Global Prevalence of Severe Neonatal Jaundice among Hospital Admissions: A Systematic Review and Meta-Analysis

Evidence regarding the adverse burden of severe neonatal jaundice (SNJ) in hospitalized neonates in resource-constrained settings is sparse. We attempted to determine the prevalence of SNJ, described using clinical outcome markers, in all World Health Organization (WHO) regions in the world. Data were sourced from Ovid Medline, Ovid Embase, Cochrane Library, African Journals Online, and Global Index Medicus. Hospital-based studies, including the total number of neonatal admissions with at least one clinical outcome marker of SNJ, defined as acute bilirubin encephalopathy (ABE), exchange blood transfusions (EBT), jaundice-related death, or abnormal brainstem audio-evoked response (aBAER), were independently reviewed for inclusion in this meta-analysis. Of 84 articles, 64 (76.19%) were from low- and lower-middle-income countries (LMICs), and 14.26% of the represented neonates with jaundice in these studies had SNJ. The prevelance of SNJ among all admitted neonates varied across WHO regions, ranging from 0.73 to 3.34%. Among all neonatal admissions, SNJ clinical outcome markers for EBT ranged from 0.74 to 3.81%, with the highest percentage observed in the African and South-East Asian regions; ABE ranged from 0.16 to 2.75%, with the highest percentages observed in the African and Eastern Mediterranean regions; and jaundice-related deaths ranged from 0 to 1.49%, with the highest percentage observed in the African and Eastern Mediterranean regions. Among the cohort of neonates with jaundice, the prevalence of SNJ ranged from 8.31 to 31.49%, with the highest percentage observed in the African region; EBT ranged from 9.76 to 28.97%, with the highest percentages reported for the African region; ABE was highest in the Eastern Mediterranean (22.73%) and African regions (14.51%). Jaundice-related deaths were 13.02%, 7.52%, 2.01% and 0.07%, respectively, in the Eastern Mediterranean, African, South-East Asian and European regions, with none reported in the Americas. aBAER numbers were too small, and the Western Pacific region was represented by only one study, limiting the ability to make regional comparisons. The global burden of SNJ in hospitalized neonates remains high, causing substantial, preventable morbidity and mortality especially in LMICs.


Introduction
Severe neonatal jaundice (SNJ) in a neonate may manifest as acute bilirubin encephalopathy (ABE) [1] with a range of symptoms including difficulty feeding, tone abnormalities, abnormal cry and the kernicteric facies [2] scored using the bilirubin-induced neurological dysfunction (BIND) score or modified BIND [3,4]. Persistent abnormalities which are now known as the Kernicterus Spectrum Disorder (KSD) [1], occur in 70% of survivors beyond the neonatal period including choreo-athetoid cerebral palsy, deafness, speech and language processing disorders, enamel dysplasia, and learning difficulties [5][6][7].
The Global Burden of Disease study ranks SNJ among the top 5-10 causes of neonatal deaths in countries with the highest number of neonatal deaths [8]. Previous attempts at providing global and regional estimates of SNJ burden have been challenged by limited data. Bhutani et al. estimated 481,000 global cases of SNJ among term/near-term neonates with 114,000 deaths and 75,000 of survivors developing kernicterus [9]. These figures derived using mathematical models with limited data have limitations inherent in such estimates. A previous population-based systematic review and meta-analysis including some authors in our current team (TS, DA, BL) reported a pooled incidence of SNJ at 244 per 100,000 live births [10]. A major drawback of this review was the disproportionate representation of high-income countries with lesser burden of disease. Several studies have suggested that the African and Asian regions have the highest burden of disease [9][10][11]. Factors responsible for these regional burdens include the high prevalence of glucose-6-phosphate deficiency (G6PD) deficiency, late presentation due to the high incidence of out-of-hospital births, inability of caregivers to promptly identify jaundice, caregivers' decision to seek alternative treatments; lack of or ineffective phototherapy and unavailable or unreliable access to bilirubin estimations [11]. Unfortunately, most data on SNJ in low-resource countries is hospital-based without true population-based data making the actual burden of SNJ unknown. However, this review of hospital-based data covers a wider representation of literature from diverse countries to ascertain, though still imperfect, the burden in low and lower-middle-income countries (LMICs). Our intent is to be the first comprehensive, current systematic review and meta-analysis that provides rigorous, worldwide appraisal of SNJ for all neonatal hospital admissions which included adverse clinical outcomes seen in SNJ, to compare regional geographic differences, and to provide representation from low/lower-resource areas. These data are critical not only to meet the global Sustainable Development Goals (SDGs) but also to assist in identifying region-specific strategies to decrease disability-adjusted life years (DALYs) from KSD morbidities [12].

