Prevalence data were identified from studies conducted between 1985 and 2017 in populations representative of the general adult (15–49 year-old) population in South Africa (S1 File) [11,12]. Eligible study populations included pregnant women in ANC care (routine screening or surveys), women attending family planning clinics, and women and men sampled in household surveys and other community-based studies. For women, all data from ANC and non-ANC women were pooled, assuming no systematic prevalence differences between them, as in the 2012 WHO methodology [19] and supported by a recent global meta-analysis of risk factors for adult syphilis [20].

Data were adjusted for diagnostic test performance to reflect ‘active’ syphilis, defined as concurrent positivity on both non-treponemal Rapid Plasma Reagin (RPR) and treponemal tests (e.g. the Treponema pallidum haemagglutination assay (TPHA)[21], using the same assumptions as other Spectrum estimates [22]. Prevalence data from studies that used both a treponemal and non-treponemal test were not adjusted. Prevalence values for studies using rapid treponemal tests were multiplied by 0.70, studies using only treponemal or only non-treponemal tests without confirmatory test were multiplied by 0.53, and studies where the diagnostic test was unknown by 0.75. Each adjustment multiplier was assumed to have a standard error of ±25%. In addition, each syphilis data point data increased by 10% to account for under-sampling of high-risk populations [19], an adjustment in keeping with key group prevalences from Integrated Bio-Behavioural Surveys in South Africa [9,23,24], and estimated population sizes [9]. No adjustments were made for age or location. The adjusted data points were each assigned a weight to reflect national coverage and representativeness. Nationally representative data were assigned a weight of 100% whilst data representative of smaller areas, or from a subset of sentinel surveillance sites were weighted accordingly [22] (S1 File).

The Spectrum-STI model fitted the prevalence data by segmented polynomials regression [25,26] applying a second-order spline curve. Both the number and positions of the knots were estimated, and the Akaike Information Criterion was used for model selection’ [22]. Estimated prevalence was constrained to be within 0–20% [22]. Uncertainty bounds were generated using bootstrapping, with 10,000 replications [22].

Annual incidence was derived by dividing the estimated prevalence by the average duration of infection [19] using the 2012 WHO average durations of infection values for countries with low and intermediate treatment access (S1 File). In the default scenario the proportion of symptomatic episodes treated was set at 35% (the WHO estimates for countries with low treatment access [19]) between 1990 and 1997 and then assumed to increase linearly to 60% in 2006 and to remain constant after this. The increase between 1997 and 2006 was included to reflecting the roll-out of syndromic management [2,12,13] in the mid-1990s [27].

Prevalence data were identified from studies conducted between 1990 and 2017 in female and male populations representative of the general adult 15–49 year-old population in South Africa (S1 File) [11,12]. Eligible populations included pregnant women attending ANC, women and men attending family planning clinics or PHCs, and adult women and men tested as part of household or community surveys. Eligible diagnostic tests were Nuclear Acid Amplification Tests (e.g. Polymerase Chain Reaction and Ligase Chain Reaction) and culture.

Prevalence data from each study were adjusted for diagnostic test performance [19,28,29], geographical location and age, using the same approach and parameters as the WHO’s 2012 STI estimates [19]. Diagnostic tests were adjusted using fixed sensitivities and specificities, and studies conducted in exclusively rural or urban sites were converted into a national prevalence by applying a rural-to-urban ratio of 0.9, and South Africa’s annual urban and rural population sizes [30]. Each prevalence data point was also increased by 10% to account for the contribution of higher-risk populations [19], an adjustment in keeping with key group prevalences measured in South Africa [24,31], and estimated population sizes [9].

