Objective To explore the association between hyperglycaemia and adverse outcomes in very preterm infants.
Design Systematic review and meta-analysis. Data were pooled separately for adjusted and unadjusted odds ratios (ORs) using random-effects model. Subgroup analysis was conducted based on study design (cohort and case control).
Main outcome measures Association between hyperglycaemia in preterm neonates (<32 weeks or <1500 g) and mortality and morbidities.
Findings Forty-six studies (30 cohort and 16 case control) with data from 34 527 infants were included. Meta-analysis of unadjusted ORs from cohort studies found hyperglycaemia to be significantly associated with mortality, any-grade intraventricular haemorrhage (IVH), severe IVH, any-stage retinopathy of prematurity (ROP), severe ROP, sepsis, chronic lung disease and disability. However, pooling of adjusted ORs found significant associations only for mortality (adjusted OR (CI): 2.37 (1.40 to 4.01); I2: 36%; 6 studies), ‘Any grade IVH’ (adjusted OR (CI): 2.60 (1.09 to 6.20); I2: 0%; 2 studies) and ‘Any stage ROP’ (adjusted OR (CI): 3.70 (1.55 to 8.84); I2: 0%; 2 studies). Meta-regression analysis found glucose levels >10 mmol/L to be associated with increased odds of mortality compared with <10 mmol/L. Pooled analysis from case–control studies were similar to cohort studies for most outcomes but limited by small sample size. Longer duration of hyperglycaemia was associated with adverse outcomes. GRADE of evidence was ‘Low’ or ‘Very low’.
Conclusion Hyperglycaemia in very preterm infants is associated with higher odds of mortality, any-grade IVH and any-stage ROP. A limitation was lack of availability of adjusted ORs from many of the included studies.
PROSPERO registration number CRD42020193016.
Data availability statement
Data are available on reasonable request. Data are available from the corresponding author and would be provided on reasonable request.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
What is already known on this topic?
Neonatal hyperglycaemia is a common finding in very preterm and very low birth weight infants.
Individual observational studies have explored the associations between hyperglycaemia and adverse outcomes in very preterm infants.
What this study adds?
This systematic review found that hyperglycaemia is significantly associated with mortality, any-grade intraventricular haemorrhage and any-stage retinopathy of prematurity in very preterm infants.
Glucose levels >10 mmol/L were associated with increased odds of mortality compared with <10 mmol/L.
Longer duration of hyperglycaemia was associated with adverse outcomes.
Hyperglycaemia is a common finding in very preterm and very low birth weight (VLBW) infants during their stay in neonatal intensive care units. Various thresholds of blood glucose levels (>7, >8.3, 10 and 12 mmol/L) have been used to define neonatal hyperglycaemia.1–5 A survey found six different definitions of neonatal hyperglycaemia, with majority using a cut-off of 10 mmol/L.6 The incidence of hyperglycaemia varies from 10% to more than 80% in published studies depending on the definition.7–10
Irrespective of the controversy surrounding its definition, various observational studies have explored the association between hyperglycaemia and adverse outcomes in very preterm infants.5 11–16 Evidence from animal models suggests that hyperglycaemia is associated with increased mortality and morbidity such as retinal inflammation, intraventricular haemorrhage (IVH), apoptosis and reduced brain weight (especially of the hippocampus).17–22 A systematic review did not show a significant association (OR 1.08; 95% CI 0.97 to 1.20) between mean glucose levels and retinopathy of prematurity (ROP) in human preterm infants when adjusted ORs were pooled.23 It also reported that upon pooling of three studies there was a “borderline significant association” between duration of hyperglycaemia and ROP (adjusted OR = 1.08, 95% CI: 1.01–1.15, I2 = 49%, P = 0.03). Subsequent to that systematic review,23 many observational studies have evaluated the association of hyperglycaemia with ROP in preterm infants and reported contradictory results in almost equal numbers. Currently, there are no systematic reviews evaluating the association between neonatal hyperglycaemia and other important outcomes such as mortality, sepsis, IVH, periventricular leucomalacia (PVL), chronic lung disease (CLD), necrotising enterocolitis (NEC) and long-term neurodevelopment. Hence, we conducted a systematic review to evaluate the current evidence in this area.
This systematic review was conducted using the COSMOS-E guidance24 and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) statement25 and MOOSE guidelines.26 It was registered on the international prospective register of systematic reviews.
