Chlamydia and gonorrhoea in pregnant Batswana women: time to discard the syndromic approach?

Urine was checked on site with a dipstick; all other specimens were analysed at the National Health Laboratory in Gaborone. Cervical swabs were obtained for ligase chain reaction (LCR) amplification technology for detection of C trachomatis and N gonorrhoeae. The swabs were placed in LCx^® transport media, transported to the laboratory the same day, and stored at -20°C prior to batch processing. The LCx^® Assays (Abbott Laboratories, IL) were performed according to the manufacturer's instructions. A case of C trachomatis or N gonorrhoeae infection was defined as an individual with a positive LCR analysis.

Gram-stained vaginal smears were scored for bacterial vaginosis according to Nugent's criteria [20]. Culture of Candida species was initiated by direct inoculation of a high vaginal swab on Saboraud plates on site, and Gram-stained smears and wet-mounts from high vaginal swabs were examined for budding yeast cells and pseudohyphae. A cervical smear was gram-stained to count polymorphonuclear leukocytes per high power field (PMN/HPF).

Data were analysed with the statistical package SPSS Version 11. We performed univariate analyses on all independent variables in our dataset which could be associated with cervical infection. The factors were within four categories: sociodemographic and behavioural factors, symptoms, clinical signs, and microscopy results. The dependent variable "cervicitis" was defined as infection with C trachomatis or N gonorrhoeae, or both. Factors which in the univariate analysis were associated with infection at a 0.2 level (p-value of odds ratio (OR)) were included in multiple logistic regression analyses. We performed multiple logistic regression analyses on sociodemographic factors, symptoms, clinical signs and laboratory results, separately and combined. Due to the high number of variables, the full analysis of different factors' independent association with cervicitis had to be performed in two steps. First, multiple logistic regression analyses for the four variable categories were performed separately. The analysis was then extended to combining the factors from the different categories which were significantly independently associated with cervical infection in the first step.

Three levels of risk scores were computed and retrospectively applied, and each of them were assessed for their usefulness as a diagnostic tool to manage N gonorrhoeae and C trachomatis infections in the antenatal care. The first risk score level consists of only sociodemographic factors. At the second risk score level we added findings from the gynaecological examination. At the third level, we also allowed the results from microscopy of vaginal and cervical smears. Microscopy can be done on site, and thus all three risk score levels – the sociodemographic, the clinical and the microscopy score – can theoretically be performed during an antenatal care visit. The weights for each factor used in the risk scores were based on correspondent multiple logistic regression analysis models. The log of the OR for the variables in the model multiplied by 10 and rounded to the nearest whole number were used as weights for the respective factors in the risk scores [9]. The factors included and their respective weights are shown for each risk score in Table 1.

Receiver Operating Characteristic (ROC) curves were used to compare the different risk scores. The ROC curve shows the sensitivity and specificity that correspond to each possible cut-off for the risk score. The validity of different diagnostic strategies (the existing STI management guidelines, diagnosis based on symptoms or signs alone, and diagnosis based on a risk score) was assessed by measuring sensitivity, specificity, positive and negative likelihood ratios (LR+ and LR-) and positive and negative predictive values (PPV and NPV). The LCR-based laboratory diagnosis of cervical infection was used as the reference standard. In the evaluation and comparison of diagnostic strategies, we present two cut-offs for the risk scores, taken at a sensitivity of minimum 0.40 and 0.70.

The study was approved by ethical committees in Botswana and in Norway.

In spite of the high prevalence of cervical and vaginal infections in this study, few women confirmed symptoms when probed. Complaints of vaginal discharge were elicited from 119 (17%) of the women and lower abdominal pain from 58 (8%). Vaginal discharge was the most common clinical sign; candidalike discharge was found in 81 (12%) and other abnormal discharge was found in 227 (32%) of the women.

Demographic, behavioural, and obstetric factors; symptoms, signs, and simple laboratory tests; and their association with C trachomatis and/or N gonorrhoeae infection are shown in Table 2. Age was the strongest predictor of cervical infection. The prevalence of infection was highest among teenagers (22%, 95% confidence interval (CI): 13 to 32), whereas two-thirds of the infections were within the largest age group: women 20–29 years. None of the evaluated symptoms predicted cervical infection. Symptoms of vaginal discharge or lower abdominal pain were therefore not included in any of the risk scores. There were 65 (9%) women who reported believing that they had a genital illness, but the prevalence of cervicitis was not significantly higher in this group (OR = 1.6, 95% CI: 0.8 to 3.4). Of clinical findings, vaginal (excluding candida-like) discharge was significantly associated with increased prevalence of cervicitis (OR = 1.8, 95% CI: 1.1 to 3.0). Several discharge characteristics (thin, smelly, foamy and increased amounts) were also significantly associated with cervical infection. Two laboratory results were associated with cervicitis; white blood cells in the cervical smear, and infection with T vaginalis.

The factors which were independently associated with cervical infection are shown in Table 3.

