Annual numbers of live births and pregnancies

Annual number of live births were based on the annual live births projection (2017 World Population Prospects, 2015–2020 estimates and 2015–2020 projections in the medium (i.e. intermediate-growth) variant [18]), adding to these country-specific annual stillbirths occurring from 28 or more weeks of gestation in each year within 2011 to 2015 previously estimated for 195 WHO member states [19]. For 2016 we used each country’s 2015 stillbirths estimate. For non-WHO member states, we assumed stillbirths to be 1.9% of live births, the median value over 2012–2015 across WHO member states [19].

Maternal syphilis prevalence

Country estimates of maternal syphilis prevalence were based on data extracted from the Spectrum-STI adult prevalence database [20–22] developed by Avenir Health with support from WHO. The database synthesizes earlier multi-country databases and country data reported to the WHO, UNAIDS and UNICEF, through GAM. The vast majority of adult prevalence data comes from sentinel surveys and routine testing in pregnant women in ANC. In addition, there are a few data points from community studies and surveys among non-ANC (pregnant and/or non-pregnant) women, or adult men that were thought to be representative of the general adult population in the country [17,23]. For a subset of countries data from blood donor screening and published reports [24–34] were also included. When information was stratified by type of donor, we used the data from first-time donors only, since repeat donors may have a relatively lower syphilis risk, having already been screened for not having had blood-transmittable infections.

Prevalence data were standardized before trend fitting to ensure that they reflected active syphilis, defined as concurrent positivity on both a non-treponemal (e.g., Rapid Plasma Reagin (RPR) or Venereal Disease Research Laboratory (VDRL) test) and a treponemal test [35], as in previous WHO estimations [15,36]. Each data point was assigned a weight according to its national coverage and representativeness, and we assumed no systematic prevalence differences between the data types included [17].

Maternal syphilis prevalence was estimated using one of five approaches, depending on data availability:

1. A. Countries with at least one study either in ANC women (sentinel survey or routine syphilis screening in ANC, with a sample size of >100) or in the general population from 2011 or later AND 3 or more syphilis prevalence measurements in ANC women or general adult populations from 2000 and 2017: Spectrum-STI model was used to generate a prevalence trend estimate for pregnant women. Data were fitted with smoothed-splines polynomial regression [17], using second order segmented polynomials. Both the number and positions of the knots were estimated, and the Akaike Information Criterion was used for model selection [17].
2. B. Countries with one or two ANC data points from 2011 or later but fewer than three data points from 2000 to 2017: the average of the available data within 2011–2017 was used as the national estimate and assumed to be time constant for 2011–2016.
3. C. Countries with no ANC or general population data from 2011–2017 but with one or more syphilis blood donor screening data from 2011 or later, and where including the post-2010 blood donor data meant that there were three or more measurements over 2000–2017: Spectrum-STI model used to generate a prevalence trend estimate, pooling pre-2011 ANC and general population data with the 2011–2017 blood donor data [17].
4. D. Countries with no ANC or general population data from 2011–2017, and one or two blood donor screening data points from 2011–2017 and where even after including the post-2010 blood donor data there were fewer than three data points from 2000: a simple average of the prevalence in blood donors in any years reported within 2011–2017 was used as the national estimate and assumed to be time constant.
5. E. Countries with no ANC, general population or blood donor data from 2011 or later: the value was the imputed median prevalence for all of the countries in the same WHO region that had an estimate based on national data—changing over time as the WHO regional median.

For countries where Spectrum-STI was used to generate maternal estimates (A and C) 95% confidence intervals (CI) were generated using the bootstrap resampling method, with methods and results described in detail before [17,20]. For each country, 400 samples were drawn and, at each iteration, both surveys’ observed prevalence and key parameters were re-sampled. The CIs obtained in this manner reflect stochastic variations in data based on sample sizes, uncertainties in diagnostic test adjustments, and the statistical model. For the other countries we assumed a 95% CI ranging from 50% of the point estimate as lower-bound, to 100%+75% of the point estimate as upper-bound, which was the median width of the 95% CIs across countries in group A.

Service coverage

Based on data from a variety of sources covering the period 2008 to 2016, we developed a database of the three key service indicators: national ANC1 coverage, the proportion of pregnant women attending ANC screened for syphilis, and the proportion of pregnant women diagnosed with syphilis during ANC who were adequately treated. Data sources included: GAM data sets and other national programmatic and surveillance data sent in response to GAM data cleaning, national survey results, data provided during Spectrum-STI country workshops [35–38], and other published and unpublished national program reports, and regional WHO reports [10,11]. Country data review was conducted in collaboration with WHO regional STI advisors to complete the country coverage database and validate final data used.