For chlamydia in women, prevalence data were adjusted for age reflecting the decline with age of chlamydia prevalence observed in recent surveys of women in South Africa [32,33], the United Kingdom [34] and the United States of America [35] and among women in South Africa presenting to clinics for vaginal discharge syndrome (VDS) [5,36]. This marked age pattern reflects acquired immunity against reinfection with chlamydia [37]. In order to obtain prevalence estimates for the 15–49 age range, prevalence data from studies that only sampled younger populations (15–24 years of age) prevalence estimates were therefore multiplied by 0.60, and studies of exclusively older populations (25 and over) by 1.39. In the absence of similar data for men, we did not assume an age pattern for chlamydia in men. No adjustments were made for age for gonorrhea in either men or women, lacking clear evidence.

Each data point was assigned a weight to reflect its representativeness. Studies that used NAAT diagnostic tests in populations representative of the overall adult population were assigned a weight of 1.0; all other studies were assigned a weight of 0.10 reflecting uncertainties in diagnostic tests, age adjustors and the representativeness of the population sampled.

Spectrum-STI was used to generate estimates for gonorrhea and chlamydia in women and in men separately. A moving average time trend curve was fitted through the adjusted and weighted prevalence data, as there was insufficient data to use a segmented polynomials regression. Uncertainty bounds were calculated by bootstrapping (10,000 replications)[14].

Trends in annual incidence for both infections were estimated using the same approach as syphilis, and the same default assumptions for treatment coverage. Duration estimates were based on the WHO 2012 assumptions with one change: the probability of a man with chlamydia developing symptoms was set at the value used in South African transmission dynamic models, 0.33 [10,38], rather than the WHO global value of 0.54.

Spectrum-estimated numbers of symptomatic cases of gonorrhea and chlamydia in 2017, and the subsets of these symptomatic cases treated, were compared to the national-level MUS case reports for 2017 collated by the South African District Health Management Information System (S2 File). Based on etiological data on the causes of MUS from South Africa [5,36,39,40] we have assumed 76% of MUS cases are due to gonorrhea and 24% to chlamydia, considering that 5% of MUS cases do not have a gonococcal and/or chlamydial etiology, but that the etiological studies found considerable rates of dual gonococcal plus chlamydia etiologies.

The Spectrum-estimated incident cases in men were multiplied by antibiotic resistance prevalence estimates from the NICD gonococcal antimicrobial resistance surveillance among men with MUS in four provinces [4,5,41,42] to estimate the incident numbers of antibiotic-resistant gonorrhea in men from 2007 to 2017.

Univariate sensitivity analyses looked at relaxing the data inclusion criteria to include blood donors, who were excluded from the default estimation because of their likely non-representativeness of overall populations, as transfusion services often exclude donors with self-reported risks or observed infections (syphilis only); removing the representativeness weight adjustment (all three infections); and assuming no change in treatment coverage over time (all three infections). More general assumptions of the Spectrum-STI methodology have been addressed in sensitivity analyses in earlier publications [14,18,22].

A total of 44 prevalence data points from pregnant women attending ANC were identified from 1985 and later: 5 data points from nation-wide routine ANC screening, and 39 national and sub-national sentinel surveys and prevalence studies (S2 File). In addition, 11 prevalence measurements from non-pregnant women (1990–2017) were also identified that met the study inclusion criteria. Fig 1A shows the adjusted prevalence data and the Spectrum estimation for women.

The estimated prevalence in 2017 was 0.50% (0.32–0.80%) (Table 1), a marked decrease from the 1990 estimate of 11.3% (8.5–16.4%).

Fig 1B shows the corresponding prevalence estimates for men. In men, owing to a lack of data pre-1995 the prevalence from 1990 to 1995 was fixed at the 1995 level. Between 1995 and 2017 prevalence fell from 8.2% (4.1–13.2) to 0.97 (0.19–2.38). These estimates were based on the 8 data points identified in male general populations.

The number of new cases in 2017 (Table 1) was estimated to be 23,175 (9,900–62,500) in women and 475,000 (5,100–173,000) in men. The annual incidence (Fig 1C) mirrored the decline in prevalence, with incidence in 2017 9-fold lower in women and 5-fold lower in men than in 1995. The decline in incidence was less than the decline in prevalence reflecting a decrease in the average duration of infection between 1997 and 2006 owing to our assumption that treatment coverage increased over this period.

