Impact of neonatal sepsis on neurocognitive outcomes: a systematic review and meta-analysis

Abstract

Introduction

Sepsis is associated with neurocognitive impairment among preterm neonates but less is known about term neonates with sepsis. This systematic review and meta-analysis aims to provide an update of neurocognitive outcomes including cognitive delay, visual impairment, auditory impairment, and cerebral palsy, among neonates with sepsis.

Methods

We performed a systematic review of PubMed, Embase, CENTRAL and Web of Science for eligible studies published between January 2011 and March 2023. We included case–control, cohort studies and cross-sectional studies. Case reports and articles not in English language were excluded. Using the adjusted estimates, we performed random effects model meta-analysis to evaluate the risk of developing neurocognitive impairment among neonates with sepsis.

Results

Of 7,909 studies, 24 studies (n = 121,645) were included. Majority of studies were conducted in the United States (n = 7, 29.2%), and all studies were performed among neonates. 17 (70.8%) studies provided follow-up till 30 months. Sepsis was associated with increased risk of cognitive delay [adjusted odds ratio, aOR 1.14 (95% CI: 1.01—1.28)], visual impairment [aOR 2.57 (95%CI: 1.14- 5.82)], hearing impairment [aOR 1.70 (95% CI: 1.02–2.81)] and cerebral palsy [aOR 2.48 (95% CI: 1.03–5.99)].

Conclusion

Neonates surviving sepsis are at a higher risk of poorer neurodevelopment. Current evidence is limited by significant heterogeneity across studies, lack of data related to long-term neurodevelopmental outcomes and term infants.

Introduction

Sepsis is a major cause of mortality and morbidity among neonates [1,2,3,4]. Young infants especially neonates, defined by age < 28 days old, have a relatively immature immune system and are susceptible to sepsis [5, 6]. Annually, there are an estimated 1.3 to 3.9 million cases of infantile sepsis worldwide and up to 700,000 deaths [7]. Low-income and middle-income countries bear a disproportionate burden of neonatal sepsis cases and deaths [7, 8]. While advances in medical care over the past decade have reduced mortality, neonates who survive sepsis are at risk of developing neurocognitive complications, which affect the quality of life for these children and their caregivers [9].

Previous reviews evaluating neurocognitive outcomes in neonates with infections or sepsis have focused on specific types of pathogens (e.g., Group B streptococcus or nosocomial infections [10]), or are limited to specific populations such as very low birth weight or very preterm neonates [11], and there remains paucity of data regarding neurocognitive outcomes among term and post-term neonates. There remains a gap for an updated comprehensive review which is not limited by type of pathogen or gestation. In this systematic review, we aim to provide a comprehensive update to the current literature on the association between sepsis and the following adverse neurocognitive outcomes (1) mental and psychomotor delay (cognitive delay (CD)), (2) visual impairment, (3) auditory impairment and (4) cerebral palsy (CP) among neonates [11].

Methods

We performed a systematic review using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines [12]. This study protocol was registered with Open Science Framework (https://doi.org/10.17605/OSF.IO/B54SE).

Eligibility criteria

We identified studies which evaluated neurocognitive outcomes in neonates less than 90 days old (regardless of gestational age) with sepsis. While the neonatal period is traditionally defined to be either the first 28 days postnatally for term and post-term infants, or 27 days after the expected date of delivery for preterm infants [13], serious late onset infections in the young infant population can present beyond the neonatal period [14], hence we defined the upper age limit as 90 days old to obtain a more complete picture of the burden of young infantile sepsis [15]. Post-term neonates was defined as a neonate delivered at >  = 42 weeks of gestational age in this study [16]. We included studies that either follow international sepsis definitions such as Surviving Sepsis Campaign guidelines definitions [17], or if they fulfilled clinical, microbiological and/or biochemical criteria for sepsis as defined by study authors. The primary outcome of interest was impaired neurocognitive outcome defined by the following domains of neurodevelopmental impairment (NDI) [11]: (1) CD, (2) visual impairment, (3) auditory impairment and (4) CP. We selected these domains because they were highlighted as key neurocognitive sequelae after intrauterine insults in a landmark review by Mwaniki et al. [18]. The authors’ definitions of these outcomes and their assessment tools were captured, including the use of common validated instruments (e.g., a common scale used for CD is the Bayley Scales of Infant Development (BSID) [19] while a common instrument used for CP was the Gross Motor Function Classification System (GMFCS) [20]. Specifically for BSID, its two summative indices score – Mental Development Index (MDI) and Psychomotor Development Index (PDI) were collected. The MDI assesses both the non-verbal cognitive and language skills, while PDI assess the combination of fine and gross motor skills. The cut-off points for mild, moderate and severe delay for MDI and PDI were < 85 or < 80, < 70 and < 55 respectively [21]. There were no restrictions on duration of follow-up or time of assessment of neurocognitive outcomes to allow capturing of both short- and long-term neurocognitive outcomes.

