Pneumococcus was initially thought to cause only pneumonia, until in 1886, Albert Fraenkel, who had previously coined the term “Pneumokokkus,” described pneumococcal meningitis in association with pneumonia.1 Soon, it was documented that this organism was also responsible for other invasive syndromes, that is, osteomyelitis, septic arthritis, peritonitis, abscesses in the brain, liver and spleen and bacteremia without a recognized focus.1 It is now understood that after pneumococcal colonization, the bacterium can cause invasive disease by either bloodstream invasion or direct lower respiratory tract invasion followed by dissemination to various sites in the human body.
Globally, pneumococcus has been estimated to cause 541,000 deaths (uncertainty range: 376,000–594,000), in the year 2008, in children younger than 5 years.2 However, the morbidity and mortality caused by pneumococcus in neonates has not been estimated nor has the variation in disease burden estimates across the world been quantified. Most of the published etiologic data among neonates from developing countries are from hospital-based studies. This is also true among those in developed countries because the assessment of an ill neonate necessarily requires equipment available in health facilities. However, it may be assumed that in developing countries, the likelihood of seeking medical care from an appropriate health provider for an ill neonate is low; as for example, only 59% of children younger than 5 years with suspected pneumonia are taken to an appropriate healthcare provider in developing countries in 2005–2009 according to United Nations International Children’s Emergency Fund.3
To improve the representativeness, this project focused on community-based longitudinal studies and surveillance systems from around the world that had data on invasive pneumococcal disease (IPD) among neonates, including data that had not been published. Hence, the burden of IPD among neonates was estimated by conducting a systematic review of published and unpublished studies and employing meta-analysis techniques. The estimation of IPD incidence in neonates during the pre-pneumococcal conjugate vaccine (pre-PCV) era, in the general population and in high-risk populations, globally and in the 3 United Nations (UN) country development strata was undertaken.
Pneumococcal invasive disease was defined as pneumococcus identified by a standard laboratory technique such as culture or polymerase chain reaction from a normally sterile site, that is, blood, cerebrospinal fluid or pleural fluid. Pre-PCV era of a country was defined as the time period before PCV introduction into the routine immunization schedule of the country. Studies were classified by their level of development at the time of study data collection into the 3 UN country development strata (least, less and more developed). Neonatal mortality rate (NMR) was defined as the number of neonatal deaths per 1000 live births.4 Studies were included based on the inclusion and exclusion criteria (Fig. 1).
Searches were conducted in Medline, Embase and World Health Organization Library and Information Networks for Knowledge Database databases, using Medical Subject Headings (MeSH), EMTREE terms and text words (detailed search strategy available on request). Hand search of reference lists of included studies and bibliographies of associated systematic reviews were carried out. Authors of included published studies were contacted for missing data and/or additional data, which also yielded updated or unpublished data.
Selection of Studies and Data Abstraction
Based on the inclusion and exclusion criteria, titles and abstracts of all the articles were reviewed after duplicate records were deleted. For studies with multiple publications, only the primary publication was included. Full texts of included studies were reviewed, and the data were obtained using a data abstraction form. Data on NMR for the year corresponding to data collection period of studies were obtained from the World Health Organization neonatal mortality database, and if data were unavailable for certain countries, United Nations International Children’s Emergency Fund statistics and monitoring database was searched.5,6 Abstracted data were then transferred to STATA version 10.1 (StataCorp LP, College Station, TX) and checked for errors and inconsistencies.
Each article was reviewed and evaluated for sources of systematic bias. Biases were assessed using a bias assessment questionnaire following the Cochrane Collaboration’s tool for assessing the risk of bias.7 Bias assessment questionnaire was answered as yes, no or unclear indicating high risk, low risk and unclear risk of bias, respectively. Studies were assessed for study quality based on risk of bias in case ascertainment, follow-up rate and diagnostic methods followed for the detection of pneumococcus. A study was given a high risk of bias in case ascertainment if the study may have missed a significant proportion of cases. A study was given a high risk of bias in diagnostic methods if the method used may have been unreliable. Finally, a study was given a high risk of bias in attrition if the longitudinal study had a loss to follow-up rate of more than 20% leading to emigrative selection bias or if censoring was not adequately handled. The methodological quality of studies and the impact on the quantitative result of the review were assessed based on study characteristics and study quality presented in the included studies.
