This review was conducted with the standard methods of Cochrane Handbook for Systematic Reviews of Interventions Version 6.5 (19)). Reporting was in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (20)). The review has been submitted to the International Prospective Register of Systematic Reviews (PROSPERO CRD42024512148).

We searched the following electronic databases: MEDLINE, Google Scholar, EMBASE, the Cumulative Index of Nursing and Allied Health Literature, Clinical Trials.gov, and the Cochrane Central Register of Controlled Trials. We used a predefined algorithm with the search terms neonate, newborn, cerebral oximetry, cerebral regional oxygenation, near-infrared spectroscopy, NIRS, delivery room, resuscitation, transition, after birth, intervention, care, and treatment. Additionally, we performed a manual search of references in articles identified by our search strategy. No language or publication period restrictions were applied, and the search was performed through December 2024. Studies conducted in older patients or in settings outside the delivery room were excluded.

Two review authors (MB and ES) independently screened titles and abstracts assessed full-text articles for eligibility and resolved any disagreements through discussion. Any discrepancies regarding inclusion were resolved through consensus. Then, the full-text articles were retrieved and included based on the eligibility criteria. This systematic review included only RCTs involving preterm infants with a gestational age less than 37 weeks, who underwent resuscitation in accordance with current neonatal resuscitation guidelines and additionally received interventions based on CrSO2 measurements. The primary outcome was survival without cerebral injury. Cerebral injury was defined as IVH any grade and/or cystic PVL diagnosed at any moment until discharge. Secondary outcome parameters were the incidence of necrotizing enterocolitis, retinopathy of prematurity, and bronchopulmonary dysplasia.

Data extraction was performed using a standardized data collection form that included study design, methods, patient characteristics, interventions, and outcomes. We used Microsoft Excel (Version 16, Microsoft Corporation, Redmond, Washington, USA) to document the mode of randomization, allocation concealment, blinding, and adherence to the intention-to-treat principle. Two independent investigators (MB and ES) extracted the data and resolved any discrepancies through discussion and additional review of the case report forms.

The methodological quality of the included trials was assessed using the risk of bias in randomized trials (RoB2) of the Cochrane Collaboration tool, which evaluated the risk of bias (21). The domains assessed included randomization, allocation concealment, blinding, and adherence to the intention-to-treat principle.

We aimed to obtain individual patient data of included studies, which allowed us to perform Bayesian analysis for the primary outcome, e.g., survival without cerebral injury (IVH any grade and/or cystic PVL) as well as for the secondary outcome parameters including mortality, IVH, PVL, necrotizing enterocolitis, retinopathy of prematurity, and bronchopulmonary dysplasia. Individual patient data were recoded into common format and checked for completeness of records, values, and variables, and internal consistency (i.e., out-of-range values), external consistency with published reports, compliance, validity, plausibility, and duplicate entries.

We performed a Bayesian meta-analysis to evaluate the efficacy of interventions based on NIRS monitoring for survival without cerebral injury across two RCTs. A Bayesian hierarchical model was used to describe the posterior distribution of the probability to observe the primary and secondary outcomes. Hereby, the effect sizes of the individual studies are modelled as realizations of an overarching normal distribution. The mean corresponds to the pooled effect size and the variance describes the between study heterogeneity. However, due to the small number of studies included in the meta-analysis and since both RCT are direct replications and stem from the same center, the between-study heterogeneity was not evaluated and set to zero. This approach corresponds to the fixed effect model for meta-analyses (22, 23). A binomial distribution was used as a likelihood function to model the binary data, i.e., primary outcome observed, or primary outcome not observed, accommodating the number of events within the NIRS and control groups.

For the pooled effect size, a weakly-informative uniform prior was used. The uniform prior distribution allows the data to drive the estimates by restricting the probability for the primary outcome to feasible values between zero and one, without favoring particular values within this range. Since a uniform prior was used, a sensitivity analysis was deemed unnecessary. The posterior distribution for the primary outcome effect size was derived by combining the likelihood function and the prior distribution using Bayes’theorem for both, the standard care group as well as the NIRS-guided group.

The computation of the posterior distributions as well as posterior inference was carried out using numerical integration, implemented in Python. Posterior estimates were summarized by the posterior mean, 95% credible interval, and probability of effect in the direction of interest.

Posterior predictive checks were conducted to evaluate model fit, confirming that the observed data were well-represented by the posterior model. Detailed statistical report is available in supplementary material.