Evaluation and optimization of the syndromic management of female genital tract infections in Nairobi, Kenya

Two-thirds of the women presenting with LGTS at outpatient clinics in Nairobi had at least one confirmed LGTI, more than one third of the infected had multiple infections, and a majority reported at least 3 episodes in a year. VVC was the most frequent infection (2 of 5 women) while BV prevalence was remarkably low. Symptoms of vaginitis were predominant in these mostly young women, but clinical signs were scanty. Contraceptive use was high; condom use was low, in line with a predominance of reported monogamous relationships. The vaginal discharge syndrome algorithm used in Kenya proved to be insufficient for the management of genital infections, but our proposed alternative achieved only modest improvement.

The frequencies of specific genital infections in our study vary from rates detected previously at similar clinics in Nairobi in which VVC was 6% higher and TV was double, but NG and CT rates were lower [22]. With vaginal discharge and itch being the commonest clinical presentations and coupled with recurrent symptoms, we speculate that a high usage of vaginitis (VVC and BV-TV) treatment with delayed opportunity for STI treatment over the years may explain these variations. Our detection of an infection in two-thirds of patients is similar to proportions in studies from elsewhere in Africa [8, 9], but somewhat lower than studies from India (80%) [7]. For the specific female genital infections, our findings do not concur with other studies from Africa where BV was predominant [2, 8, 9, 17]. These variations are likely due to study population differences. Indeed, we noted associations between patient characteristics and the infections. Influence by study population characteristics such as sexual risk behaviour, level of education, age, condom use, prior use of antimicrobials, and perhaps genetics have been cited as determinants of aetiology in other studies [3, 8, 9, 17, 23].

About one-third of the patients in our study tested negative to the six common LGTI despite being symptomatic. We think that the probability of false negative test results due to prior antimicrobial use is small because we used very sensitive testing methods. We employed PCR for detection of the four STIs; this technique would identify even antimicrobial-suppressed bacteria and TV. For Candida infection we employed 3 techniques i.e., KOH, gram stain and culture, hence the possibility of false negative cases was low. For bacterial vaginosis we used the Nugent score, which is the gold standard.

Although the vaginal discharge syndrome tool has the advantage of providing treatment to patients at an opportune time and without the laboratory-testing-associated delays and costs, we identified concerning discrepancies between the syndromic predictions and actual infections. Hence, we sought to improve the algorithm’s accuracy by determining patient characteristics more predictive of the infections. The symptoms of vulvovaginal itch for VVC and repulsive vaginal discharge for BV-TV were crucial in delineating the vaginitis syndrome to guide specific treatment for VVC and BV-TV, so as to reduce the over-use of metronidazole given the disparate burden of VVC and BV-TV. It is however worth noting that although important for detection of STI, LAP was an infrequent symptom hence contributing to the low sensitivity and PPV by both algorithms.

Our study showed no association between contraceptive use (including hormonal) and STI in general or with specific STI. Although controversial, studies in the past have pointed toward higher likelihood of some STI in clients using hormonal contraceptives. Indeed, a recent systematic review and meta-analysis on studies investigating the influence of hormonal contraceptives on STI reports mixed findings that included no effect, a protective effect, and increased risk [24].

The significant association between symptom recurrence and absence of infection was unexpected, especially because we employed DNA detection for most microbes. We speculate that, in addition to the effects of prior antimicrobial use in almost one quarter our participants, vaginal pathobionts (not tested in our study) could partly explain this; additionally women may not be able to distinguish between physiologic and abnormal leucorrhoea, hence overreport vaginal discharge as has been shown elsewhere [25]. Future studies are necessary, to elucidate this.

A syndromic-only approach can be misleading as a diagnosis and treatment tool. Indeed, we demonstrate here that the vaginal discharge syndrome algorithm used in Kenya is poor at detecting or excluding infections. With the algorithm’s low specificity and PPV, about two-thirds of patients in our study received unnecessary metronidazole and antifungal treatment, while a similar proportion of patients requiring treatment for bacterial STI did not receive it. Both algorithms had low sensitivity and poor PPV scores for STI. Our findings are in line with other studies which have revealed the inadequacies of the syndromic flowcharts in diagnosis and treatment of female genital infections [18, 19, 26]. Our substitute algorithm had advantages over the current algorithm. By avoiding blanket treatment of VVC, BV and TV, our algorithm performed better for BV-TV treatment by lowering the unnecessary use of metronidazole, and somewhat for VVC treatment too. Our substitute algorithm also recognized women without infection leading to less overtreatment.