Criteria for Article Inclusion
We included hospital-based studies that had neonatal hospital admissions for any cause and provided information about at least one clinical marker of SNJ, including number of exchange blood transfusions (EBTs); ABE; abnormal brainstem audio-evoked response (aBAER); or jaundice-related death. No patients or members of the public were involved in any way, and data were from published sources; therefore, investigational review board or ethics committee approval was not needed.

Criteria for Article Exclusion
Articles were excluded if (1) the entire data collection period was prior to 1997 or later than 2020, (2) sample size was <10, (3) period of data collection was not defined, (4) jaundice was conjugated, from metabolic or neonatal liver disease, (5) EBT was done for conditions unrelated to SNJ, (6) non-English, (7) non-neonatal, (8) publication type was a review article, questionnaire or survey or study design was case-control or experimental study on a subset of neonates with jaundice, and (9) missing critical data (total number of neonatal admissions). In the case of missing data, we (FA, TS) contacted the authors for further information and excluded the article if the requested information was not supplied. Magnetic resonance imaging (MRI) was excluded due to no returned results.

Outcome Definition
Our primary outcome was the prevalence of SNJ in hospitalized neonates (both inborn and outborn) clinically defined as having at least one clinical indicator of SNJ noted above. We also looked at the prevalence of SNJ in hospitalized jaundiced neonates again using clinical markers noted above. LMIC status was defined using the 2020 World Bank Criteria [13].

Search Criteria
We conducted a comprehensive search including both natural language and controlled vocabulary terms to reflect concepts of a neonatal population and jaundice, including both serum bilirubin and clinical indicators. The search was conducted across five databases: Ovid Medline, Ovid Embase, Cochrane Library via Wiley, African Journals Online, and Global Index Medicus (Figure 1). This search strategy was translated across the different databases to ensure appropriate use of the available controlled vocabulary and unique search functionality. The search was conducted in June 2018 and updated in September 2020. The protocol was registered in Prospero (CRD42018100214). A PRISMA checklist was completed (Table S1).

Data Extraction
Articles were screened using the Rayyan software for systematic reviews [14]. One author (CB) conducted the literature search and uploaded all eligible abstracts onto the software. Three groups of reviewers comprising of two authors per group [group A (UD

Data Extraction
Articles were screened using the Rayyan software for systematic reviews [14]. One author (CB) conducted the literature search and uploaded all eligible abstracts onto the software. Three groups of reviewers comprising of two authors per group [group A (UD and TO), group B (FU and LH) and group C (KS and FA)] were involved in the literature review.
During screening, each group member independently assessed the allocated articles' titles and abstracts for eligibility. Discrepancies were resolved via group dialog and, when necessary, a third author (TS) acted as an arbitrator. This was followed by full-text screening with reasons for exclusion recorded (Figure 2). Data extraction forms were developed, piloted, and refined. Data were extracted using Qualtrics ® , including citation information, country and World Health Organization (WHO) region, study duration, total number of neonatal admissions and neonatal jaundice (NNJ) admissions, gestational age (all, only term, only near-term, term and near-term combined, only preterm or unspecified), how jaundice was determined (clinical definition or using bilirubin assay), markers of SNJ: number of EBTs, aBAER, and reported jaundicerelated deaths. Microsoft Excel spreadsheet was used to collect data for analysis. Table 1 shows the profile of articles selected for the meta-analysis.  Data extraction forms were developed, piloted, and refined. Data were extracted using Qualtrics ® , including citation information, country and World Health Organization (WHO) region, study duration, total number of neonatal admissions and neonatal jaundice (NNJ) admissions, gestational age (all, only term, only near-term, term and near-term combined, only preterm or unspecified), how jaundice was determined (clinical definition or using bilirubin assay), markers of SNJ: number of EBTs, aBAER, and reported jaundice-related deaths. Microsoft Excel spreadsheet was used to collect data for analysis. Table 1 shows the profile of articles selected for the meta-analysis.