Case–control, cohort studies and cross-sectional studies published between January 2011 and March 2023 were included. Because the definition and management of sepsis has evolved over the years [22], we chose to include studies published from 2011 onwards. Case reports, animal studies, laboratory studies and publications that were not in English language were excluded. Hand-searching of previous systematic reviews were performed to ensure all relevant articles were included. To avoid small study effects, we also excluded studies with a sample size of less than 50 [23].

Information sources and search strategy

Four databases (PubMed, Cochrane Central, Embase and Web of Science) were used to identify eligible studies. The search strategy was developed in consultation with a research librarian. The first search was conducted on 4 December 2021 and an updated search was conducted on 3 April 2023. The detailed search strategy can be found in Supplementary Tables 1A and B.

Study selection process

Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia) [24] was utilized during this review. Five reviewers (WJO, BJY, NM, CLN and GH) independently conducted the database search and screened the title and abstracts for relevance. Following training on inclusion and exclusion eligibility, 4 reviewers (WJO, NM, CLN and GH) subsequently assessed the full text of shortlisted articles for eligibility. All full texts were independently assessed by at least 2 reviewers. Any conflict related to study eligibility were resolved in discussion with the senior author (S-LC). We recorded the reason(s) for exclusion of each non-eligible article.

Data collection process and data items

Four reviewers (WJO, NM, CLN and GH) independently carried out the data extraction using a standardized data collection form, and any conflict was resolved by discussion, or with input from the senior author (S-LC). A pilot search was performed for the first 200 citations to evaluate concordance among reviewers and showed good concordance among reviewers of 94%. For studies with missing data required for data collection or meta-analyses, we contacted the corresponding authors of articles to seek related information. If there was no reply from the authors, the data were labelled as missing.

Study risk of bias assessment

Three reviewers (BJY, GH and WJO) independently carried out the assessment of risk of bias using the Newcastle–Ottawa Scale (NOS) for all observational studies [25]. Studies were graded based on three domains namely, selection, comparability and outcomes. Studies were assigned as low, moderate and high risk of bias if they were rated 0–2 points, 3–5 points and 6–9 points respectively. Any conflict was resolved by discussion or with input from the senior author (S-LC).

Statistical analysis

All outcomes (i.e. CD, visual impairment, auditory impairment and CP) were analysed as categorical data. Analyses were done for each NDI domain separately. To ensure comparability across scales, results from different studies were only pooled if the same measurement tools were used to assess the outcomes and hence sub-group analyses were based on different scales and/or different definitions of neurocognitive outcomes used by authors. Both unadjusted and adjusted odds ratios (aOR) and/or relative risk (RR) for each NDI domain were recorded. Where source data were present, we calculated the unadjusted OR if the authors did not report one, together with the 95% confidence interval (CI). For adjusted odds ratio, these were extracted from individual studies and variables used for adjustment were determined at the individual study level.

Meta-analysis was conducted for all outcomes that were reported by at least 2 independent studies or cohorts. Studies were included in the meta-analysis only if they reported outcomes for individual NDI domains within 30 months from sepsis occurrence. For each domain, all selected studies were pooled using DerSimonian-Laird random effects model due to expected heterogeneity. Studies were pooled based on adjusted and unadjusted analyses. Case–control and cohort studies were pooled separately. The pooled results were expressed as unadjusted odds ratio (OR) or adjusted odds ratio (aOR) with corresponding 95% confidence interval (95% CI). If there was more than 1 study that utilized the same population, we only analysed data from the most recent publication or from the larger sample size, to avoid double counting. Standard error (SE) from studies with multiple arms with same control group were adjusted using SE = √(K/2), where K refers to number of treatment arms including control [26]. Heterogeneity across studies was evaluated using the I^2 statistic, for which ≥ 50% is indicative of significant heterogeneity. With regards to publication bias, this was performed using Egger’s test and funnel plots only if the number of studies pooled were 10 or more for each outcome.

