Types of studies

We included randomised (individual and cluster‐randomised) and quasi‐randomised controlled trials of zinc supplementation versus placebo or no intervention in preterm infants with low birth weight. We excluded observational and cross‐over trials.

Types of participants

We included studies that enrolled infants born preterm (gestational age < 37 weeks) and at low birth weight (birth weight < 2500 grams) and admitted to the NICU or the special care unit or a comparable setting after birth. We excluded infants who underwent GI surgery during their initial hospital stay, or had a GI malformation or another condition accompanied by abnormal losses of GI juices, which contain high levels of zinc (including, but not limited to, stomas, fistulas, and malabsorptive diarrhoea). If studies included participants with and without such presumed high GI zinc losses, we contacted study authors to request data on the former and to exclude the other participants from the analysis. If this information was not available, we excluded the respective study as a whole.

Types of interventions

We included zinc supplementation in any formulation, regimen, or dose administered via the enteral route, in addition to a standard nutrition regimen (partial or full enteral feeds, breast milk, or formula) versus placebo or no intervention, starting at any time from birth to hospital discharge. We included trials in which participants received additional macronutrient and micronutrient supplementation and/or multicomponent milk fortification, as long as supplementation was the same in both intervention and non‐intervention/placebo groups, and as long as the only difference between the groups were differences in zinc (and possibly copper) intake.

Types of outcome measures

We included studies even if they did not report all outcomes. If a study did not report all outcomes, we sought further information from trial authors.

Electronic searches

After the initial approval of the review protocol, we conducted a comprehensive search including: Cochrane Central Register of Controlled Trials (CENTRAL 2017, Issue 8) in the Cochrane Library; MEDLINE via PubMed (1966 to 29 September 2017); Embase (1980 to 29 September 2017); and CINAHL (1982 to 29 September 2017) using the following search terms: (zinc), plus database‐specific limiters for RCTs and neonates (see Appendix 1 for the full search strategies for each database). We did not apply language restrictions. We searched clinical trials registries for ongoing or recently completed trials (clinicaltrials.gov; the World Health Organization’s International Trials Registry and Platform www.whoint/ictrp/search/en/, and the ISRCTN Registry).

We conducted a comprehensive updated search in February 2020 including: Cochrane Central Register of Controlled Trials (CENTRAL 2020, Issue 2) in the Cochrane Library; Ovid MEDLINE(R) and Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, Daily and Versions(R) (1 January 2017 to 20 February 2020); and CINAHL (1 January 2017 to 20 February 2020). We have included the search strategies for each database in Appendix 2. We did not apply language restrictions.

We searched clinical trial registries for ongoing or recently completed trials. We searched The World Health Organization’s International Clinical Trials Registry Platform (ICTRP) (www.who.int/ictrp/search/en/) and the U.S. National Library of Medicine’s ClinicalTrials.gov (clinicaltrials.gov) via Cochrane CENTRAL. Additionally, we searched the http://www.isrctn.com/ for any unique trials not found through the Cochrane CENTRAL search.

Searching other resources

We also searched the reference lists of any articles selected for inclusion in this review in order to identify additional relevant articles.

Selection of studies

Two review authors (ES, KE) independently assessed the eligibility of trials against inclusion and exclusion criteria. We selected studies as potentially relevant by screening title and abstract and, if relevance could not be ascertained by the latter method, by retrieving the full text. For all articles identified as potentially relevant in this first step, we assessed their eligibility from their full‐text version independently in accordance with the specified inclusion and exclusion criteria. We resolved disagreements by discussion and documented studies excluded from the review in the Characteristics of excluded studies table, along with the reasons for exclusion.

Data extraction and management

Two review authors (ES, KE) independently extracted data from full‐text articles using a specifically designed spreadsheet to manage the information. We resolved discrepancies through discussion or, if required, via consultation with a third review arbiter. We entered data into Review Manager 5 software and checked data for accuracy (Review Manager 2020). When information regarding any of the above was missing or unclear, we attempted to contact authors of the original reports to clarify and obtain additional details.

Assessment of risk of bias in included studies

Two review authors (ES, KE) independently assessed the risk of bias (low, high, or unclear) of all included trials using the Cochrane ‘Risk of bias’ tool (Higgins 2011), for the following domains.

• Sequence generation (selection bias);

• Allocation concealment (selection bias);

• Blinding of participants and personnel (performance bias);

• Blinding of outcome assessment (detection bias);

• Incomplete outcome data (attrition bias);

• Selective reporting (reporting bias);

• Any other bias.

We resolved any disagreements by discussion or by a third assessor. See Appendix 3 for a more detailed description of risk of bias for each domain.

Measures of treatment effect

We followed standard methods of Cochrane Neonatal for data synthesis, using Review Manager 5 software (Review Manager 2020). We reported dichotomous data or categorical data using risk ratios (RRs), relative risk differences (RDs), and, for significant risk difference, number needed to treat for an additional beneficial outcome (NNTB) or for an additional harmful outcome (NNTH). We obtained means and standard deviations (SDs) for continuous data and performed analysis using mean differences (MDs) and standardised mean differences (SMDs) to combine trials that measured the same outcome using different scales. For each measure of effect, we provided the corresponding 95% confidence interval (CI). For SMDs, we used Cohen's rule of thumb to interpret effect size: 0.2 small effect, 0.5 moderate effect, 0.8 large effect (Cohen 1988).