Studies that provided population-based incidence rates of pneumococcal invasive disease among neonates were stratified by the UN country classification by level of development (more-developed regions, less-developed regions and least-developed countries). Incidence data were transformed to logits log (P/(1 − P)), where P represents a probability, in all meta-analyses and meta-regressions and then back transformed to incidence (expressed as per 100,000 live births) and rounded up to 1 decimal place. Random effects model using the method of DerSimonian and Laird, a simple noniterative procedure,8 which takes into account both within study variance and among-study variance, was selected. I2 statistic was calculated holding a cutoff of 50%, with values above 50% considered to exhibit statistical heterogeneity of studies. Sources of heterogeneity among studies were explored by constructing univariate random effects models with the log odds of neonatal IPD and each study covariate using the metareg command in STATA and determining the variance explained by each of these covariates, using the formula: ((1 − (τ2 in the model/τ2 in the model without covariates)) × 100), where τ2 is the variance of the distribution of the true effects across studies. Studies with missing data were excluded if we did not receive a response from study authors after 2 reminders. A continuity correction of k/N was chosen, where k was a proportionality constant of a chosen size, and N was the sample size. We set k = 1, making the continuity correction 1/N. This method of adjustment introduces less bias when comparison groups are severely unbalanced and is more conservative when compared with the traditional Yates continuity correction of 0.5.9 Subgroup meta-analyses were performed for high-risk neonates (HIV positive, sickle cell disease and indigenous populations). The association between IPD incidence in neonates and NMR, a country level covariate, was explored by constructing a random effects meta-regression model.
Electronic searches of databases yielded a total of 2104 records (Fig. 2). Figure 2 includes search results for pneumococcal colonization, IPD and case fatality rates in neonates, younger than 2 month and younger than 1 year (as part of a larger project). The systematic review process resulted in a total of 124 included studies providing 174 data points, as a single study could provide more than 1 data point. If a study reported data disaggregated by ethnicity, vaccination status or year of data collection, it would yield more than 1 data point for modeling (174 references available on request). Considering only pneumococcal invasive disease among neonates in the pre-PCV era, there were 26 data points. Table 1 presents the characteristics of the 26 IPD incidence data points among neonates in the pre-PCV era. Data collection period reported ranged within 1980–2010 time frame.
Assessment of Methodological Quality
Studies on neonatal IPD were assessed for inherent biases. In the least-developed UN country stratum, there was only a single study from The Gambia (See Table, Supplemental Digital Content 1, https://links.lww.com/INF/C301, RefID: 2376).10 Although the study had a low risk of case ascertainment bias and diagnostic bias the authors concluded that these were minimal rates as not all ill neonates receive medical care and the population denominator was not known for uncertain. In the less-developed UN country stratum, except for data from 2 countries, the studies in this UN country stratum had low risk of case ascertainment bias and diagnostic bias as robust active surveillance systems were in place with standard specimen collection, transportation and laboratory techniques for the isolation of pneumococcus (See Table, Supplemental Digital Content 1, https://links.lww.com/INF/C301). There were several data points from the more-developed UN country stratum, which represented the general population and/or the high-risk population in the respective countries (See Table, Supplemental Digital Content 1, https://links.lww.com/INF/C301). A majority of these countries had active national surveillance systems in place but a few had passive surveillance systems. Active surveillance systems allow for more complete capture of all cases of IPD than passive surveillance systems but do not in themselves change physician diagnostic practices. Furthermore, in countries with active population-based surveillance systems coupled with high care-seeking behavior and active collection of diagnostic specimens among ill neonates, the IPD incidence is likely to more closely reflect the true burden of bacteremic disease among neonates in these countries (See Table, Supplemental Digital Content 1, https://links.lww.com/INF/C301). In addition to the prospective, active surveillance and the passive surveillance systems, there was yet another type of surveillance—the retrospective surveillance design that provided IPD incidence data among neonates (See Table, Supplemental Digital Content 1, https://links.lww.com/INF/C301). These surveillances were laboratory-based and were described differently by the respective investigators. The difference between the active and passive surveillance systems was whether project personnel actively contacted microbiology laboratories periodically to collect data (an active system) or if laboratories reported to the surveillance system (a passive system). In a retrospective system, investigators analyzed already existing data in a surveillance system.