For neurocognitive related outcomes, subgroup analyses were performed based on the severity of the NDI domain outcomes and distinct, non-overlapping populations of septic infants (such as late onset vs early onset sepsis, culture positive sepsis vs clinically diagnosed sepsis, term and post term patients).

All analyses were done using ‘meta’ library from R software (version 4.2.2) [27]. The statistical significance threshold was a two tailed P-value < 0.05.

Certainty of evidence

The certainty of evidence for outcomes in this review was performed during the GRADE criteria [28] which is centred on the study design, risk of bias, inconsistency, indirectness, imprecision, and other considerations.

Results

Study selection

From 7,909 studies identified, a total of 24 articles were included (Fig. 1) [29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52]. A total of 101,657 and 19,988 preterm and term infants were included in this review.

Fig. 1

PRISMA flowchart of the study selection process for search

Study characteristics

There were 2 case–control studies and 22 cohort studies, with a total of 121,645 infants (Table 1). Studies were conducted in 16 different countries (Fig. 2), with the most studies conducted in the United States of America (USA) (7 studies, n = 92,358 patients) [30, 33, 37, 41, 42, 47, 52]. There were no studies that were conducted solely on term infants. 5 studies reported data specifically on ELBW infants (27,078 infants) and 6 studies on VLBW infants (3,322 infants). All studies were performed among neonates.

Table 1 Study characteristics

Fig. 2

World map depicting distribution of studies that evaluate neurocognitive outcomes in infantile and neonatal sepsis

Risk of bias 

Overall, all 24 studies were classified as low risk (Supplementary Table 2). 5 papers scored high risk for outcome bias for having greater than 10% of initial population being lost to follow-up [29, 32, 40,41,42].

Outcome measures reported by domain

As the number of studies pooled for each outcome was less than 10, publication bias was not analysed in the meta-analyses.

Cognitive delay (CD)

Among 24 studies that assessed for CD, 16 studies reported either the incidence of CD among young infants with sepsis compared to those without, and/or the odds ratio (adjusted and/or unadjusted) comparing the two populations [29, 31,32,33,34,35,36,37,38, 40,41,42, 45, 46, 48, 49]. The scales used, authors’ definition of CD, incidence of CD among those with sepsis and those without are described in Table 2. The most common tools used for assessment of CD were the Bayley Scales of Infant Development (BSID) (n = 13) and Denver Development Screening Test II (n = 2).

Table 2 Cognitive delay assessment and definition

Infantile sepsis was associated with increased risk of overall CD delays [aOR 1.14 (95%CI: 1.01, 1.28)], overall PDI delay (aOR 1.73 (95%CI: 1.16, 2.58)) and moderate PDI delay [aOR 1.85 (95%CI: 1.01, 3.36)]. Conversely, infantile sepsis was not associated with increased risk for severe PDI delay nor overall MDI delay [aOR 1.30 (95%CI: 0.99, 1.71)] or its subgroups. There were no significant differences in outcomes between different subgroups of infections as well as culture-proven or clinically defined sepsis for either MDI or PDI (Table 8, Fig. 3A and B).

Fig. 3

A Forest plot on adjusted odds ratios for neurocognitive outcomes related to MDI, PDI, visual impairment, hearing impairment and cerebral palsy. B Forest plot on unadjusted odds ratios for neurocognitive outcomes related to MDI, PDI, visual impairment, hearing impairment and cerebral palsy. Legend: MDI: Mental Developmental Index; PDI: Psychomotor Developmental Index. Foot note: Mild MDI or PDI: < 85 or < 80; Moderate MDI or PDI < 70; Severe MDI or PDI < 55

Visual impairment

Seven studies reported data on visual impairment (Table 3) [31, 33, 41, 42, 47, 49]. The most common definition of visual impairment utilized was “visual acuity of < 20/200” (n = 4, 66.7%).

Table 3 Visual impairment assessment and definition

In the meta-analysis, infantile sepsis was associated with significantly increased risk of visual impairment [aOR 2.57 (95%CI: 1.14, 5.82)] but there were no statistically significant differences in visual impairment between subgroups of early or late onset sepsis, and blood culture negative conditions as compared to the non-septic population (Table 8, Fig. 3A and B).

