Neonatal encephalopathy is a broad term including different etiologies, while hypoxic-ischemic encephalopathy may be used when hypoxia-ischemia is the most likely cause of the encephalopathy (44, 45). In this systematic review, we used the term “newborn encephalopathy” to emphasize our interest in insults surrounding the birth and the uncertainty with underlying causes. However, when discussing the individual studies, we continued the use of their specific terminology. In addition, no consensus definition of neonatal sepsis currently exists (46). Positive blood culture remains the golden standard of diagnosis. However, several neonates are believed to have sepsis without ever having isolated the specific pathogen (47, 48). Again, we opted to continue the terminology and definition of sepsis used in each individual study.

The search strategy was developed by the authors and tested before use. The primary search was conducted 12 April 2021. Pubmed, Embase, Web of Science, and The Cochrane Database were searched by use of subject headings and free texts related to newborns, encephalopathy, and infection (Supplementary B). Our search was limited to human studies. References and citations from each included study were manually scrutinized for additional relevant studies.

The search results from each database were pooled using Endnote X9® and duplicates were removed. Rayyan QCRI was used in the screening process (51). Studies found by search were screened by title and abstract, while references and citations from included studies were screened by title only. All studies that seemingly met the eligibility criteria or provided insufficient information were extracted for full-text analysis. If any information regarding the eligibility criteria was missing, the study was excluded (Supplementary C).

Data were extracted by a predefined data-collection form. To reduce errors and missing data plots, the data form was piloted prior to data extraction. Following items were extracted from each study: (1) title, authors, country, journal, year of publication, references, citations, funding sources, and conflicts of interest; (2) aim of the study, design, setting, and time-period; (3) number and characteristics of the newborns; (4) assessments of encephalopathy and whether the newborns received therapeutic hypothermia; (5) assessments of maternal and newborn infections; (6) assessments of neurological outcome; and (7) statistical methods and results (Supplementary D).

The Newcastle-Ottawa Scale for cohort studies was used to evaluate the risk of bias. The studies were evaluated on the risk of bias in the domains: selection process, comparability between groups, and assessment of outcome. A total of four, two, and three points could be awarded in each domain, respectively. The risk of bias in the studies was rated as “low” when awarded 3–4 points in selection, 1–2 points in comparability, and 2–3 points in outcome; “fair” when awarded 2 points in selection, 1-2 points in comparability, and 2–3 points in outcome; and “high” when awarded 0–1 point in selection or 0 point in comparability or 0–1 point in outcome. When assessing comparability, one point was awarded if the study controlled for malformations. An additional point was awarded if the study controlled for gestational age, birth weight, gender, or metabolic diseases.

No meta-analysis was performed due to the heterogeneity between the studies with regard to the assessment of encephalopathy, infection, and neurological outcome. A narrative synthesis of the results was made in accordance with Popay et al. (52). In studies with inadequate summary measures, a Chi-square test or Fischer's exact test was performed to assess odds ratio with 95% confidence interval. Analyses were performed by GraphPad Prism version 8.0.0 for Mac (GraphPad Software, San Diego, California USA, www.graphpad.com).

The quality of evidence was assessed in accordance with GRADE (53, 54). A total of five domains were evaluated including the risk of bias, directness of evidence, heterogeneity, precision of effect estimates, and risk of publication bias. Due to their observational design, the studies started with an initial rating of low quality. The studies were then downgraded for serious limitations in any of the five domains and would only be upgraded if no downgrading had occurred. Accordingly, the quality of evidence could be rated as high, moderate, low, or very low.

Publication bias was assessed qualitatively based on the characteristics of the included studies as an insufficient number of studies was included for formal test of asymmetry. Selective reporting bias was assessed by comparing the outcomes reported in the method section and the result section of the included studies.

A total of 4,256 studies were identified across all databases. After removal of duplicates, 3,254 studies were screened for inclusion. Of these, 55 studies were evaluated by full-text and seven studies were eligible for inclusion. Two additional studies were included following screening of 1,030 references and citations from already included studies. This led to inclusion of nine studies for this systematic review (55–63). An overview of the selection process is illustrated in the PRISMA flow diagram (Figure 1).

Table 5 provides an overview of the points awarded by the Newcastle-Ottawa Scale. The studies failed to report on several of the selected risk factors for quality assessment and three studies failed to report on any of these (59, 60, 63). Rao et al. (55) excluded newborns with major congenital anomalies and found no difference between newborns with and without early-onset infection with regard to gestational age and gender, but observed that newborns with infections had higher birth weights. Hakobyan et al. (56) found no difference between newborns with and without early-onset sepsis with regard to gestational age, birth weight, and gender. Orrock et al. (58) excluded newborns with major congenital abnormalities. Three studies excluded newborns with either chromosomal or metabolic disorders (57, 61, 62). In total, six studies were classified as having low risk of bias (55–58, 61, 62), while three studies were classified as having high risk of bias (59, 60, 63).

We found that the quality of evidence across all outcomes was very low according to the GRADE assessment. This was mainly due to a concern related to risk of bias within the studies and the small number of included newborns. Furthermore, early MRI findings as proxy measures for neurodevelopment subjected the studies to downgrading for indirectness (53, 54).

Selective reporting bias was not identified. Several of the studies reported no difference in neurological outcome between newborns with and without markers of perinatal infections. Furthermore, the association between perinatal infections and neurological outcome was not the primary aim in several studies (55, 57–63). Therefore, publication bias seems improbable.

Based on the included studies, the presence of perinatal infection does not seem to impact neurological outcome in newborns with encephalopathy. However, the quality of evidence was very low.