Forty-nine eligible prevalence measurements (36 female and 10 male) were identified between 1990 and 2017 (S1 File, Fig 2A and 2B). The data represented both urban and rural locations in all nine provinces.

The Spectrum estimated gonorrhea prevalence in 2017 was 6.6% (95% CI: 3.8–10.8%) in women and 3.5% (1.7–6.1%) in men (Fig 2A and 2B). Prevalence over 1990–2017 was stable in both sexes. The time trend in incidence of new infections mirrored this (Fig 2C), except for a slight increase between 1997 and 2006 in men when treatment coverage was assumed to increase to reflect the roll-out of syndromic treatment.

Spectrum estimated there were 2.3 (1.1–5.0) million and 2.2 (1.1–3.8) million new gonorrhea cases (i.e. episodes) in women and men in 2017 (Table 1). The case incidence rate and number of new cases were similar in men and women, despite the prevalence being higher in women, reflecting the longer average duration of infection in women.

For chlamydia, 46 prevalence data points (35 female and 11 male) were identified from the period 1994 to 2017 from urban and rural locations across all 9 provinces of South Africa.

Chlamydia prevalence in 2017 was estimated at 14.7% (9.9–21%) in women and 6.0% (3.8–10.4%) in men (Fig 3A & 3B). The time trend in prevalence over 1990–2017 was stable for both sexes and the incidence trend mirrored this apart from an increase over 1997 to 2006 when treatment coverage was assumed to increase.

Spectrum estimated that there were 1.9 (1.1–3.8) million new chlamydia cases in women and 3.9 (2.2–6.3) million new infections in men in 2017 (Table 1). The chlamydia case incidence rate and number of new cases were higher in men than in women, despite chlamydia prevalence being higher in women, reflecting the longer average duration of infection in women.

Fig 4A shows the evolution of antibiotic resistance for gonorrhea, based on sentinel surveillance data from 2007 to 2017. In response to antimicrobial resistance trends, South Africa changed its first-line treatment regimen for MUS in 2008 from ciprofloxacin (used since 1993) to oral cefixime, and in 2014/15 from oral cefixime to dual injectable single-dose ceftriaxone & oral azithromycin [2], in accordance with WHO recommendations [43]. Dual antimicrobial cover for gonorrhoea is a strategy designed to curtail the emergence of extensively-drug resistant N. gonorrhoeae that is resistant to the extended-spectrum cephalosporins. This gained additional importance in 2012 with the first reports of cefixime resistance and resultant treatment failure for male urethritis in South Africa in two MSM [42], in recognition of the fact that there is a paucity of gonococcal antimicrobial resistance surveillance data from key population groups in the country.

The observed antibiotic resistance prevalence estimates from four provinces were applied to the population-level male gonorrhea prevalence estimates for 2007 to 2017. In each year the majority of cases were susceptible to the prevailing year’s first-line treatment (Fig 4B). The continued increase in the prevalence of N. gonorrheae resistance to ciprofloxacin after its withdrawal from use in STI syndromic management algorithms may reflect ongoing antimicrobial selection pressure arising from its use for other indications or its continued use by public and private STI treatment providers during shortages or stock-outs of the first line treatment [44]. In addition, there may be a fitness benefit conferred by ciprofloxacin-resistance associated mutations [45].

In 2017, of the 2.2 million Spectrum-estimated gonorrhea cases in adult men, 1.4 million were symptomatic and 850,000 million were treated (Table 2). For chlamydia Spectrum estimated 3.9 million cases in adult men in 2017 of which 1.3 million were symptomatic and 765,000 were treated. Using the gonorrhea estimates and assuming 74% of symptomatic MUS cases are due to gonorrhea the estimated number of treated MUS cases in 2017 was 1.1 million. A similar calculation for chlamydia, assuming 26% of symptomatic cases are due to chlamydia, comes to 3.2 million treated MUS cases.