Risk of Bias (Quality) Assessment
Each article was scored based on five parameters that were modified from those used in a prior population-based study also using clinical parameters to assess the burden of disease from SNJ [11]. Scoring was in line with recommendations by the modified quality assessment tool for systematic reviews of observational studies (QATSO) scoring system [98]. These included: (1) if the sampling method was representative of the target population i.e., covered the whole nursery population (scored 3), term and near-term only (scored 2), premature or term only (scored 1); (2) the method used to define jaundice, categorized based on bilirubin assay (scored 2), visual clinical assessment (scored 1) or not stated (scored 0); (3) whether the study excluded any of the following conditions: Glucose -6 Phosphate Dehydrogenase deficiency, ABO incompatibility, Rhesus incompatibility or sepsis and was grouped as yes (scored 0) or no (scored 1); (4) if total number of SNJ cases was reported and classed as yes (scored 1) or no (scored 0); and (5) whether clinically significant jaundice was clearly defined in the Methods section (including use of AAP/NICE guidelines) and classified as yes (scored 1) or no (scored 0). Each study's quality was judged based on aggregate points, with a maximum obtainable score of 10, and classified as "good quality" (7-10 points), "fair quality" (4-6 points) and "poor quality" (0-3 points).

Statistical Analysis
The summary estimate for the meta-analysis was prevalence/proportion, which was transformed using Freeman-Tukey double arcsine transformation to enable them to correspond to probabilities under the standard normal distribution and enhance significance testing [99]. The double transformations adequately addressed issues of variance instability as well as confidence intervals (CIs) of proportions falling outside the possible range of 0 to 1 for binomial data [100]. Pooled estimates were calculated using DerSimonian and Laird's random-effects method, weighting individual study estimates using the inverse of the variance of their transformed proportion as study weight, with their 95% CI determined using the Clopper-Pearson exact binomial method [101]. Statistical heterogeneity among studies was assessed using Cochran's Q test and I 2 with a p-value of <0.10. The I 2 quantifies the proportion of the dispersion that is real and not spurious [102]. Possible sources of heterogeneity were also explored via subgroup analysis.
Additionally, mixed-effects meta-regression analysis was used to determine whether study-level covariates, such as publication year, country-level income and methodologi-cal domains for assessing study quality, explained some of the observed between-study heterogeneity. A formal test of publication bias was assessed using Begg's adjusted rank correlation [102] and Egger' regression asymmetry tests [103], as well as through visual interpretation of funnel plots. Analyses were conducted using R software (version 4.3.0; R Foundation for Statistical Computing, Vienna, Austria).

Results
The electronic databases search identified 4436 distinct articles (after removing 1497, duplicates) ( Figure 2). An additional 3729 articles were excluded after reviewing titles and abstracts. Seven hundred and six (706) articles were selected for full article review and 700 (99%) were retrieved and reviewed. The remaining six articles were unavailable from any source that we could access. Eighty-four hospital-based studies involving a total of 2,210,043 neonatal admissions and 5986 neonates with at least one marker of SNJ were included in the meta-analysis (Table 1) [6,.
Only nine studies reported aBAERs, making the reported results likely a gross underestimate ( Figure 6). For comparison, eight studies reported aBAERS among neonates admitted with jaundice ( Figure 7).  (Tables 4 and 5).
There was evidence of potential publication bias influencing the reporting of prevalence of SNJ. Studies that report a lower proportion of neonates with SNJ were less likely to be published. The funnel plot appeared largely asymmetrical (Figure 8) with empirical evidence supporting this observation (Begg test p < 0.001, Egger's bias = 13.0, p < 0.001).  (Tables 4 and 5).
There was evidence of potential publication bias influencing the reporting of prevalence of SNJ. Studies that report a lower proportion of neonates with SNJ were less likely to be published. The funnel plot appeared largely asymmetrical (Figure 8) with empirical evidence supporting this observation (Begg test p < 0.001, Egger's bias = 13.0, p < 0.001).
Of the six methodological domains included (5 domains used for assessing study quality and type of study facility), only facility type showed significant differences in estimates of the prevalence of SNJ in subgroup analysis (Table 6). In meta-regression analysis, publication year (<0.001), country income level (p = 0.009), representativeness of the sample to the target population (p = 0.04), and type of healthcare facility (p = 0.001) significantly explained 17.00% of the between-study heterogeneity in the observed prevalence of SNJ (Table 7). Of the six methodological domains included (5 domains used for assessing study quality and type of study facility), only facility type showed significant differences in estimates of the prevalence of SNJ in subgroup analysis (Table 6). In meta-regression analysis, publication year (<0.001), country income level (p = 0.009), representativeness of the sample to the target population (p = 0.04), and type of healthcare facility (p = 0.001) significantly explained 17.00% of the between-study heterogeneity in the observed prevalence of SNJ (Table 7).      A one-year increase in publication year was found to predict a decrease in prevalence of SNJ by 0.6% (coefficient: −0.006 [95% CI: −0.010, −0.002]), indicating more recent studies tended to publish lower prevalence for SNJ compared to earlier published studies. Upper-middle-, lower-middle-, and low-income countries all had a prevalence of SNJ that was 4.10%, 8.70% and 3.60% higher than the prevalence of SNJ reported among highincome countries. Studies conducted on all neonates or term and near-term neonates had 6.20% and 7.50% lower prevalence than studies that only included term only or preterm only respectively. The prevalence of SNJ in tertiary/referral hospitals was 7.0% higher (coefficient: 0.070 [95% CI: 0.031, 0.109]) than studies that did not report type of healthcare facility. In multivariable meta regression analysis, year of publication, income level of country, sample representative of target population (whether it included term, preterm or whole neonate population), method used to define jaundice, study having reported total number of NNJ cases, whether clinically significant jaundice was clearly defined or not together explained 58% of the variation in SNJ prevalence across countries.

