Literature review > Issue_2 > Review Witt et al. 

Bacterial vaginosis (BV) results from a shift in the normal vaginal flora that is predominated by lactobacilli to one that is predominated by Gardnerella vaginalis and anaerobic Gram negative bacilli and peptostreptococci. The number of lactobacilli in women with BV is not decreased compared to normal women, however, hydrogen peroxide producing lactobacilli predominate in the latter group. Current theory suggests that in some way the hydrogen peroxide lactobacilli are supplanted by non-producing strains, which permits the overgrowth of other organisms. Presenting symptoms of bacterial vaginosis are a thin homogenous skim milk-like discharge, fishy odor and vaginal irritation. However, many women are asymptomatic or, at least, don't recognize their symptoms as being abnormal. It has been shown that BV is associated with pelvic inflammatory disease, postoperative gynecologic surgery infections, and abnormal pregnancy outcome. Moreover, there is mounting evidence that bacterial vaginosis increases host susceptibility to other sexually transmitted diseases, including C. trachomatis, N. gonorrhoeae, and HIV. These associations imply that there may be a need to screen asymptomatic women for BV in order to prevent complications though as yet, there is no direct evidence in support of such a hypothesis.

Diagnosis is based on clinical findings (characteristic discharge, release of a strong fishy odor on mixing vaginal secretions with potassium hydroxide, vaginal pH >4.5, and the presence of clue cells) or on vaginal secretion Gram stains scored according to the Nugent criteria. Both of these definitions are relatively subjective, resulting in an "imperfect gold standard" for use in the evaluation of new BV diagnostic tests. Such tests are needed, as the clinical presentation is not very specific and there are few laboratory technicians trained to read vaginal secretion Gram stains, so this test is generally not available in the clinical setting.

This paper is designed to evaluate the Affirm VP III G. vaginalis DNA hybridization assay (Becton Dickinson, Sparks, MD, USA) as a test for the diagnosis of BV. This system is a DNA hybridization test that is positive only for concentrations of G vaginalis in excess of 2 X 105 bacterial cells and therefore should be positive most often in women with BV and seldom in women with normal vaginal flora. The Affirm system, using different reagents, also can detect the presence Candida spp. and Trichomonas vaginalis in the same specimen, making it quite attractive as a tool for evaluating women with vaginal discharge. The procedure involves three steps: 1) vaginal secretion sample preparation to release organism nucleic acids, 2) automated assay processing, 3) reading of the results. It requires several hands on steps and between 30 and 45 minutes to complete the assay. The test could be done in an office setting but this is not very efficient and it is usually better done in a laboratory setting were specimens can be batched. Nonetheless, the rapid turn-around time permits return of results to clinicians within 24 hours.

This study evaluated 1,725 pregnant women between the 12th and 36th weeks of gestation attending a clinic in Vienna, Austria. All patients had symptoms of vaginal discharge or complications of pregnancy. Separate vaginal swabs were tested by Gram stain and by the DNA hybridization assay. Gram stained secretions were scored according to the Nugent system which was then categorized by the authors as follows: grade 0 = no bacteria on smear, grade 1 = a Nugent score of 1 to 3, grade 2 = 4 to 6 and grade 3 = 7 to 10. Grade 2 corresponds to what is usually referred to in the literature as an intermediate Gram stain while grade 3 corresponds to BV. Gram stain results were then used to construct two different infected patient definitions for the purpose of assessing the performance of the Affirm test.

In the first analysis the authors combined grade 2 and 3 Gram stains in the infected patient definition. Based on this definition the Affirm test's sensitivity, specificity, positive and negative predictive values were as follows: 73.2%, 97.1%, 88.7% and 92.2%. In the second analysis the authors defined the infected patient as a grade 3 Gram stain only, which is the most appropriate definition given the fact that the studies suggesting significant disease associations have all used this definition of BV. However, at this point the authors commit a significant error in their analysis. They dropped all grade 2 patients from the analysis, defining the uninfected patient as having a grade 0 or 1 Gram stain. By so doing the results were quite lovely with the sensitivity rising to 89.5% and the specificity remaining high at 97.1%. However, it is clearly inappropriate to remove a significant number of cases from the data set. Not only is this a violation of accepted statistical data analysis standards but creates a population that is not reflective of the real world i.e. all female populations have a spectrum of vaginal bacterial ecologies ranging from normal to intermediate to BV and this fact cannot be ignored in assessing a new diagnostic test's performance. If an investigator wishes to change the definition of the infected patient by narrowing the infected patient definition, then those cases excluded as infected patients become members of the uninfected patient category. The appropriate analysis would have revealed a sensitivity, specificity, positive and negative predictive value of 89.5%, 88.3%, 45.7% and 98.7% respectively for the Affirm test. The major change between the analysis presented by the authors and this analysis is that the specificity is significantly lower as is the positive predictive value. Indeed, given the negative predictive value of 98.7%, the test is very good at excluding BV as implied in the article's title but the authors do not discuss the benefit of excluding the diagnosis; what clinicians really want to do is to make the diagnosis.

In conclusion, this article illustrates one of the pitfalls of assessing the performance of new molecular based diagnostic tests for infectious diseases i.e. removing a category of cases from the analysis because they do not meet the investigators notion of infected or uninfected patient status. Inappropriate analytic methods may give the impression of better performance characteristics of the evaluated test than are actually the case. In a low prevalence population such as the one studied here, the Affirm test has a low positive predictive value for BV. Among symptomatic women, over-treatment based on such a test result may not be a problem; most of the false positives in this study were in women with grade 2 Gram stains and studies have suggested that about a third of these will go on to develop BV over time. However, should one decide that screening and treating asymptomatic cases is reasonable, basing treatment on a test with a low positive predictive value will result in many women being treated unnecessarily. It is important to note that the data that we have concerning significant clinical disease associations with BV are based on defining BV as Gram stain Nugent score of >6 and that as yet, controlled trials demonstrating benefit of treating such women are not available. Currently, there is no rational for treating women with intermediate or grade 2 Gram stains, especially if they are asymptomatic. Therefore, the Affirm VPIII test for high levels of G. vaginalis in vaginal secretions should be helpful to clinicians managing women with vaginal discharge in whom they suspect BV as the cause but, until there is a strong enough rational for diagnosing and treating BV in asymptomatic women that would justify treatment of a significant number of uninfected women, it should not be used for screening asymptomatic pregnant or non-pregnant women.

Author's comment to expert review (submitted by Dr Armin Witt):

1. We do not think that it is inappropriate to remove grade 2 from the *second* analysis because we described the results of grade 1-3 in the *first* part. The intention was to show the good sensitivity/specificity for patients with a definite BV, but we did not hide other results. And it is very helpful for clinicians to exclude a disease, especially in times when overtreatment is common.

2. Data on the benefit of treating BV in pregnancy are available, the latest publication was published in the LANCET 2003 (Ugwumadu et al.; 361: 983-88).
   

about SDI | newsletters | grants | publications | literature reviews

WHO Home - WHO Search - TDR Home - SDI Home - SDI Contact us
(c) WHO/OMS 2001