The actual number of MUS cases reported by the South African STI program from public clinics was 310,921 (S2 File). Comparing the actual MUS cases to the Spectrum estimates of MUS cases treated suggests that case reporting completeness, among treated MUS cases, is somewhere between 10% and 28%.

Table 3 summarizes the results of the sensitivity analyses. Relaxing the study eligibility criteria for syphilis to include national blood donor screening data from 2015–2018 (S1 File) reduced the estimated prevalence of syphilis in women from 0.50% to 0.38% and in men from 0.97% to 0.35%. Including blood donor data had less of an impact on the prevalence in women than men owing to the large number of data points for women from ANC surveys and ANC-based routine screening. Including blood donors markedly reduced the male-to-female prevalence ratio, from almost 1.9:1 to 0.92:1, a figure much closer to the 1:1 ratio found in studies within South Africa [46], across East and Southern Africa [47–49], and elsewhere [20].

Assigning the same weight to all studies increased the prevalence in 2017 of gonorrhea and chlamydia in women from 6.6% to 8.4% and from 14.7% to 17.3% (Table 3) reflecting the prevalence in two studies from 2017 that documented high prevalences for both infections in young women [50] and in a rural area with relatively poor access to health care [51]. For men, the prevalence of gonorrhea decreased from 3.5% to 2.7% and of chlamydia from 6.0% to 5.1% reflecting data from a study in young men [50]. Gonorrhea and chlamydia incidence estimates in men decreased proportionally; relative to the incidence estimates in this scenario, we would have estimated a correspondingly higher reporting completeness of routine national MUS case reporting: 11–35% of symptomatic treated cases, compared to 10–28% in the default estimates. The weighting did not materially change results for syphilis (Table 3).

Changing the treatment coverage assumptions such that treatment coverage had not improved between 1997 and 2006 and instead remained at the 1997 level reduced the estimated number of new cases in men in 2017 from 45,000 to 37,100 for syphilis, from 2.2 million to 1.8 million for gonorrhea, and from 3.8 to 3.4 million for chlamydia. The lower number of new cases in 2017 in this scenario reflects the assumed longer average duration of infection (constant throughout 1990–2017), driven by a lower (constant) treatment coverage throughout 2017.

The Spectrum-STI South African estimates are limited by the quality and representativeness of prevalence data, and by assumptions underlying the model and those made in the absence of data. Whilst there was a sufficient base to generate time trends for the three STIs for both men and women, there is considerable uncertainty for gonorrhea and chlamydia as the data are from studies done in different populations using different study designs. Many of the prevalence data points, especially for gonorrhea and chlamydia, are from research projects and the study population may not be representative of the general population. In addition, the studies are from different age groups and we have only made age adjustments in the case of chlamydia in women and these are crude adjustments. Other limits to the data include the range of different diagnostic tests used and whilst we have corrected for diagnostic sensitivities and specificities the correction factors may not be correct. For example, our prevalence estimates for 2017 may be too high if the studies included for recent years over-sampled higher-risk populations, or for chlamydia younger populations, or if recent studies used diagnostic tests of higher sensitivity than we assumed. The lack of data also precludes any analysis at the sub-national level, which with the current non-systematic data set would risk to confound diversity in demographic, socio-economic and behavioural variations with real geographical patterns.

The present estimates focused on data from the general population and were not informed by prevalence data from high-risk groups. A new version of the Spectrum model is under development that incorporates risk groups and means that the current, crude approach of adding 10% to the general population estimate to reflect these missing populations will no longer be required.