National ANC1 coverage estimates were based on national reproductive and health surveys, collated in the DHS STAT compiler (https://www.statcompiler.com/en/) and the UNICEF global database (https://data.unicef.org/topic/maternal-health/antenatal-care/).

For all three indicators national estimates for each year between 2012 and 2016 were based on all of the identified data. Country data were reviewed by two of the authors (ELK and JR) for coherence across the indicators (e.g., number of women treated was smaller than the number diagnosed during ANC) and over time. Data provided from the WHO regional advisors or directly from countries and data from CS elimination reports took priority over the GAM data, which in turn took priority over other data sets.

For all three indicators we assigned the identified data to the year it was collected and then estimated the missing years using the closest year with data, with later years taking precedence over earlier years. For those countries with no eligible data point, an estimate was used (imputed). For high income countries, we used the mean value for each year of the high-income countries with data. For non-high-income countries, WHO region specific averages from countries with data were used. The regional means were generated for each year based on those countries with data, with each country weighted equally.

95% CIs on national coverages were calculated as (100%) ±25% of the proportion not ANC-covered, and (100%) ±50% of the proportions of pregnant women not attending ANC screened for syphilis, and those women diagnosed with syphilis during ANC but not adequately treated. ANC1 attendance and syphilis screening coverage in ANC were assumed to be independent of a woman’s syphilis status.

Congenital syphilis and adverse birth outcomes

The WHO CS definition [13] was used to estimate total CS cases. This definition includes all syphilis-related ABOs in infants born to treated and untreated women as well as infants born alive, without clinical signs, to untreated syphilis-infected women (termed non-clinical cases). The risk probabilities of the various ABOs in untreated syphilis-infected mothers were assumed to be the same as in the previous WHO estimates and were based on a meta-analysis of various types of birth outcomes of pregnant women with syphilis. For pregnant women with untreated syphilis the risk probability of an ABO was 52%. ABOs included early fetal deaths or stillbirths (21%), preterm (born alive before 37 weeks’ gestation) or low birthweight (weighing <2500g; 6%), neonatal death (death in first 28 days; 9%), and clinical disease in infants (29–365 days; 16%) [1].

For women whose syphilis infection was adequately treated during pregnancy (defined as at least one dose of 2.4 million units of benzathine penicillin [4]), the WHO’s 2008 and 2012 estimates assumed that treatment reduced early fetal deaths or stillbirths by 82%, preterm or low birthweights by 65%, neonatal deaths by 80%, and clinical disease in infants by 97% [37]. This resulted in an estimated total risk of any syphilis-attributable ABO among treated mothers of 8% [14,15], fixed across countries. In the current estimates, the risk of developing an ABO in treated women was refined to take into account country-specific average timing of ANC enrollment, used as a proxy for ANC-based syphilis screening and treatment [38,39], implying that a country with early ANC enrolment has lower ABO rates than a country with late ANC enrolment. For countries with data on the proportion of pregnancies each year with early ANC attendance this was translated into an average time for first ANC visit and used as a proxy for the time of syphilis treatment among mothers diagnosed during ANC (S1 File). For countries without survey data, the timing of ANC1 was extrapolated and a time-varying ABO risk function in line with recent meta-analyses of ABO risk variation by pregnancy trimester [38,39] was applied to obtain a distribution of ABO risks across countries, with regional medians of 6.8% (range, 3.7–10.4%) in the African Region, 4.4% (2.7–6.8%) in the Region of the Americas, 5.5% (3.5–12.5%) in the Eastern Mediterranean Region, 3.4% (2.3–6.0%) in the European Region, 6.0% (2.9–7.9%) in the South-East Asia Region and 4.7% (2.3–12.2%) in the Western Pacific Region (S1 File).

Confidence intervals (95% CIs) for year-specific, national CS and ABO point estimates were generated by combining the variances of the key country-level input data—Maternal syphilis prevalence, ANC1 coverage, testing coverage and treatment coverage—and assuming that 95% CIs on the ABO probabilities were ±25% of the point estimate for both untreated and for treated mothers within a country. Uncertainties in annual numbers of pregnancies and live births were not considered. When generating regional and global uncertainty bounds we assumed the uncertainties were independent across countries. For translating uncertainties in individual parameters into uncertainties in outcome values we used the delta method (‘error formula method’) i.e. we linearized the formula for the outcome around the best value for the individual parameters and then applied the standard method of calculating variances of linear combinations of (assumed mutually uncorrelated) variables.