Three random effects models were constructed (model 1: combining general population and the high-risk population; model 2: comprising the general population only; model 3: high-risk population only). The results of these models are given in Table 2. In model 2, there was only 1 data point in the least-developed UN country stratum; therefore, it was not possible to estimate the pooled neonatal IPD incidence for that stratum. Comparatively, there were more number of data points in the less and more-developed UN country strata for analysis (Table 2; Fig. 3). The pooled random effects estimate of neonatal IPD incidence in the less-developed stratum was 16.0 per 100,000 live births [95% confidence interval (CI): 3.9–65.6 per 100,000] with considerable heterogeneity among data points (I2 of 94.3%). The pooled random effects estimate of neonatal IPD incidence in the more-developed stratum was 41.1 per 100,000 live births (95% CI: 29.1–58.1 per 100,000). Here, again there was high heterogeneity among data points was with an I2 of 95.7%. The overall pooled neonatal IPD incidence across all 3 UN country strata was estimated to be 36.0 per 100,000 (95% CI: 20.0–64.7; I2 = 99.0%). The pooled IPD incidence among neonates in the high-risk group was estimated to be 411.5 per 100,000 (95% CI: 75.7–2203.2 per 100,000; I2=69.3%) from the more-developed UN country stratum (Table 2, model 3).
Sources of Heterogeneity
The high level of heterogeneity among studies seen in the aforementioned models warranted the investigation of the sources of heterogeneity (Table 3). The study covariates that contributed to the heterogeneity among data points were mainly data collection period (74.1%) and UN country strata (72.2%). Country level covariate, NMR was also found to add to the heterogeneity among studies and explained 51.7% of the variance among studies.
Neonatal IPD incidence reported in studies from The Gambia (RefID: 2376)10 and indigenous populations from the more-developed UN country stratum (RefID: SS_11, b, e; K.L. O’Brien, surveillance data from investigator, personal communication, 2011) was higher than in other studies and so omitting these data points led to the greatest drop in the overall pooled neonatal IPD incidence estimate below 46.2 per 100,000 in the meta-influence plot (Fig. 4). On the other hand, when data points such as the surveillances from US (RefID: 1418)15 with low neonatal IPD incidence were omitted, it led to the most increase in the overall pooled estimate above 46.2 per 100,000. Therefore, Figure 4 indicates that the overall pooled IPD incidence in neonates ranged from 40.0 per 100,000 (95% CI: 22.7–70.5 per 100,000) to 51.7 per 100,000 (95% CI: 29.3–91.4 per 100,000).
Association Between Neonatal IPD Incidence and NMR
The meta-regression results did not indicate a statistically significant association between neonatal IPD incidence in the general population and NMR [odds ratio: 1.02; 95% CI: 0.96–1.1; P: 0.5; (graph of the relationship not shown, can be provided on request)].
This systematic review and meta-analyses synthesized the burden of IPD incidence in neonates in the pre-PCV era around the world incorporating published data and unpublished data from surveillance programs. The more-developed UN country stratum included a fair number of countries from various continents and may be more likely to be representative of its respective stratum. The less-developed UN country stratum had data from only 4 countries, and the dearth of data from this stratum needs to be taken into consideration when interpreting the results of the meta-estimates. Furthermore, there was only 1 data point on IPD incidence among neonates in the least-developed UN country stratum, which was from The Gambia. Hence, a pooled estimate for this stratum was noncomputable. The unweighted IPD incidence from The Gambia was very high; subsequent to this finding a case control study was undertaken to investigate the reasons for the very high IPD incidence.25 The reasons cited included access to care, case diagnostic thresholds and laboratory capacity factors. Hence, a generalization of this IPD incidence to other countries within the least-developed UN country stratum is not known, although the epidemiological conditions identified by this study are prevalent in the least-developed countries of the world.