Data sources and searches
Three reviewers independently searched the following electronic bibliographic databases since their inception until August 2020: PubMed, EMBASE (through OVID), EMCARE (through OVID), MEDLINE (through OVID), The Cochrane Library and Google Scholar. The ClinicalTrials.gov website was searched to identify ongoing studies. Grey literature was searched on ‘Opengrey’ and ‘Mednar’ (http://mednar.com/mednar/desktop/en/search.html) databases. PubMed was searched using the following broad keywords: (((preterm infant) OR (neonate)) OR (very low birth weight)) AND ((Hyperglycaemia) OR (hyperglycaemia)). PubMed was also searched using the following Mesh terms: ((“Infant, Premature”[Mesh] OR “Infant, Extremely Premature”[Mesh]) OR (“Infant, Low Birth Weight”[Mesh] OR “Infant, Extremely Low Birth Weight”[Mesh] OR “Infant, Very Low Birth Weight”[Mesh])) AND (“Hyperglycaemia”[Mesh]). Similar terms were used for searching other databases. There were no restrictions on the search with regards to the publication date or language.
The following types of studies were included in the review: (1) cohort and case–control studies that evaluated the association between neonatal hyperglycaemia (present vs absent) and clinical outcomes (present vs absent); (2) randomised controlled trials (RCTs) that provided information on the association between hyperglycaemia and adverse outcomes; (3) studies that evaluated the association between duration of hyperglycaemia and clinical outcomes. Outcomes of interest were (1) mortality before hospital discharge, (2) IVH (any grade), (3) severe IVH (grade III or IV based on Papille’s classification),27 (4) ROP (any stage), (5) severe ROP defined as ≥stage 3 or requiring treatment, (6) CLD (need for respiratory support or oxygen at 36 weeks post-menstrual age), (7) late-onset sepsis (LOS): positive blood culture on a sample collected after 72 hours of birth, (8) PVL, (9) any-stage NEC, (10) NEC ≥stage II (as per modified Bell’s classification28) and (11) long-term developmental outcomes based on validated tools.
Data extraction and quality assessment
Titles and abstracts identified in the initial broad search were read by two independent reviewers. Full-text articles of the potentially eligible studies were read in detail by two reviewers to confirm their eligibility for inclusion. A standardised form was used to extract data. The incidences of the clinical outcomes of interest in the two groups (hyperglycaemia; no hyperglycaemia) were abstracted. If the authors had provided ORs or risk ratios (adjusted or unadjusted) for those outcomes, they were recorded. Information about the association between duration of hyperglycaemia and clinical outcomes (adjusted and unadjusted ORs) was also collected.
All authors were contacted to provide additional information; seven acknowledged our request, of which two13 14 provided additional information. Each included article was reviewed by three independent reviewers to assess the methodological quality using the Newcastle–Ottawa scale (NOS).29
Meta-analysis was performed using the Review Manager V.5.4 (Cochrane Collaboration, Nordic Cochrane Centre, Copenhagen, Denmark) and Stata V.16.0 software (StataCorp. 2019. Stata Statistical Software: Release 16. College Station, TX: StataCorp LLC). We separately pooled the reported adjusted and unadjusted ORs from included studies using the inverse-variance method (Cochrane Handbook section 10.3.3). Subgroup analysis was carried out based on study design (cohort and case control). Random-effects model (DerSimonian and Laird) was used for meta-analysis since heterogeneity was expected. If the published manuscripts of the included studies did not have information on ORs, we used the raw numbers to calculate unadjusted ORs using contingency tables prior to pooling. For dichotomous outcomes, the pooled effect size estimates were presented as pooled ORs with 95% CIs. Qualitative synthesis was done for studies where meta-analysis was not possible. Publication bias was assessed using visual inspection of the contour-enhanced funnel plots,30 Egger’s test31 and Begg’s test32 if ≥10 studies were included for any individual meta-analysis. If these results suggested the possibility of publication bias, non-parametric trim-and-fill analysis was conducted.33 Statistical heterogeneity was assessed using visual inspection of the forest plots and quantified using the I2 statistic. The I2 result was interpreted as follows: 0% to 40%: might not be important; 30% to 60%: may represent moderate heterogeneity; 50% to 90%: may represent substantial heterogeneity; 75% to 100%: considerable heterogeneity (Cochrane Handbook).34 Contingent on availability of adequate data, where necessary, meta-regression was performed35 to evaluate the association between severity of hyperglycaemia and adverse outcomes after adjusting for the gestational age and birth weight. Some studies had given results for different thresholds of blood glucose levels (eg, 8.3 and 15 mmol/L).11 Some had given results separately for duration of hyperglycaemia, for example, within the first 72 hours of life and first 1 week of life.36 In such situations, only the result with least SE was used for pooling. This approach was chosen because if multiple results from the same study were considered as results of separate study while pooling, it would have spuriously exaggerated the sample size. However, a limitation of this approach was the loss of information which was overcome by conducting a sensitivity analysis wherein such multiple results from the same study were considered as results of separate studies while pooling.