Discussion
Our data demonstrate that adverse clinical outcomes of SNJ remains a significant public health concern in LMICs. It continues to be a leading cause of neonatal admissions and death. SNJ contributes substantially to neonatal mortality worldwide, with the highest burden in the African (1.49%) and South-East Asian (0.82%) WHO regions. Our study highlights the global prevalence of SNJ with ranges varying from 3.34% in the African and 2.58% in the South-East Asian regions to 1.73%, 1.42%, 1.31% and 0.74% in the Americas, Eastern Mediterranean, European and Western Pacific regions. SNJ is associated with a substantial risk of long-term disability [2,8,9]. Of note, the prevalence declined slowly over time by 0.6% per year.
Focusing only on those with NNJ, our data show a prevalence of SNJ among this cohort ranging between 8.3% and 31.4%, with the highest burden of disease in the Western Pacific and African regions. However, the Western Pacific region was only represented by one report from China (upper middle-income) and the America's had only five studies [USA-high-income (n = 3), Brazil-upper middle-income (n = 1), Bolivia-lower middleincome (n = 1)]. Despite efforts to find worldwide data there is still selection bias due to underreporting with only 27 of the 195 official countries providing any data at all.
Although still unevenly distributed with many countries without data, our review more accurately represents the global burden of SNJ than previous studies/reviews have done with 64 (76.19%) of the articles from LMICs and an additional 14 articles (16.67%) from upper-middle-income countries. This is a stark contrast to the previous populationbased systematic review and meta-analysis, where 76% of the included studies were from high-income countries and, thus, much less representative of the actual income distribution globally than this present study [10]. Our current work included 84 studies representative of all WHO regions, including more country diversity and income levels within most regions.
Additionally, most articles included in our review studied both term and preterm populations. Important because preterm infants have a higher prevalence of NNJ and a higher risk of neurological damage at lower bilirubin levels [27,104]. Of note, studies that included only preterm neonates reported a higher prevalence of SNJ than other studies, however, this difference did not attain statistical significance; an observation differing from other reports [27,105].
This review also reported a higher prevalence of SNJ in higher-tier health facilities when compared to primary and secondary health facilities, possibly because SNJ is usually managed in higher-tier facilities due to these facilities having more phototherapy devices and manpower [105]. The actual burden is likely underreported as many neonates do not reach tertiary centers in LMICs.
The current review highlights that NNJ is noted in 21.99% of all neonatal admissions across WHO regions in the studies included in our review, consistent with prior studies [23,43,82]. Of all neonatal admissions with jaundice, those that had clinical evidence of severe disease ranged from 8.31-31.49% with variability across regions with areas with higher prevalence in regions where neonates often present late to the hospital, likely attributable to previously identified factors [8,106,107].
Striking differences persist between WHO regions for individual SNJ markers again with wide ranges for both ABE and percentages of neonates requiring EBTs. Many complications are likely underreported due to the lack of follow-up and/or the ability to perform specialized testing including BAER or MRI. For these reasons, along with the lack of representation of many countries, we expect that these data significantly underestimate the true burden of severe disease. With known effective treatment strategies, including intensive phototherapy and EBT, likely coupled with maternal education, early timely diagnosis, and treatment [106,108,109], these complications are preventable in almost all neonates.
Our study also looked at other factors potentially associated with SNJ. With the metaregression, three factors-publication year, type of study facility and country income level accounted for 58% of the heterogeneity. More recent studies tended to report a lower prevalence of SNJ which may reflect modest gains in recently introduced national programs such as the "Every Newborn Action Plan (ENAP)" focusing on newborn risk assessment, identification of cases with prompt referrals, maternal education and postnatal visits and having the potential of reducing behavioral factors that contribute to SNJ [110].
Study limitations include the continued underrepresentation of several regions/countries in this data set and the decision to limit the search to English only based on the lack of any population-based data in other languages in the previous population-based review. Inability to accurately ascertain place of birth and uniformly determine how many neonates were readmissions versus admissions from outside of healthcare facilities. Bilirubin levels were not required because bilirubin levels are not uniformly available in all hospitals and definitions for severe hyperbilirubinemia vary widely. Another limitation of this review is in the observed high degree of heterogeneity of pooled prevalence which is not unexpected for prevalence studies with marked methodological differences. Though we tried to deal with the high heterogeneity by looking at the effect of design, year and population characteristics in the meta-regression, we still found significant heterogeneity. Despite the high heterogeneity, it is clear that the burden of disease remains high, with a much higher proportion of the disease in LMICs. This does not alter the significance of our study as it is a representation of available research done thus far in hospitalized neonates. We also failed to link prevalence of SNJ in this review to the predicted prevalence of long-term sequelae. A disappointing limitation is the small number of studies from three of WHO regions (Americas, Eastern Mediterranean, and Western Pacific), decreasing generalizability of the findings in these regions. Of note, the Americas did have stronger representation in the previous population-based review and did not have a high prevalence of SNJ [10]. This study could have been potentially strengthened by adding additional weighting based on the prevalence of known factors, such as G6PD deficiency, Rhesus disease and neonatal sepsis, which vary among different WHO regions in neonates with SNJ. This additional weighting should be included in future systematic reviews and meta-analyses to help determine the global burden of SNJ. A strength of this study is that it included global data, and most of articles analyzed were adjudged to be of high quality which strengthens the validity of our findings. The relatively high representation from the African, European, and South-East Asian regions and middle-income countries enhances generalizability of our findings to these regions. Overall, this review had better representation than our previous population-based study although attempts to get population-based data should continue.
As we highlighted in the limitations section above, our data demonstrated high heterogeneity but, despite that limitation, provides the best representation of the burden of disease, especially in LMICs/LICs available at this time. Country and regional registries and population data are urgently needed but only largely available in a few high-income countries globally [10]. Should true population-based data become widely available, they will provide more robust and generalizable data. However, if we wait for that population data to come, it will likely be years if not decades before important stakeholders, such as the WHO and United nations Children's Fund (UNICEF), move SNJ to the top of their list of global neonatal priorities. Using mathematical modeling, Bhutani et al. [9] predicted that in 2010, there were 240 million infants at risk for neonatal hyperbilirubinemia-related adverse outcomes, and 750,000 with KSD. With increasing populations in Africa and other LMICs where the burden of SNJ is highest, these estimates will increase if mitigation factors are not implemented. More studies are also needed that factor in the medical standards and risks for developing jaundice in each country along country and even within-country regionalspecific guidelines based on the risks and treatment available within a given country or region. As highlighted in American Academy of Pediatrics (AAP) 2022 guidelines, and also highlighted in a recent perspective piece, LMICs need to base treatment on their own risks and resources [111,112]. Using AAP guidelines would potentially lead to a substantially higher burden of both ABE and KSD than we currently see in LMICs.