Hearing impairment

Seven studies reported data on hearing impairment (Table 4) [31, 33, 41, 42, 47, 49]. Two studies defined hearing impairment as permanent hearing loss affecting communication with or without amplification [42, 47]. Other definitions included “sensorineural hearing loss requiring amplification” (n = 1), “bilateral hearing impairment with no functional hearing (with or without amplification)” (n = 1), “clinical hearing loss” (n = 1).

Table 4 Hearing impairment assessment and definition

In the meta-analysis, sepsis was associated with increased risk of hearing impairment [aOR 1.70 (95% CI: 1.02–2.81)]. However, in the subgroup analyses, there were no differences in risk of hearing impairment between patients with late onset sepsis as compared to the non-septic population (Table 8, Fig. 3A and B).

Cerebral palsy

Nine studies [29, 32, 33, 41, 42, 47,48,49,50] reported data on CP (Table 5), of which 5 studies [41, 42, 45, 49, 50] used the GMFCS scale. In the meta-analysis, infantile sepsis was associated with significantly increased risk of CP [aOR 2.48 (95%CI: 1.03; 5.99)]. There was no difference in rates of CP among patients with proven or suspected sepsis, as compared with infants with no sepsis (Table 8, Fig. 3A and B).

Table 5 Cerebral palsy assessment and definition

Differences in neurocognitive outcomes between neonates with culture-proven or clinically diagnosed sepsis as well as early or late onset sepsis

Tables 6 and 7 showed data related to differences in neurocognitive outcomes between neonates with culture-proven or clinically diagnosed sepsis as well as early or late onset sepsis. Meta-analyses were not be performed due to significant heterogeneity in definitions of sepsis, time of assessment of outcomes.

Table 6 Characteristics of study categorised by early vs late onset sepsis

Table 7 Characteristics of study categorised by culture positive vs clinical sepsis

Differences in neurocognitive outcomes between term and post-term neonates

There were no studies which evaluated neurocognitive outcomes between term and post-term neonates and infants.

Certainty of evidence

We found that the certainty of evidence to be very low to low for the four main neurocognitive outcomes selected. (Supplementary File 3).

Discussion

In this review involving more than 121,000 infants, we provide an update to the literature regarding young infant sepsis and neurocognitive impairment. Current collective evidence demonstrate that young infant sepsis was associated with increased risk of developing neurocognitive impairment in all domains of CD, visual impairment, auditory impairment and cerebral palsy.

Cognitive delay

In this review, higher rates of cognitive delay were noted among infants with sepsis [29, 31, 33,34,35,36,37,38, 40,41,42, 45, 46, 48, 49, 52]. We found that infants with sepsis reported lower PDI scores (Table 8), which measures mainly neuromotor development. On the other hand, young infant sepsis was not associated with lower MDI scores (Table 8), which assesses cognitive and language development. The pathophysiological mechanism of young infant sepsis and its preferential impact on PDI remains unclear. Postulated mechanisms include development of white matter lesions which may arise from the susceptibility of oligodendrocyte precursors to inflammatory processes such as hypoxia and ischemia [53]. Future studies should look into evaluating the causes of the above findings. A majority of included studies focused on early CD outcomes while no studies evaluated long-term outcomes into adulthood. CD is known to involve complex genetic and experiential interactions [54] and may evolve overtime with brain maturation. Delays in speech and language, intellectual delay and borderline intellectual functioning are shown to be associated with poorer academic or employment outcomes in adulthood [55, 56], and early assessment of CD may not fully reveal the extent of delays. The only study with follow-up to the adolescent phase showed a progressive increase in NDI rate as the participants aged, which provides evidence of incremental long-term negative outcomes associated with infantile sepsis [44]. Moving forward, studies with longer follow-up may allow for further examination of the long-term effects of neonatal sepsis on CD.

Table 8 Meta analysis related to neurocognitive outcomes among infants with sepsis

There were different versions of the BSID instrument (BSID-II and BSID-III) [19, 57, 58]. BSID-II lacked subscales in PDI and MDI scores, leading to the development of BSID-III with the segregation of PDI into fine and gross motor scales and MDI into cognitive, receptive language, and expressive language scales [59]. Although we pooled results of both BSID-II and BSID-III in our study, we recognize that comparisons between BSID-II and BSID-III are technically challenging due to differences in standardised scores [59, 60]. In addition, the BSID-IV was created in 2019 which has fewer items, However, none of our studies utilized this instrument. Future studies should consider this instrument, as well as standardising the timepoints for assessment of CD.