The overall pooled estimate of neonatal IPD incidence in the general population was not influenced to a large extent by the data from the least-developed UN country stratum because there was only 1 data point from this stratum. The IPD incidence in the least-developed UN country stratum is likely to be higher than that reported from the less-developed and more-developed UN country strata. The more-developed stratum had a higher pooled IPD incidence estimate than the less-developed stratum, which most likely was because of differences in data collection periods of studies (1980s–1990s in the more-developed stratum and 1990s–2000s in the less-developed stratum). This confounding by time reflects changes over time in a population such as changes in socioeconomic conditions and epidemiological conditions (ie, prevalence of risk factors of IPD). In addition, the investigation for heterogeneity among data points indicated that data collection period explained much of the variance among data points. Moreover, there were data from only 4 studies in the less-developed UN country stratum, which may not have adequately represented their respective stratum.
There was wide variation in the IPD incidence reported by studies within and across UN country strata, and the reasons for this variability may have been both methodological and epidemiological. Methodological sources of heterogeneity included differences in case definitions followed in hospitals, clinical practices for the selection of cases for blood culture and lumbar puncture, diagnostic methods for the detection of pneumococcus and laboratory operating procedures. The population denominator estimation may have varied across data points, with some studies adjusting the denominator for NMR. Epidemiological sources of heterogeneity would include differences in host conditions, prevalence of risk factors, prevalence of antibiotic resistant strains and health service provision, quality and utilization.
IPD cases are frequently underestimated in studies of neonates. The incidence of IPD may have most likely been underestimated in countries where there is poor recognition of symptoms of illness and delay in recognition of danger signs in neonates. The underestimation is augmented by low care-seeking as a consequence of cultural beliefs,26 financial constraints or poor access to care.27 Furthermore, insufficient blood culture volumes from neonates, high growth rate of contaminants contribute to underestimation. Hence, IPD incidence estimates must be considered minimum estimates and not a true reflection of the IPD incidence among neonates.
We wanted to understand the way in which neonatal IPD incidence correlated with NMR for 2 reasons. First, if pneumococcus is a significant cause of neonatal mortality, then such a correlation should exist. Second, if a positive correlation existed, then pneumococcal disease burden among neonates could be inferred from knowing the NMR of a country. Although the meta-regression modeling did not indicate a statistically significant association between neonatal IPD incidence and NMR, the association between neonatal o colonization and NMR was statistically significant as reported elsewhere (M.E. Billings et al, unpublished data, 2015). This finding reiterates that the IPD incidence among neonates is highly likely an underestimation and thereby resulting in the inability to exhibit a statistically significant association with NMR.
There were several limitations to this study. The representativeness of countries in the least-developed and less-developed UN country strata was greatly restricted as there was very limited data from these strata. Countries where 49% of children younger than 5 years deaths occur—India, Nigeria, Democratic Republic of the Congo, Pakistan and China did not contribute to this analysis.28 Additional modeling to predict the IPD incidence in unrepresented countries was not possible as there were insufficient data points within UN country strata to undertake multivariate modeling. Furthermore, the available data on IPD incidence among neonates were almost certainly underestimated for aforementioned reasons. Finally, there was a high level of heterogeneity among data points, but further stratification on the covariates contributing to heterogeneity was not possible because of very few number of data points. For these reasons, the pooled estimates should be seen as minimum estimates.
Despite such limitations, there are clear benefits of conducting a systematic review and meta-analysis. The variation in IPD incidence among neonates across the world has been quantified; sources of heterogeneity have been investigated; pneumococcus has been shown to be an important pathogen even in the neonatal period especially in countries from the least-developed UN country strata. The need for population-based surveillance programs in the least-developed and less-developed strata is re-emphasized and the need for clear criteria for diagnostic testing, clear case definitions and quality standards in the laboratory. This study is a step forward in quantifying IPD incidence in neonates in the world with incidence in the least-developed UN country stratum being several-fold higher than that of the less-developed and more-developed UN country strata.