Literature search and study selection
A PRISMA flow chart of screening and selection results is shown in online supplemental efigure 1. The initial search identified 1775 articles of which 46 studies were included after application of the selection criteria.1 5 7–9 11–16 36–70 Out of these studies 18, 19, 21, 10, 10, 10, 4 and 3 studies were included in the meta-analysis for mortality, IVH, ROP, LOS, CLD, NEC, PVL and neurodevelopmental outcomes, respectively. The total sample size was 34 527 and the number of infants in individual studies ranged between 30 and 859, but one large multinational database study had a sample size of 24 548. Among the 46 studies, 16 were case–control studies,1 14 15 37 44 45 49 50 52 54 56 61 62 65 69 70 13 were prospective cohort studies7–9 11 13 39 40 47 51 55 57 60 67 and 17 were retrospective cohort studies5 12 16 36 38 41–43 46 48 53 58 59 63 64 66 68 (table 1 and online supplemental etable 1). Ten out of 46 included studies specifically addressed the issue of association between duration of hyperglycaemia and adverse outcomes.14 40 41 47 54 56 58 60 67 70 The median number of stars in the cohort studies as assessed by NOS was 9 (IQR 7 to 9). The median number of stars in the case–control studies as assessed by NOS was 9 (IQR 9 to 9) (tables 2 and 3).
Table 4 provides an overview of results of individual studies. Pooling of unadjusted ORs from cohort studies found hyperglycaemia to be significantly associated with mortality, any-grade IVH, severe IVH, any-stage ROP, severe ROP, LOS, CLD and disability. However, pooling of adjusted ORs from cohort studies found significant associations for mortality (figure 1), any-grade IVH and any-stage ROP (online supplemental efigures 2 and 3). Hyperglycaemia was not associated with NEC, severe NEC and PVL, either on adjusted ORs or unadjusted ORs. The detailed results of meta-analysis are given in table 5 and online supplemental efigures 2–11.
Association between duration of hyperglycaemia and adverse outcomes
Ten studies that examined the association between duration of hyperglycaemia and the adverse outcomes reported a significant association.14 40 41 47 54 56 58 60 67 70 Of them, four studies reported a significant association between duration of hyperglycaemia and ROP,40 41 54 70 of which the association remained significant on multivariate analysis in two studies.40 54 One study reported a significant association between duration of hyperglycaemia and severe IVH on both univariate and multivariate analysis.14 Two studies reported that infants with >5 days of hyperglycaemia were significantly lighter, shorter and had smaller occipital-frontal head circumference at 4 months and 24 months corrected age.58 60 These findings remained significant after correcting for nutritional and illness factors.60 Two studies reported that prolonged duration of hyperglycaemia was associated with poorer cognition, language and motor performance at 1–2.5 years and the association remained significant in multivariate analysis.47 56 We could not pool the aforementioned information in an exclusive meta-analysis since the outcomes of interest were heterogenous.
Results of the sensitivity analysis were similar to the primary analysis (online supplemental etable 2).
The meta-regression analysis found that blood glucose levels ≥10 mmol/L were associated with higher odds of mortality compared with <10 mmol/L (regression coefficient 1.0563 (95% CI 0.2193 to 1.8933), p=0.013) (online supplemental efigure 12). On the other hand, there was no association between blood glucose level and the odds of any-grade IVH (regression coefficient −0.2460 (95% CI −1.3729 to 1.3237), p=0.971) (online supplemental efigure 13). Meta-regression could not be done for the remaining outcomes since there were <10 studies.
Publication bias was assessed only for the outcomes of mortality (unadjusted) and any IVH (unadjusted) since they had ≥10 studies in the meta-analysis. Visual inspection of the contour-enhanced funnel plots and the results of Begg’s test or Egger’s test suggested publication bias was unlikely for unadjusted mortality (Egger test p=0.59, Begg test p=0.49). The contour-enhanced funnel plot for studies reporting unadjusted mortality is depicted in figure 2. The results of Begg’s and Egger’s test (p values 0.02 and 0.01) suggested the possibility of funnel plot asymmetry for unadjusted any-grade IVH (online supplemental efigure 14). The results of trim-and-fill analysis that imputed three additional studies continued to show significant association between hyperglycaemia and any-grade IVH (original OR 2.30 (1.55 to 3.40); new OR 1.85 (1.18 to 2.88)). We revisited our literature search after noticing funnel plot asymmetry but did not find any additional studies.