Time trend evaluation

The absolute difference between 2012 and 2016 estimates was used to indicate the 5-year time trend. We were not able to determine p-values of trends in CS outcome, owing to a lack of data on how uncertainties in coverage parameters correlate across countries and across years. Making the assumption of independence across years did not appear to be realistic. For example, countries overreporting ANC attendance rates in one year are likely to do the same in another year. Information about relevant correlations was hard to come by unfortunately. We expect trends of sufficient magnitude to be of public health relevance, however.

Ethics statement

No individual patient records were used in this study; we exclusively used aggregated data about numbers of patients, without any individual-patient information. All of these data had already been published or officially reported before, by national STI and/or maternal and child health programs and/or in scientific publications, this was all secondary material. Hence, no ethics review or approval was applicable

Maternal syphilis prevalence

The Spectrum syphilis database as of 24 August 2018 contained over 1,400 data points collected post-1990 from 186 countries [17].

Countries were stratified into five groups based on the type and amount of data available (S2 File). Of all 205 countries, 134 had one or more ANC or general population data points from 2011 or later. Of these, 133 had 3 or more data points post-2000, and Spectrum-STI was used to generate maternal prevalence trends (group A). For one country the average of its available 2011–2017 ANC-based data points was used (B). Another 46 countries had no ANC data from 2011 or later and blood donor data was included in the analysis. Of these, 34 countries had sufficient data for a national Spectrum-STI trend estimate (C), and for 12 countries we used the average prevalence from their eligible 2011–2017 data points (D). For another 24 countries with no data, apart from the USA, prevalence trends were imputed as the regional estimate (E). For the USA, we used data provided by the US Centers for Disease Control and Prevention; maternal prevalence was assumed to be 0.14% based on the 0.17% adult prevalence of active syphilis measured in national household surveys over 2001–2004 [40], and held constant over 2012 to 2016.

The estimated global prevalence of maternal syphilis was 0.70% (95% confidence interval: 0.63–0.77%) in 2012 and 0.69% (0.57–0.81%) in 2016 (Table 1). Prevalence was also stable within most regions, except for non-significant increases in the Region of the Americas from 0.64% (0.56–0.71%) to 0.86% (0.76–1.03%) and in the Region of the Eastern Mediterranean from 0.69% (0.57–0.82%) to 0.77% (0.46–1.07%), and a non-significant decrease in the South-East Asian Region from 0.32% (0.18–0.47%) to 0.21% (0.10–0.32%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Estimated maternal and congenital syphilis and ABO cases and associated ANC service coverages, by WHO region.

Annual country estimates of maternal syphilis prevalence, service coverage and CS case rates were weighted by the number of pregnancies in the country to generate regional and global estimates.

https://doi.org/10.1371/journal.pone.0211720.t001

Between 2012 and 2016, the total number of pregnancies in the world increased from 142.4 million to 143.2 million. This increase combined with the stable maternal syphilis prevalence resulted in a slight decrease in the number of pregnant women with active syphilis: from 1.00 (0.89–1.23) million in 2012 to 0.99 (0.81–1.41) million in 2016 (Table 1).

Service coverage

National data on ANC1 coverage, ANC-based syphilis testing coverage and syphilis treatment coverage from 2011 or later were identified for 182, 135 and 121 countries covering 65%, 76% and 75% of global pregnancies in 2016. Overall, 103 countries had data for all three indicators. Between 2012 and 2016 we estimated that global ANC1 attendance improved from 85% to 88%, syphilis testing coverage from 59% to 66% and treatment coverage from 74% to 78%. At the regional level, coverage rates improved or were stable across all regions for all three indicators. Only one region, the European Region, very nearly achieved the median regional service coverage rates required by WHO to achieve EMTCT (all three indicators above 95%), with 97% ANC1 attendance and 94% coverage of both syphilis screening and treatment in ANC.

Congenital syphilis and adverse birth outcomes

The estimated total number of CS cases globally fell from 748,000 (684,000–812,000) in 2012 to 661,000 (538,000–784,000) in 2016, and the CS cases per 100,000 live births fell from 539 (493–585) to 473 (385–571; Table 1). Over the same period ABOs fell from 397,000 (361,000–432,000) to 355,000 (290,000–419,000) and non-clinical CS cases fell from 351,000 (321,000–381,000) to 306,000 (249,000–363,000).