STI surveillance and control in South Africa would clearly benefit from regular, systematic prevalence screening. For surveillance, the ideal would be periodic (e.g. 3-yearly) cluster-sample surveys, in men and women, in selected urban and rural sites. Securing funding for expanded etiological STI screening will be challenging, but there are opportunities to add to existing or planned activities, such as in pregnant women, adolescent girls, young women and other high-risk women, and priority populations such as MSM [60]. In both routine health care settings and through outreach, this would ideally use point-of-care tests (POCTs) with acceptable performance characteristics (affordability, equipment-free testing and delivery of results within 30 minutes), and be accompanied by partner notification and treatment. Of commercially available POCTs for gonorrhoea and chlamydia screening, the Cepheid geneXpert CT/NG (duplex for Neisseria gonorrhoeae and Chlamydia trachomatis) has demonstrated acceptable performance in terms of sensitivity and specificity, including when used on non-invasive specimens such as first-void urine [61]. With the Cepheid Xpert MTB/Rif assay now widely implemented for tuberculosis screening across public health sector laboratories in South Africa, existing geneXpert diagnostic capacity for TB could be leveraged to include screening for asymptomatic STIs in at-risk populations, which would also enable pathogen-directed expedited partner therapy. Additionally, there is a need to reinforce the training of both public and private sector healthcare workers in correct STI syndromic management, through the use of continuing medical education activities and educational material, including visual aids. Studies evaluating STI services at both private and public healthcare facilities in South Africa have revealed that general knowledge regarding STI aetiologies and management and compliance with national treatment guidelines among healthcare workers is suboptimal [44,62], and this compromises quality of care resulting in the prescription of incorrect treatment regimens for STI syndromes.

There is also a need to strengthen case reporting in both the public and private sector if trends in case reports to be a meaningful indicator. For a start, conducting regular, periodic data quality audits at facility-level could assess the quality of existing reporting systems, and determine the degree of under-reporting. Also, MUS could be added to South Africa's Rapid Internal Performance Data Audit (RIPDA) tool [63,64], used by health facilities for internal auditing before the Auditor General evaluates the Department of Health performance by comparing data as found in facility-level source documents with data in the District Health Information System 2. Third, quarterly feedback to provinces and districts about STI indicator performance, could stimulate action. Fourth, printed booklets of the STI surveillance tools will promote consistent and frequent data collection in sentinel sites. Finally, setting case finding targets can be a helpful mechanism to promote consistent case reporting, especially when coupled with period evaluation of actual performance against those targets.

The stable high rates of gonorrhea and chlamydia highlight that syndromic case management is not sufficient to control STIs. One policy change that would make a difference is to remove the current age constraint on STI treatment among women presenting with VDS; at present a woman presenting with VDS is treated for an STI only if she is below 35 years of age, gives a history of a male partner with urethral discharge, or if she is a returning patient and reports no improvement in symptoms [2].

For gonorrhea treatment, the regularly revised treatment guidelines are adequate in view of antimicrobial resistance patterns in public-sector surveillance. However, antimicrobial resistance monitoring clearly remains a priority area–and there may be scope to extend screening for extensively-drug resistant gonorrhea to key populations such as MSM, to expand sampling to include symptomatic and asymptomatic men at both genital and extra-genital sites, and to expand resistance surveillance and prescription patterns in the private sector where ciprofloxacin continues to be prescribed [44].

South Africa’s high STI rates also underscore the national policy to not limit VMMC to HIV-negative men, but to also enable HIV-infected men, after clinical assessment and enrolment into HIV care, to receive VMMC, which may reduce their and their partners’ STI rates [65,66].

The Spectrum-STI estimates, based on STI surveillance and monitoring data, highlight the high burden of STIs in South Africa. Whilst the prevalence and incidence of syphilis in adults has fallen between 1990 and 2017, the prevalence and incidence of gonorrhea and chlamydia, with high burden especially in women who are often asymptomatic, have remained almost unchanged and are high relative to other countries. Both gonorrhea and chlamydia have health consequences for the infected individual and for babies born to infected mothers, and are established co-factors for HIV transmission [67], especially in partnerships where the HIV-infected partner is not on ART. Intensification of actions to prevent, detect and control STIs alongside HIV are important if the burden of STIs is to be reduced and would benefit from strengthened surveillance.