Estimates of the Burden of Group B Streptococcal Disease Worldwide for Pregnant Women, Stillbirths, and Children

We aimed to provide the first comprehensive estimates of the burden of group B Streptococcus (GBS), including invasive disease in pregnant and postpartum women, fetal infection/stillbirth, and infants. Intrapartum antibiotic prophylaxis is the current mainstay of prevention, reducing early-onset infant disease in high-income contexts. Maternal GBS vaccines are in development. Methods. For 2015 live births, we used a compartmental model to estimate (1) exposure to maternal GBS colonization, (2) cases of infant invasive GBS disease, (3) deaths, and (4) disabilities. We applied incidence or prevalence data to estimate cases of maternal and fetal infection/stillbirth, and infants with invasive GBS disease presenting with neonatal encephalopathy. We applied risk ratios to estimate numbers of preterm births attributable to GBS. Uncertainty was also estimated. Results. Worldwide in 2015, we estimated 205 000 (uncertainty range [UR], 101 000-327 000) infants with early-onset disease and 114 000 (UR, 44 000-326 000) with late-onset disease, of whom a minimum of 7000 (UR, 0-19 000) presented with neonatal encephalopathy. There were 90 000 (UR, 36 000-169 000) deaths in infants <3 months age, and, at least 10 000 (UR, 3 000-27 000) children with disability each year. There were 33 000 (UR, 13 000-52 000) cases of invasive GBS disease in pregnant or postpartum women, and 57 000 (UR, 12 000-104 000) fetal infections/stillbirths. Up to 3.5 million preterm births may be attributable to GBS. Africa accounted for 54% of estimated cases and 65% of all fetal/infant deaths. A maternal vaccine with 80% efficacy and 90% coverage could prevent 107 000 (UR, 20 000-198 000) stillbirths and infant deaths. Conclusions. Our conservative estimates suggest that GBS is a leading contributor to adverse maternal and newborn outcomes, with at least 409 000 (UR, 144 000-573 000) maternal/fetal/infant cases and 147 000 (UR, 47 000-273 000) stillbirths and infant deaths annually. An effective GBS vaccine could reduce disease in the mother, the fetus, and the infant. We summarize our methods according to our 4 objectives as follows: We conceptualized the full burden of GBS disease (Figure 2) to include pregnant and postpartum women, fetal infections (based on stillbirths), and infants, as described in the first article in this supplement [15]. We took a compartmental model approach to modeling infant invasive GBS disease, deaths, and disability, with 4 steps as illustrated in Figure 3. For the first step in the model (maternal GBS colonization), the step where most data were available for national prevalence estimation, we also attempted a multivariable regression model to predict national maternal GBS colonization, as an alternative to using a subregional estimate when national-level data were limited (Appendix). We sought data inputs from the published literature through systematic reviews and unpublished sources through research databases and investigators worldwide, as summarized in the previous 10 articles (Figure 1). The specific methods used for each of these (database searches, inclusion and exclusion criteria, data characteristics, criteria used to assess bias and sensitivity analyses) are described in general [15] and reported elsewhere [16–23, 30]. We performed meta-analyses, to obtain estimates of maternal GBS colonization prevalence [16], the ratio of late-onset to early-onset invasive GBS disease [19], case fatality risks (CFRs) [19], proportion of cases with meningitis [19], proportion of infants with GBS meningitis who had moderate to severe neurodevelopmental impairment [20], incidence of maternal GBS disease in pregnant/postpartum women [17], prevalence of GBS disease in stillbirth [18], prevalence of GBS disease in neonatal encephalopathy [22], and the association between maternal GBS colonization and preterm birth [21]. We calculated pooled estimates using random-effects models [31] to allow for heterogeneity across studies by use of a statistical parameter representing the variation between studies. For the first step of the compartmental model, we determined maternal GBS colonization prevalence for countries, subregions (South America, Central America, Caribbean, Western Asia, Southern Asia, South-Eastern Asia, Eastern Asia, Oceania) and regions (Latin America, Asia, Africa, Oceania, developed) as described elsewhere [16], to apply to estimates of live births in 195 countries for 2015, using latest United Nations data [32]. The colonization data were adjusted for sampling site (rectal and/or vaginal) and laboratory culture methods [16]. Where data were considered sufficient (≥1000 mothers tested for rectovaginal colonization), we used an estimate for individual countries. Where data were limited (<1000 mothers tested for rectovaginal colonization), we used a subregional estimate, and where no subregional estimate was available, we used a regional estimate (Supplementary Table 1 for inputs by country). For the second step of the compartmental model, we assessed IAP policies and their implementation in countries as described elsewhere in this supplement [23], and categorized 89 countries with data available into 1 of 4 categories, which were (1) microbiological screening for maternal GBS colonization with IAP and high implementation coverage (>50% of mothers screened and given IAP if appropriate); (2) clinical risk factor approach with IAP