GRADE of evidence was ‘Low’ or ‘Very low’ for all outcomes (online supplemental etable 3).
This systematic review, which included 46 studies (n=34 527), found that neonatal hyperglycaemia is associated with mortality, any-grade IVH and any-stage ROP based on pooled adjusted ORs in very preterm infants. The evidence was inadequate for other outcomes as very few studies had reported adjusted ORs. While pooled unadjusted ORs suggested a significant association between hyperglycaemia and majority of the adverse outcomes such as severe ROP, late-onset sepsis, CLD and disability, these results are probably less reliable given that the confounding factors were not adjusted for.
There is a physiological plausibility that the duration and the severity of hyperglycaemia might contribute to mortality and morbidity due to proinflammatory effect, changes in osmolality, fluid shifts, and direct cellular effects in various organs, particularly in the fragile preterm brain.
A recent systematic review involving critically ill adult patients found intensive glucose control significantly reduced the risk of all-cause mortality, length of ICU stay and acquired infections.71 A meta-analysis of six studies in critically ill children concluded that tight glycaemic control does not reduce mortality but reduces the need for dialysis.72 Overall, these systematic reviews found tight control group to have higher incidence of hypoglycaemia. A recent observational study including 580 extremely preterm infants concluded that insulin treatment was associated with lower mortality in infants with hyperglycaemia.5 A RCT of prophylactic infusion therapy in 389 VLBW infants reported that insulin reduces hyperglycaemia but may increase the risk of hypoglycaemia.73 In a RCT (n=88) of insulin for treatment of hyperglycaemia in preterm infants (<30 weeks), the ‘tight glycaemic control with insulin’ group showed better weight gain and head growth but reduced linear growth and increased risk of hypoglycaemia.74 However, nearly 64% of the infants in the control group also had received insulin infusion. At 7 years of follow-up, there was no difference in the incidence of survival without disability, but the tight control group had reduced height, increased height-adjusted lean mass and lower fasting blood glucose concentrations.75
The current strategies to treat hyperglycaemia using low glucose infusion rates and insulin therapy are not without problems. Reduced glucose infusion results in poor nutrition delivery which may have consequences for neurodevelopment and growth. On the other hand, insulin therapy can increase the risk of hypoglycaemia, leading to poor outcomes. It is also possible that comorbidities like hypoxia, inflammation, infection or ischaemia causing hyperglycaemia might directly contribute to the morbidity and hence treating hyperglycaemia may not improve outcomes. Strategies such as continuous glucose monitoring to titrate insulin therapy, appropriate insulin therapy to target a liberal glucose level, targeting novel pathophysiological pathways or their combinations need further evaluation.
An important limitation of our review was the lack of data from some studies in a format suitable for pooling, especially for adjusted ORs. Future observational studies should endeavour to report ORs after adjusting for confounders. Another limitation was the presence of statistical heterogeneity in some outcome measures. We tried to address heterogeneity using three approaches: (1) random-effects model in the meta-analysis; (2) meta-regression wherever there were more than 10 studies in the meta-analysis; (3) analysis of cohort and case–control studies separately. While interpreting the results of our review, it is also important to be aware that association does not always mean cause-and-effect relation. The hyperglycaemia might just be a passenger/marker in another disease process rather than being the causative agent. The strengths of our review include its rigorous methodology, separate pooling of adjusted and unadjusted ORs, sensitivity analyses, meta-regression, the use of contour-enhanced funnel plots, formal statistical tests to assess funnel plot asymmetry and the trim-and-fill analysis. To our knowledge, it is the first systematic review that addresses the association between hyperglycaemia and various adverse outcomes in preterm infants.
Neonatal hyperglycaemia in preterm infants is associated with higher odds of mortality, any-grade IVH and any-stage ROP. RCTs evaluating the efficacy and safety of strategies to treat hyperglycaemia are needed.
Data availability statement
Data are available on reasonable request. Data are available from the corresponding author and would be provided on reasonable request.
Patient consent for publication
Contributors CPR and SCR conceptualised and designed the study, data collection instruments, drafted the initial manuscript, carried out the initial analyses, and reviewed and revised the manuscript. MS, SM and CPR collected data, reviewed and revised the manuscript. SP coordinated and supervised data collection, and critically reviewed the manuscript.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.