Conclusions
SNJ remains an important contributor to neonatal morbidity and mortality, especially in the African and South-East Asian regions. As we work towards the SDGs of improving neonatal mortality and the goal of decreasing morbidity, SNJ needs to be addressed as a preventable cause of both-most effectively addressed with a package approach which includes maternal, community and healthcare provider education; country specific guidelines based on risk and resources; accurate reliable low-cost methods of screening and diagnosis including not only bilirubin levels but also blood grouping and Rhesus as well as G6PD screening, effective phototherapy, capabilities to do safe EBT's when indicated and comprehensive follow-up and treatment for all children with KSD.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/jcm12113738/s1. Author Contributions: T.M.S. conceptualized and designed the study, coordinated and supervised data collection, drafted the initial manuscript, and reviewed and revised the manuscript. U.M.D., F.U., L.H., T.O. and F.A., designed the study, collected data, drafted the initial manuscript, and reviewed and revised the manuscript. K.M.S. collected data, drafted the initial manuscript, and reviewed and revised the manuscript. C.J.B. designed the study including the data collection instruments, collected data, drafted the initial manuscript, and reviewed and revised the manuscript. D.A. designed the study, analyzed the data, drafted the initial manuscript, and reviewed and revised the manuscript. B.W.L. designed the study, supervised data collection especially the quality assessment aspects and reviewed and revised the manuscript. All authors have read and agreed to the published version of the manuscript.