given to mothers with risk factors before delivery and high implementation coverage (>50% with risk factors receiving IAP); (3) microbiological screening for maternal GBS colonization with IAP and low implementation coverage (<50%); (4) clinical risk factor approach with IAP given to mothers with risk factors before delivery and low implementation coverage (<50%), or no IAP strategy in place. We assigned countries in the developed region with no data to category 1 as a conservative approach, and of those countries reporting these data, 21 of 31 developed countries were in group 1. We assigned countries, not in the developed region and with no data to group 4, as 51 of 59 countries not in the developed region reporting these data were in this group. We then assessed the risk of EOGBS disease in studies reporting maternal GBS colonization data, and the use of IAP, as described elsewhere in this supplement [30]. We used the linear association between IAP use and risk of EOGBS disease described in [23] to estimate the risk of EOGBS disease in each of the 4 contexts, with specific risks for each group as follows: group 1 = 0.3% (95% CI, .0–.9%); group 2 = 0.6% (95% CI, .10%–1.2%); group 3 = 0.9% (95% CI, .4%–1.5%); group 4 = 1.1% (95% CI, .6%–1.5%). For each country, the number of cases of EOGBS was estimated by multiplying the estimated number of exposed babies by the appropriate risk for that country. We used regional estimates of the ratio of early-onset to late-onset GBS cases [19] to then estimate the number of LOGBS cases. For Oceania, where data were lacking, we applied the estimate for Asia, as the most similar regional context. There were variations in estimates, with the highest ratio in Asia (5.99 [95% CI, 2.40–14.9]) suggesting more EOGBS than LOGBS, and lowest in Africa (1.02 [95% CI, 0.82–1.27]). We give parameters for each region in Table 1. These regional estimates could, however, be affected by low case ascertainment. This could reduce EOGBS disease cases, particularly those with home delivery, inadequate access to care and/or high rapid CFR, and/or late-onset cases, particularly if cerebrospinal fluid sampling is not undertaken, and cases of GBS meningitis are thus not detected. We therefore did a sensitivity analysis applying a worldwide ratio of early-onset to late-onset GBS disease from high-quality studies worldwide (1.11 [95% CI, 0.90–1.30] / 3.92) [19]. Data Inputs to the Compartmental Model to Estimate Cases of Infant Group B Streptococcal Disease, Deaths, and Disability Data in parentheses represent the 95% confidence interval. Abbreviations: EOGBS, early-onset group B Streptococcus; GBS, group B Streptococcus; IAP, intrapartum antibiotic prophylaxis; LOGBS, late-onset group B Streptococcus; NDI, neurodevelopmental impairment. For the third step of the compartmental model, we applied region-specific CFRs to 3 different groups that differ considerably in terms of outcome: EOGBS cases delivered without a skilled birth attendant, EOGBS cases delivered with a skilled birth attendant, and LOGBS cases. Case fatality risk for EOGBS: We applied percentages of skilled birth attendance for each country to EOGBS cases to determine EOGBS cases which would, and would not, have been attended by a skilled birth attendant. We applied a CFR of 0.9 (0.3–1.0) to estimated EOGBS cases born without a skilled birth attendant, based on expert opinion as to the likely high CFR in these “unseen” cases. To estimate deaths from EOGBS born with a skilled birth attendant (and for all developed countries), we estimated regional CFRs for EOGBS from facility-based data, as described elsewhere in this supplement [19]. We applied these regional CFRs to cases of EOGBS disease with skilled birth attendance. The highest CFR for EOGBS with skilled attendance was in Africa (0.27 [0.15–0.37]), then Latin America (0.17 [0.05–0.30]), Asia (0.14 [0.06–0.23]), and developed countries (0.05 [0.04–0.07]) (Table 1). For Oceania, where even regional data were lacking, we applied the risk in Asia, being the most geographically proximal. Case fatality risk for LOGBS: We also estimated regional CFRs for LOGBS from facility-based data, as described elsewhere in this supplement [19]. Regional CFRs for LOGBS were lower than EOGBS overall, with the highest again in Africa (0.12 [0.05–0.19]) (Table 1). Due to insufficient data from Oceania, we applied the CFR for Asia. We estimated moderate to severe neurodevelopmental impairment (NDI) after meningitis, only, because data were insufficient to estimate NDI after sepsis, as described elsewhere in this supplement [20]. To do this, we applied the percentage of infant cases of GBS disease which were meningitis, for early (12% [8%–15%]) and late-onset (42% [30%–55%]) GBS disease [18] to estimates of EOGBS and LOGBS survivors. We then applied an incidence risk of moderate to severe NDI at 18 months of age of 0.18 (0.13–0.22) [20]. These data were limited to developed countries; however, we applied this proportion worldwide, on the basis that this would be a minimum estimate as NDI was unlikely to be lower in settings with reduced levels of care. We compared the results from the compartmental model for infant GBS disease cases with those estimated using incidence data on infant GBS disease [19]. To do this, we calculated subregional incidence, or regional incidence where subregional data were not available, of EOGBS and LOGBS disease. We applied these to estimates of live births for each country in 2015. Data inputs are given