The majority of cases of CS were in the African Region—which accounted for 62% and 61% of total CS cases in 2012 and 2016, respectively (Table 1). In four of the regions the number of ABOs and non-clinical CS cases fell. The two exceptions were the Eastern Mediterranean Region and the Region of the Americas. The largest decline in terms of absolute numbers was in the African Region (65,000) but the greatest proportionate decline was in the South-East Asia Region (42% drop). The estimated number of CS cases fell in 133 of the 205 countries.

The CS case rate was highest in the African Region (Table 1 and Fig 1), and lowest in the European Region, reflecting variations in the respective regions’ maternal prevalence and service coverages (Table 1). Across the 205 countries, in 2012 and 2016, 84 and 85 countries, respectively, had less than 50 CS cases per 100,000 live births, the WHO criterion for certifying EMTCT of syphilis. Countries with low CS case rates were mostly in the European and Western Pacific Regions; countries with high CS case rates in the African and Eastern Mediterranean Regions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Distribution of CS case rates per 100,000 live births, by type, WHO region and calendar year.

AFR = WHO African Region; AMR = WHO Region of the Americas; EMR = WHO Eastern Mediterranean Region; EUR = WHO European Region; SEAR = WHO South-East Asia Region; WPR = WHO Western Pacific Region.

https://doi.org/10.1371/journal.pone.0211720.g001

Restricting the analysis to the 100 countries that had a Spectrum-based maternal prevalence trend estimate as well as national data for all three ANC service coverage indicators, the estimated CS case rate in 2012 was 557 per 100,000 live births and 427 in 2016. This 23% decline is greater than that the CS case rate decline for all 205 countries (12%).

Early fetal deaths or stillbirths accounted for 143,000 of the estimated 355,000 global ABOs in 2016. Clinical disease in infants accounted for another 109,000, neonatal deaths for 14,000 and pre-maturity/LBW for 41,000 (Fig 1).

Of the estimated global ABO cases in 2016, 74,000 (21%) were in mothers not enrolled in ANC, 203,000 (57%) in mothers who attended ANC but were not screened, 55,000 (16%) in mothers who had been screened but not treated and 23,000 (6%) in women screened and treated (due to treatment failures or reinfection during pregnancy). Comparable figures for 2012 were 89,000 (22%), 237,000 (60%), 55,000 (14%) and 16,000 (4%; Fig 2). The 42,000 drop in the number of ABOs between 2016 and 2012 was a composite effect of a drop in ABO cases among both women not enrolled in ANC (14,500) and in women who attended ANC but were not screened or screened but not treated (34,000), and an increase in ABOs occurring to women who had been treated (6,500) reflecting improvements in screening and treatment coverage, especially in the African Region.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Estimated ABO cases in 2012 and 2016, by access to ANC care and treatment, WHO region and year.

AFR = WHO African Region; AMR = WHO Region of the Americas; EMR = WHO Eastern Mediterranean Region; EUR = WHO European Region; SEAR = WHO South-East Asia Region; WPR = WHO Western Pacific Region.

https://doi.org/10.1371/journal.pone.0211720.g002

Sensitivity analysis

We conducted a series of univariate sensitivity analyses, focusing on the sensitivity of results to assumptions and imputations made for countries and indicators with limited or missing data (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Analyses of the sensitivity of estimates to uncertainties in country data.

https://doi.org/10.1371/journal.pone.0211720.t002

Restricting the analysis to the 100 countries with a Spectrum-based maternal prevalence trend estimate and national data for all three ANC service coverage indicators the estimated CS case rate fell from 557 per 100,000 live births in 2012 to 427 in 2016. This decline is greater than the decline for the complete data set of 205 countries (23% vs 12%).

Changing the maternal prevalence estimation approach for countries in Groups B & D from time-constant based on national data), to time-varying based on regionally imputed trends resulted in a negligible increase in maternal prevalence and resulting CS and ABO numbers.

In the best estimates, countries missing service coverage data were assigned the regionally imputed value for that indicator and year. For ANC-1 coverage replacing the imputed data by the lowest ANC-1 coverage reported by any country for that year led to a negligible increase in CS and ABO cases and rates and replacing them by the highest reported value resulted in a small decrease in CS and ABO cases.

Replacing screening coverage or treatment coverage with the lowest or highest recorded values resulted in more marked changes to CS and ABO estimates. In all cases, however, the estimated global decline in CS case rates and numbers was maintained; across all analyses this ranged from 9% to 24% for both the CS rate and number.