 febrile patients admitted to the pediatric ward had culture-proven typhoid fever, a negative Widal test result would have a good predictive value for the absence of disease but a positive result would have a low predictive value for typhoid fever, making the use of the Widal test in our setting questionable.
In a previous paper describing the clinical aspects of the children included in this study  older age and long duration of fever were predictive of typhoid fever in this group.
There are several difficulties associated with evaluation of the Widal test. Firstly, levels of agglutinins detectable in the non-infected populations of different areas vary considerably by time and place depending on the endemicity of the disease, which affects test performance. For example, the sensitivity and specificity of a Widal test anti-TO titer of 1:80 in Kolkata, India was 58% and 85%  compared to our findings of 69% and 98%. Secondly, test performance is also affected by cross-reacting infections. In our study, none of the 113 children with non- Salmonella bacteremia exhibited titers above 1:80 for both O and H, although cross-reactions with Klebsiella spp. and Staphylococcus aureus  have been reported. In contrast, 7 (14.3%) of the 49 children with NTS had titers above 1:80 for both O and H. There is also the possibility of cross-reactivity with non-bacterial infections such as malaria, dengue, hepatitis A, and infectious mononucleosis [2, 9, 16]. The third limitation is the choice of a satisfactory gold standard for diagnosis. We used blood culture-positive patients as our true positives. Although bone marrow culture would be the ideal gold standard, this test is difficult to perform in small rural hospitals in Africa. We found that 26 (1.7%) of 1,502 children from whom pathogenic bacteria were not isolated showed agglutination at 1:80 or higher, both for O and H antigens. These may be Widal false positive results due to cross-reaction. Alternatively, since the reported sensitivity of a single blood culture is only 40% to 60% [16–19], some of these are likely to be false negative blood culture results. The final, and what we found to be the most contentious issue, is the selection of the most appropriate control group. It is difficult to choose patients with febrile illness who are blood culture-negative and who definitely do not have typhoid fever. Furthermore, there were relatively few hospitalized children with no bacteremia in the same age range as those with typhoid fever. Thus, the control children were significantly younger than the cases. For our primary analysis, we used groups 2, 3 and 4 (i.e., all children admitted for a febrile illness who were subsequently culture-negative for S. typhi). These would be the most conservative controls for specificity since blood culture picks up only a fraction of typhoid cases, resulting in a control group that is likely contaminated with culture-negative typhoid cases. Despite this, the specificity of the Widal test was high. Using the more exclusive control groups as others had done previously [9–11] did not appreciably alter the sensitivity, specificity, and NPV but they increased the PPV.
The previous studies included in our review (Table 5) had not been performed in Africa hence different cut - off titers were applied, and the resulting sensitivity, specificity, PPV and NPV varied considerably. PPV as well as NPV are dependent on the prevalence of disease within the group of participants; the selection process of study participants has therefore direct influence on the results. The difficulty of choosing the correct control group has been noted earlier . While the gold standard, blood culture, is applied in most studies, the true negatives may be defined as febrile patients with a non-typhi laboratory-confirmed diagnosis as done by Parry et al. and Olsen et al. [9, 20]. Alternatively, some studies use healthy controls. Choo et al . considered all febrile cases with an S.typhi negative blood culture as the control group which is problematic as a number of blood culture-negative results are likely to be false-negative due to the poor sensitivity of the blood culture [17–19, 22]. Furthermore, it is difficult to compare the different test kits, as varying antigens perform differently .