Visual impairment

Young infant sepsis was associated with increased risk of developing visual impairment. This was similar to results noted by a previous systematic review published in 2014 [61] and 2019 [62] which showed that neonatal sepsis was associated with twofold risk of developing retinopathy of prematurity in preterm infants. Specifically, meningitis was associated with a greater risk of visual impairment compared to just sepsis alone [47]. The mechanism of visual impairment has not been fully described although various theories have been suggested, including sepsis mediated vascular endothelial damage, increased body oxidative stress response as well as involvement of inflammatory cytokines and mediators [63, 64].

Hearing impairment

Our meta-analysis showed an increased risk of hearing impairment for young infants with young infants with sepsis. This is consistent with a previous report that found an association between neonatal meningitis and sensorineural hearing loss [65]. One potential confounder which we were unable to account for may have been the use of ototoxic antimicrobial agents such as aminoglycosides. Additional confounders include very low birth weight, patient’s clinical states (e.g. hyperbilirubinemia requiring exchange transfusion) and use of mechanical ventilation or extracorporeal membrane support. To allow for meaningful comparisons of results across different study populations, it is imperative that a standardised definition of hearing impairment post neonatal sepsis be established for future studies.

Cerebral palsy

Our meta-analysis found an association between neonatal sepsis and an increased risk of developing CP. This is also consistent with previous systematic reviews which had found a significant association of sepsis and CP in VLBW and early preterm infants [11]. One study found that infants born at full term and who experienced neonatal infections were at a higher risk of developing a spastic triplegia or quadriplegia phenotype of CP [66]. The pathophysiology and mechanism of injury to white matter resulting in increased motor dysfunction remains unclear and more research is required in this area.

Limitations and recommendations for future research

The main limitation of this review lies in the heterogeneity in the definitions of sepsis, exposures and assessment of outcomes across studies. This is likely attributed to the varying definition of sepsis used in different countries as well as lack of gold standard definitions or instruments for assessment of each component of NDI. A recent review of RCTs [67] also reported similar limitations where 128 different varying definitions of neonatal sepsis were used in literature. Notably, there is a critical need for developing international standardized guidelines for defining neonatal sepsis as well as assessment of NDI such as hearing and visual impairment. Another important limitation relates to the inability to assess quality of neonatal care delivered as well as temporal changes in medical practices which could have affected neurocognitive outcomes for neonates with sepsis. Improving quality of neonatal care has been shown to significantly reduce mortality risk among neonates with sepsis, especially in resource-poor countries [68]. We performed a comprehensive search strategy (PubMed, Embase, Web of Science and CENTRAL) coupled with hand searching of references within included systematic reviews, but did not evaluate grey literature. Future studies should include additional literature databases and grey literature. Another area of research gap lies in the paucity of data related to differences in neurocognitive outcomes between term and post-term neonates with sepsis and future research is required to bridge this area of research gap. Likewise, there are few studies which evaluated differences in neurocognitive outcomes between early or late onset sepsis and outcomes assessed were significantly heterogenous which limits meaningful meta-analyses. Similarly, there was significant heterogeneity in study outcomes, causative organisms and severity of disease.

We found a lack of long-term outcomes and recommend that future prospective cohorts include a longer follow-up duration as part of the study design. This is important given the implication of NDI on development into adulthood. Most data were reported for preterm infants with low birth weight, and there was a paucity of data for term infants in our literature review. Since prematurity itself is a significant cause of NDI [69], future studies should consider how gestational age and/or birth weight can be adequately adjusted for in the analysis.

Apart from the domains of NDI we chose to focus on in this review, there are other cognitive domains classified by the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V) [70] and/or recommended by the Common Data Elements (CDE) workgroup [71]. Future studies may wish to look into the implications of sepsis on other neuro-cognitive domains related to executive function, complex attention and societal cognition which are studied for other types of acquired brain injury [71, 72].

Conclusion

Our systematic review and meta-analysis found that neonates surviving sepsis are at a higher risk of poorer neurodevelopment. However, the evidence is limited by significant heterogeneity and selection bias due to differing definitions used for NDI and for sepsis. There is also a lack of long-term follow-up data, as well as data specific for term and post-term infants. Future prospective studies should be conducted with long-term follow-up to assess the impact of neurodevelopmental impairment among all populations of neonates with sepsis.