We have not addressed in this project the serotype distribution of the neonatal pneumococcal cases. There is a greater data paucity problem for serotype specific disease than there is for pneumococcal disease burden overall. Nevertheless, it is likely that the serotypes in the existing pneumococcal conjugate vaccines are also important in neonatal disease, as evidenced by reductions in IPD among neonates in the US in the pneumococcal conjugate vaccine era.29,30 It is also known that there is a broader range of serotypes causing disease in the neonates than in older children, so serotype specific vaccines, although helpful, will not be sufficient to address the burden of pneumococcal disease in neonates.
Future research on characterizing pneumococcal serotypes causing invasive disease in neonates in various countries needs to be undertaken so that both impact of pneumococcal conjugate vaccine on neonatal disease can be understood and so that future vaccine and other prevention strategies can be appropriately designed. Furthermore, very few studies focus primarily on pneumococcal disease in the neonatal period; because neonatal mortality is now a priority, more research focusing on neonatal pneumococcal disease especially from countries with high rates of neonatal mortality is needed as the burden of pneumococcal disease in this young population needs to be monitored.
The findings of this project confirm the need for neonatal survival strategies and prevention strategies to include prevention of pneumococcal disease through approaches that would likely include pneumococcal vaccine. This study also supports the need to investigate maternal vaccination against pneumococcus to reduce the burden of invasive disease among neonates.
We would like to acknowledge the authors of relevant studies, who responded to the request for data that were missing, which enabled the completion of the overall project. We thank the following for their time and support in this endeavor: O. Abdullahi, H.C. Baggett, C.L. Coles, R. Dagan, G. Dbaibo, S.M. Granat, C.C. Grant, M.A. Gutierrez Rodriguez, H.M. Heffernan, B. Henriques-Normark, P.C. Hill, H.E. Hsu, C.Z. Hua, A. Kisakye, T. Kurien, R. Lagos, E. Leibovitz, C. Levy, S.M. Mattie, E. Miller, E. Molyneux, M.R. Moore, S. Mudhune, B. Ozdemir, M. Paragi, K.A. Poehling, B. Resch, I.A. Rivera-Olivero, F.M. Russell, S.K. Saha, S.S. Vedula, A. Vergison, A. von Gottberg and K. Watson.
1. Siber GR, Klugman KP, Makela PH Pneumococcal Vaccines: The Impact of Conjugate Vaccines. 2008 Washington, D.C. ASM Press;
2. World Health Organization, Immunization, Vaccines and Biologicals. Estimated Hib and Pneumococcal Deaths for Children Under 5 Years of Age, 2008. 2012 Available at: http://www.who.int/immunization/monitoring_surveillance/burden/estimates/Pneumo_hib/en
. Accessed on July 25, 2015
3. UNICEF, SOWC, 2005–2009. Available at: http://www.unicef.org/statistics/index_24183.html
. Accessed on April 5, 2011
4. World Health Organization. Neonatal and Perinatal Mortality: Country, Regional and Global Estimates. 2006 Geneva: WHO
5. World Health Organization. Available at: http://www.who.int/healthinfo/statistics/mortality_neonatal/en/index.html
. Accessed on February 8, 2011
6. UNICEF. Available at: http://www.unicef.org/statistics/index_step1.php
. Accessed on February 8, 2011
7. Higgins JPT, Green S Cochrane Handbook for Systematic Reviews of Interventions Version 5.0.1. The Cochrane Collaboration. 2008 www.cochrane-handbook.org
. Accessed December 1, 2009
8. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7:177–188
9. Sweeting MJ, Sutton AJ, Lambert PC. What to add to nothing? Use and avoidance of continuity corrections in meta-analysis of sparse data. Stat Med. 2006;23:1351–1375 Erratum in: Stat Med. 2006 Aug 15;25(15):2700
10. O’Dempsey TJ, McArdle TF, Lloyd-Evans N, et al. Pneumococcal disease among children in a rural area of west Africa. Pediatr Infect Dis J. 1996;15:431–437