The sociodemographic, clinical, and microscopy-based risk scores performed similar in the management of N gonorrhoeae and C trachomatis infection, as illustrated with their ROC-curves (Figure 1). The diagnostic accuracy of the risk scores did not increase significantly when detailed information from the clinical examination and subsequently the microscopy results were added to the sociodemographic risk factors.

Table 4 presents the evaluated options for the diagnosis of C trachomatis and N gonorrhoeae in antenatal care in the absence of specific diagnostic tests: the syndromic algorithm, symptoms, signs, and the three risk scores. The VDS algorithm did not identify women with C trachomatis and N gonorrhoeae in our study population (positive likelihood ratio (LR+) 1.1, 95% CI: 0.6 to 1.9). Symptoms of vaginal discharge or lower abdominal pain in pregnancy proved inappropriate as an entry point to the VDS algorithm, with a LR+ of 0.94 (95% CI: 0.6–1.5). The current practice of clinical screening for signs of vaginal discharge (excluding candida-like discharge) also had low discriminative ability (LR+ 1.5, 95 % CI 1.1–1.9).

All risk scores suffered from the choice between low sensitivity and low specificity. With a cut-off taken at an acceptable sensitivity of minimum 0.7, the risk score based on sociodemographic factors identifies 50 (75%) and misses 17 (25%) of the 67 cervical infections. Per true case treated, six pregnant women would be prescribed multiple antibiotic regimens unnecessarily. With a cut-off at a sensitivity of 0.4, the sociodemographic risk score identifies only 29 (43%) of the cervical infections. The majority of the cervical infections would remain untreated, and although over-treatment is reduced, four pregnant women would still be unnecessary treated with multiple antibiotics per infected women. The more comprehensive clinical and microscopy risk scores show similar results.

Elicited symptoms of increased discharge or lower abdominal pain are not predictive of cervical infection in antenatal attendees in Botswana. These symptoms are non-specific; they are common in pregnancy, and their association with cervical infections is even lower among pregnant than among non-pregnant women [23]. The VDS algorithm is used extensively to diagnose cervical infections in antenatal care in Botswana [24], but our results show that this management is no better than random treatment.

To use the genital examination at the first antenatal care visit as a clinical screening tool for cervical infections is also unadvisable. According to the newly revised VDS algorithm in Botswana, symptoms or signs of yellow discharge are risk factors which should lead to treatment for cervical infections. In our study among antenatal care attendees, neither symptoms nor signs of yellow discharge are associated with cervical infection. In pregnancy, symptoms or signs of vaginal discharge in general, yellow discharge, or other specific discharge characteristics should not be used as criteria for treating C trachomatis and N gonorrhoeae.

Other studies from sub-Saharan Africa conclude that case management with the vaginal discharge syndrome has poor discriminatory ability in the diagnosis of cervicitis [25‐27]. Several reviewers [28‐30] emphasise that the syndromic approach should not be used as a screening tool for N gonorrhoeae and C trachomatis. Our study concurs with a substantial body of knowledge indicating that the syndromic approach should be used neither as a case management of symptomatic women nor as a clinical screening tool to identify C trachomatis and N gonorrhoeae in antenatal care attendees in sub-Saharan Africa.

Despite extensive analyses, all computed risk scores were of limited value as screening tools in antenatal care attendees. They had poor discriminative ability, even in the study population in which they were computed and adapted to under optimal conditions. As the syndromic approach, the risk scores also resulted in a large number of undetected cervical infections and substantial over-treatment. Additionally, notification and treatment of sexual partners, an essential element of STI management, is difficult to justify when the majority of the identified women do not have an STI [8]. A substantial improvement of the management of cervical infections in antenatal care in developing countries seems impossible without specific diagnostic tests.

The development of simple, rapid tests for C trachomatis and N gonorrhoeae has been a high priority since the 1990s [9‐11]. Major progress has recently been made, and several tests for C trachomatis and N gonorrhoeae are now on the market [7, 32]. The Sexually Transmitted Diseases Diagnostics Initiative at the WHO has begun a programme to field test and systematically evaluate these simple, affordable rapid tests [32, 33]. So far, available tests are found to be specific (>90%), but with a variable sensitivity (25–85%) [34‐36]. In Botswana, the health system is relatively well functioning, and this country could well serve as an exploratory site for the use of rapid tests for C trachomatis and N gonorrhoeae. Through the prevention of mother-to-child transmission of HIV programme, all health posts and clinics have lay workers who perform rapid tests for HIV. Simple tests for cervical infections performed by clinicians or lay workers may prove a feasible contribution to the improvement of diagnosis and the reduction of the disease burden of these conditions in this and similar settings.

Consistent with established knowledge on STI epidemiology [6, 31], youth is the single factor most strongly associated with C trachomatis and/or N gonorrhoeae in our study population. Thus age can be useful as a screening tool in the traditional sense, to minimise the number of standard diagnostic tests by identifying people with a higher-than-average prevalence of infection [29]. If it were decided in Botswana to introduce screening for cervical infections with rapid tests in the antenatal care, selective screening of younger women should be considered.