for each country in Supplementary Table 2. To calculate the numbers of infants with invasive GBS disease and coexistent neonatal encephalopathy, we used previously published national incidences of neonatal encephalopathy and modeled uncertainties and adjusted these for births in 2015 [11]. Then using our new data, we calculated the proportion of invasive GBS disease among these cases of NE. In developed countries, among all NE cases included in cooling trials, 0.51% (95% CIs, 0.05%–0.97%) were also identified as having GBS disease [22]. Data inputs were limited for data from other regions (3/16 studies), so we used the worldwide estimate of 0.58% (95% CIs, 0.18%–0.98%) of NE cases with GBS disease to apply in Africa, Asia, Latin America, and Oceania. Since our case definition assumes that cases of NE with GBS count as a case of GBS invasive disease, we include these numbers within our estimates of GBS infant disease. Where a compartmental approach was not possible, we used incidence, prevalence, or risk ratios from pooled data applied to births in 2015 to make minimum estimates of worldwide, regional, and national estimates for cases attributable to GBS (Figure 2). We calculated the pooled incidence of maternal GBS disease per 1000 maternities and applied this to a denominator of total births worldwide to estimate cases. As described elsewhere [17], data were only available for developed countries, with a pooled estimate of 0.23 (95% CI, .09–.37) per 1000 maternities. We applied this to all regions, on the basis that maternal GBS disease was unlikely to be lower in settings with reduced levels of care. We calculated the pooled prevalence of GBS disease in stillbirths, equating also to the minimum number of fetal infections. Data were available from developed countries (1% [95% CIs, 0–2%]) and from Africa (4% [95% CIs, 2%–6%]) [18]. For regions with no data, we applied the prevalence of GBS in stillbirths from developed countries, on the basis that GBS-associated stillbirth was unlikely to be lower in settings with reduced levels of care. However, as this is a conservative approach, we did a sensitivity analysis applying the regional estimate from Africa (4% [95% CIs, 2%–6%]) [18] to regions with no data. We calculated pooled risk ratios or odds ratios for the association between maternal GBS colonization and preterm birth [21]. For cohort or cross-sectional studies, the risk ratio was 1.21 (95% CI, .99–1.48; P = .061), and for case-control studies, the odds ratio was 1.85 (95% CI, 1.24–2.77; P = .003). However, for preterm birth the results, in terms of the association between maternal colonization and preterm birth, are susceptible to confounding and bias. For preterm birth, we thus give a range for the number of cases, based on calculation of the population attributable fraction, which could be attributable to GBS given maternal GBS colonization [16] and incidence of preterm birth [33]. The ranges are based on the range in the 95% CIs of risk and odds ratios (1.0–2.8) for the association between maternal GBS colonization and preterm birth. We applied risks, without adjusting for IAP use, to estimates of live births for 2015 to calculate early-onset cases with no IAP use. We adjusted for skilled birth attendance as previously and applied regional facility CFRs to estimate deaths with no IAP use. We subtracted current cases and deaths in early infancy to calculate those currently prevented by IAP. For IAP scale-up worldwide, we assumed that all births were being attended by a skilled birth attendant, able to provide careful clinical monitoring for risk factors at delivery and administer IAP, but we did not adjust CFRs for this. Given these assumptions, we applied risks of EOGBS disease with a clinical risk factor approach, with coverage >50% and IAP worldwide where microbiological screening and IAP was not already in place. We did not calculate cases prevented with IAP for pregnant or postpartum women or stillbirths, or late-onset cases as these are not the target of IAP and any effect is likely to be limited due to the timing of IAP administration. For maternal GBS vaccination, we calculated cases prevented (with no IAP) by a maternal GBS vaccine with 80% efficacy and coverage at 50% and 90%, for births in 2015. No assumptions were made on skilled birth attendance and/or laboratory capacity. We calculated the prevalence of GBS serotypes (Ia/Ib/II–X) colonizing mothers and causing maternal and infant GBS disease from meta-analyses of proportions of each serotype reported in each disease syndrome [16, 17, 19]. We calculated the coverage of a pentavalent maternal GBS vaccine (Ia/Ib/II/III/V) based on these data. For the compartmental model, we included uncertainty at every step by taking 1000 random draws, assuming a normal distribution with a mean equal to the point estimate of the parameter, and standard deviation (SD) equal to the estimated standard error (SE) of the parameter. We present the 2.5th and 97.5th centiles of the resulting distributions as the uncertainty range (UR). For the incidence or proportional approach, we estimated uncertainty around the point estimate with the same approach, taking 1000 random draws, assuming a normal distribution with a mean equal to the point estimate of the parameter, and SD equal to the estimated SE of the parameter. We present the 2.5th and 97.5th centiles of the resulting distributions as the UR. Code used for the estimation process is available online at https://doi.org/10.17037/data.51.