11. Lagos R, Muñoz A, San Martin O, et al. Age- and serotype-specific pediatric invasive pneumococcal disease: insights from systematic surveillance in Santiago, Chile, 1994–2007. J Infect Dis. 2008;198:1809–1817
12. Russell FM, Carapetis JR, Tikoduadua L, et al. Invasive pneumococcal disease in Fiji: clinical syndromes, epidemiology, and the potential impact of pneumococcal conjugate vaccine. Pediatr Infect Dis J. 2010;29:870–872
13. Heffernan HM, Martin DR, Woodhouse RE, et al. Invasive pneumococcal disease in New Zealand 1998–2005: capsular serotypes and antimicrobial resistance. Epidemiol Infect. 2008;136:352–359
14. Vergison A, Tuerlinckx D, Verhaegen J, et al.Belgian Invasive Pneumococcal Disease Study Group. Epidemiologic features of invasive pneumococcal disease in Belgian children: passive surveillance is not enough. Pediatrics. 2006;118:e801–e809
15. Poehling KA, Talbot TR, Griffin MR, et al. Invasive pneumococcal disease among infants before and after introduction of pneumococcal conjugate vaccine. JAMA. 2006;295:1668–1674
16. Paragi M, Kolman J, Kraigher A, et al.Slovenian Meningitis Study Group. Possibility of application of new pneumococcal conjugate vaccines in children in Slovenia. Vaccine. 2003;21:4708–4714
17. Grant CC, Harnden AR, Jewell G, et al. Invasive pneumococcal disease in Oxford, 1985-2001: a retrospective case series. Arch Dis Child. 2003;88:712–714
18. Kaltoft, Zeuthen N, Konradsen HB. Epidemiology of invasive pneumococcal infections in children aged 0–6 years in Denmark: a 19-year nationwide surveillance study. Acta Paediatr Suppl. 2000;89:3–10
19. Schuchat A, Robinson K, Wenger JD, et al. Bacterial meningitis in the United States in 1995. Active Surveillance Team. N Engl J Med. 1997;337:970–976
20. Gutiérrez Rodríguez MA, Ordobás Gavín M, Ramírez Fernández R, et al. [Incidence of pneumococcal disease in the autonomous Region of Madrid (1998–2006)]. Med Clin (Barc). 2008;130:51–53
21. Johnson AP, Waight P, Andrews N, et al. Morbidity and mortality of pneumococcal meningitis and serotypes of causative strains prior to introduction of the 7-valent conjugant pneumococcal vaccine in England. J Infect. 2007;55:394–399
22. Ispahani P. Bacterial meningitis in Nottingham. J Hyg (Lond). 1983;91:189–201
23. Dahl MS, Trollfors B, Claesson BA, et al. Invasive pneumococcal infections in Southwestern Sweden: a second follow-up period of 15 years. Scand J Infect Dis. 2001;33:667–672
24. Melegaro A, Edmunds WJ, Pebody R, et al. The current burden of pneumococcal disease in England and Wales. J Infect. 2006;52:37–48
25. O’Dempsey TJ, McArdle TF, Morris J, et al. A study of risk factors for pneumococcal disease among children in a rural area of West Africa. Int J Epidemiol. 1996;25:885–893
26. Zaman K, Zeitlyn S, Chakraborty J, et al. Acute lower respiratory infections in rural Bangladeshi children: patterns of treatment and identification of barriers. Southeast Asian J Trop Med Public Health. 1997;28:99–106
27. Syed U, Khadka N, Khan A, et al. Care-seeking practices in South Asia: using formative research to design program interventions to save newborn lives. J Perinatol. 2008;28(Suppl 2):S9–S13
28. Black RE, Cousens S, Johnson HL, et al.Child Health Epidemiology Reference Group of WHO and UNICEF. Global, regional, and national causes of child mortality in 2008: a systematic analysis. Lancet. 2010;375:1969–1987
29. Weatherholtz R, Millar EV, Moulton LH, et al. Invasive pneumococcal disease a decade after pneumococcal conjugate vaccine use in an American Indian population at high risk for disease. Clin Infect Dis. 2010;50:1238–1246
30. Moore MR, Link-Gelles R, Schaffner W, et al. Effect of use of 13-valent pneumococcal conjugate vaccine in children on invasive pneumococcal disease in children and adults in the USA: analysis of multisite, population-based surveillance. Lancet Infect Dis. 2015;15:301–309