Markers of maternal infection have been associated with newborn encephalopathy (6–12). However, we found no further association with worse neurological outcome. Some studies on maternal infection were excluded from this systematic review, despite having some relevance (Supplementary C). Nelson et al. (66) found that the combination of perinatal sentinel events and markers of maternal infection was associated with an increased risk of cerebral palsy compared with perinatal sentinel events alone. Jenster et al. (67) investigated newborns with clinical and biochemical evidence of hypoxia-ischemia including 5-min Apgar score ≤5 and arterial cord pH <7.1 and base deficit >10. By contrast, they found that chorioamnionitis was associated with more favorable MRI findings at around 5 days of age and more favorable neurodevelopment assessed by Bayley-II or III at around 30 months of age. Early-onset infections may be more prevalent in newborns with encephalopathy and are associated with increased mortality (13, 14). However, as with maternal infections, we found no difference in neurological outcome between newborns with encephalopathy with and without early-onset infection. Both included studies on early-onset infections had a limited number of newborns with positive cultures. Hakobyan et al. (56) analyzed the combined number of newborns with documented and suspected sepsis. Only a smaller fraction (<1/3) considered to have an infection had positive culture, making the diagnosis less well defined. Furthermore, the study only included historical controls, which further limits the quality of evidence. Again, some studies on early-onset infection with some relevance for the topic were excluded (Supplementary C). Scheidegger et al. (68) found no difference in the Sarnat Staging during the first days of life between newborns with hypoxic-ischemic encephalopathy with and without early-onset sepsis (69). However, Jenster et al. (67) found early-onset sepsis to be associated with worse neuromotor function at around 30 months of age in newborns with clinical and biochemical markers of hypoxia-ischemia. In addition, when combining newborns with encephalopathy treated with and without hypothermia from a recent randomized controlled trial in low- and middle-income countries, culture-positive early- and late-onset sepsis were associated with increased risk of death or disability at 18 months (39).

Animal studies have suggested that infectious and inflammatory exposure before hypoxia-ischemia exacerbate newborn brain injury (24–31). Several biological mechanisms have been suggested to explain these findings (70–86). However, this systematic review of human studies was not able to substantiate these findings. This may be due to more heterogeneity inherent in clinical compared to experimental studies. A previous study found perinatal sentinel events, as potential causes of hypoxia-ischemia, only to be present in some 15% of newborns with encephalopathy (8). The timing between different perinatal insults may also have influenced the neurological outcome. Animal studies have found that lipopolysaccharides from Escherichia coli administered 4 h, 6 h, and 72 h before hypoxia-ischemia result in more severe brain damage, while lipopolysaccharides administered 24 h before hypoxia-ischemia had neuroprotective effects (24, 25, 86, 87). Therefore, both positive and negative conditioning may occur in the fetal or newborn brain when exposed to multiple insults (88). This may explain the findings of the included study by Harteman et al. (61), who found that newborns with hypoxic-ischemic encephalopathy born by mothers with chorioamnionitis had the highest incidence of both normal and most severe MRI findings.

Several newborns with encephalopathy due to presumed hypoxia-ischemia still develop unfavorable neurological outcomes despite treatment with therapeutic hypothermia (2, 3). Several reviews have postulated that therapeutic hypothermia may be contraindicated in newborns with encephalopathy and infection (19, 20, 89, 90). This systematic review found similar neurological outcome between hypothermia-treated newborns with encephalopathy with and without markers of perinatal infection (55–60, 63). However, due to the very low quality of evidence across all outcomes, this concern may still be valid. To our knowledge, no randomized clinical trial has sufficiently investigated the neuroprotective effect of hypothermia in newborns with encephalopathy and perinatal infection, although no association between hypothermia and the risk of sepsis has been found (2). Therapeutic hypothermia has been shown to delay C-reactive protein response and to suppress white blood-cell count (91, 92). These findings may partly explain the reduced efficacy of hypothermia observed in low- and middle-income countries (where infections may be more prevalent) (38, 39), and the increased risk of mortality and prolonged shock observed in hypothermia-treated adults with meningitis and sepsis (40, 41). Furthermore, animal studies have found hypothermia to have limited neuroprotective effect following lipopolysaccharide-sensitized hypoxia-ischemia (32–37, 93). Both studies on early-onset infections included in this systematic review reported an overrepresentation of Gram-positive bacteria (55, 56). Contrarily, animal studies have found hypothermia to have some neuroprotective effect following sensitization by endotoxins deriving from Gram-positive bacteria (34, 94).

We conducted this systematic review according to the PRISMA guidelines (42). We followed a preregistered protocol. We conducted a systematic search across four different databases. To minimize bias, each step was performed independently by two reviewers including screening of studies, data collection, and risk of bias and quality of evidence assessment. However, several limitations have to be considered. Newborn encephalopathy may arise from different etiologies including hypoxia-ischemia, infection and inflammation, placental pathologies, and more (5). These factors may alone or together affect the fetal or newborn brain. In the included studies, newborn encephalopathy may have occurred due to various factors and interactions. However, most studies presented criteria suggestive of both hypoxia-ischemia and perinatal infection, indicating that these factors were involved to some degree. The assessments of neurological outcomes may also be problematic. Most studies reported MRI findings, which not necessarily correlates with the neurodevelopmental outcome (95, 96). Furthermore, the longest follow-up time in the included studies was around 24 months of age. It would have been interesting to observe whether any differences in neurological outcome would develop throughout childhood and adolescence (97). Furthermore, due to inadequate reporting in several studies, we were unable to reject that the reference groups also contained mothers or newborns exposed to perinatal infections. This have likely biased the studies toward findings of no difference. At last, most included studies also contained small number of newborns and the comparability between newborns with and without perinatal infections was limited. This led to imprecision of the effect sizes and a high possibility of risk of bias. A meta-analysis could have increased the precision; however, the included studies were deemed to heterogenous.