Vaginal discharge-based syndromic approaches have been shown in the past, and confirmed in this study, to miss common bacterial STI. This however should not motivate for inclusion of bacterial STI treatment to these algorithms. Such a move would result in unnecessary use of antibiotics in three-quarters of STI-free patients, posing the risk of development of antimicrobial resistance. We rather advocate that the savings from such unnecessary antibiotic prescriptions instead be channelled to point-of-care (POC) testing costs for patients triaged to have STI by the syndromic algorithm. A broad interrogation of the syndromic approach, beyond accuracy in treatment allocation, is needed to determine the full value of our proposal.

Being symptom-dependent, the syndromic flowcharts are poor at detecting mixed infections, yet we found this to be common, particularly with STI whose mode of transmission and clinical presentation are shared. Several studies have shown coupling of some infections, especially TV with BV and with bacterial STI [27,28,29]; improved/future algorithms should thus take this into consideration. Moreover, the symptom-dependent approaches do not recognize the existence of asymptomatic infections, yet studies show that up to 80% of patients with TV or BV are asymptomatic [2, 30]; these patients remain unrecognized and not treated in the symptom-dependent algorithms.

Efforts by others elsewhere to improve the syndromic algorithm’s performance have yielded limited improvement. Such attempts included addition of sexual partner risk behavior information, and bedside tests e.g. vaginal swab pH and whiff test [17, 31,32,33]. The problems are that different etiologies share similar clinical characteristics and certain patients lack certain symptoms and signs despite having the disease; additionally, symptoms are largely subjective. For example, vulvovaginal itch is more common in women with VVC, but a large proportion of women with other infections also have it; and foul-smelling vaginal discharge is associated with BV, TV and STI [17, 26, 34, 35]. Hence the algorithm’s performance is limited by indistinguishable behavioral factors, and symptoms and signs. The result of this is a suboptimal sensitivity and specificity. Therefore, only with integration of POC into the algorithms is good discriminative power achievable.

POC testing is feasible and accepted by women [36]. Such tests would be crucial in delineating mixed infections, asymptomatic infections or deciphering infections with shared symptom(s). For example, inclusion of POC pH and biochemical testing, for BV and TV respectively, yields notable improvement in diagnostic accuracy [17]. Additionally, several studies show that real-time PCR testing for STI is very promising with high sensitivity and specificity. These rapid and accurate tests are relatively affordable, making it possible to implement them in resource-limited settings. For such settings, a combination of syndromic triage plus POC testing would be best suited [37,38,39]. However, given the long-standing funding gaps in the public sector in these settings, widespread use of POC is unlikely to be realized in the immediate future. Therefore, while use of the syndromic approach remains the most feasible option, regular review and revision of the algorithms’ performance in line with emerging evidence is vital. Relatedly, it is necessary to rethink the present approaches to algorithm evaluation; they are limited to diagnostic and treatment accuracy. We propose that future evaluations of algorithms be comprehensive and include short-term and long-term opportunity costs determinations, and cost-benefit analyses.

A limitation of our study is that a significant number of patients had multiple infections, and our analyses did not look at the influence of multiple infections on the predictors. However, we were able to interrogate the performance of the conventional vaginal discharge syndrome treatment flowchart using a large dataset. Our sizeable dataset additionally allowed us to subject our alternative algorithm to internal validation. Secondly, our study did not access patients who seek care at private healthcare facilities. We however believe that our study population is representative of women in Nairobi. While it is expected that patients of lower socioeconomic status would seek health care mainly from public health facilities, it has been shown that only one-third of patients from a slum in Nairobi seek care at public health services with a majority utilizing private facilities [40]. Notably, 43% of people in informal settlements in Nairobi have health insurance cover compared to a national proportion of 20% [41, 42]. Moreover, key bio-behavioral characteristics of the participants such as age, marital status etc., do not vary between those who use public and private facilities and therefore the prevalence and type of LGTS is not expected to vary.