Unit of analysis issues

We included cluster‐randomised trials in the analyses along with individually randomised trials using an estimate of the intra cluster correlation co‐efficient (ICC) derived from the trial (if possible). We considered it reasonable to combine the results of individually randomised and cluster‐randomised trials if we noted little heterogeneity between study designs, and if the interaction between effect of the intervention and choice of randomisation unit was considered unlikely.     

Dealing with missing data

We contacted authors of all published trials if we required clarification or additional information. In the case of missing data, we described the number of participants with missing data in the Results section and in the Characteristics of included studies table. We presented results only for available participants and explored in the Discussion section implications of the missing data.

Assessment of heterogeneity

We assessed heterogeneity of treatment effects between trials by using the following statistical models (Higgins 2020).

• I² statistic, a quantity that indicates the proportion of variation of point estimates that is due to variability across studies rather than to sampling error (i.e. to ensure that pooling of data is valid). We graded the degree of heterogeneity as none (< 25%), low (25% to 49%), moderate (50% to 74%), or high (75% to 100%).

• Chi² test: a quantity that assesses whether observed variability in effect sizes between studies is greater than would be expected by chance.

When we found evidence of apparent or statistical heterogeneity (I² ≥ 75%, p < 0.1 for Chi²), we assessed the source of the heterogeneity using sensitivity and subgroup analyses to look for evidence of bias or methodological differences between trials.

Assessment of reporting biases

We planned to explore publication bias by using funnel plots if we included at least 10 studies in the systematic review (Egger 1997; Higgins 2020). However, we only included five studies in the Cochrane Review, and so were unable to perform funnel plots.

Data synthesis

We performed statistical analyses according to recommendations of Cochrane Neonatal (http://neonatal.cochrane.org), using Review Manager 5 software (Review Manager 2020). We analysed all infants randomised on an intention‐to‐treat basis (also when authors did not report intention to treat analysis) as well as treatment effects examined in the individual trials described above. We used a fixed‐effect model to combine data, unless we found moderate heterogeneity, in which case we used a random‐effects model. We used the generic inverse variance method to synthesise risk estimates. When we judged meta‐analysis to be inappropriate (i.e. if heterogeneity was judged high, > 75%), we synthesised and interpreted individual trials separately.

Subgroup analysis and investigation of heterogeneity

If sufficient data were available, we undertook the following a priori subgroup analysis to explore potential sources of clinical heterogeneity.

• Gestational age (< 28 weeks, 28 to 32 weeks, > 32 weeks).

• Birth weight (≤ 1500 grams, not limited to ≤ 1500 grams).

• Type of enteral feeds (predominantly or exclusively human milk, predominantly or exclusively formula feeds).

• Dose of zinc supplementation (≤ 3 mg/kg/d, > 3 mg/kg/d).

• Duration of zinc supplementation (≤ four weeks, > four weeks).

• Additional micronutrient supplementation (zinc preparation alone, zinc combined with other micronutrients or vitamins).

Sensitivity analysis

We explored methodological heterogeneity by performing sensitivity analyses (if sufficient data were available). We performed sensitivity analyses by excluding trials of lower quality when we judged them to be at high risk of bias, to assess effects of bias on the meta‐analysis. We defined low quality as lack of any of the following: allocation concealment, adequate randomisation, blinding of treatment, or > 20% loss to follow‐up.

Summary of findings and assessment of the certainty of the evidence

We used the GRADE approach, as outlined in the GRADE Handbook (Schünemann 2013), to assess the certainty of evidence for the following (clinically relevant) outcomes:

• All‐cause mortality:

  • Before hospital discharge (latest time reported at or after 36 weeks' postmenstrual age);

  • Between hospital discharge and neurodevelopmental follow‐up at 18 to 24 months of age (post‐term).

• Neurodevelopmental disability at 18 to 24 months of age (post‐term), defined as a neurological abnormality including any of the following:

  • Cerebral palsy on clinical examination;

  • Developmental delay > 2 SDs below the population mean on a standardised test of development (Vohr 2004);

  • Blindness (visual acuity < 6/60);

  • Deafness (any hearing impairment requiring amplification).

• BPD, according to NICHD criteria, defined as oxygen requirement > 21% at 28 days of life and:

  • breathing room air (mild BPD);

  • oxygen requirement from 22% to 29% (moderate BPD); or

  • oxygen requirement > 30% and/or positive pressure (severe BPD) at 36 weeks' postmenstrual age (for infants born at < 32 weeks' gestation) or at 56 days of life or discharge, whichever was later (for infants born at ≥ 32 weeks' gestation) (Jobe 2001).

• ROP (any stage and stage III or IV);

• Bacterial sepsis (episodes proven by means of positive blood culture);

• NEC (any stage and stage 2 or greater);

• Change in growth between start and end of intervention, from discharge to time of neurodevelopmental follow‐up at 18 to 24 months of age (post‐term) (absolute growth or change in z‐score for weight, length, and head circumference, where z‐score was defined as deviation of an observed value for an individual from the median value of the reference population, divided by the SD of the reference population) (WHO 1995).

Two authors (ES, KE) independently assessed the certainty of the evidence for each of the outcomes above. We considered evidence from RCTs as high certainty, but downgraded the evidence one level for serious (or two levels for very serious) limitations based upon the following: design (risk of bias), consistency across studies, directness of the evidence, precision of estimates and presence of publication bias. We used the GRADEpro GDT Guideline Development Tool to create summary of findings Table 1 to report the certainty of the evidence.

The GRADE approach results in an assessment of the certainty of a body of evidence as one of four grades:

• High certainty: further research is very unlikely to change our confidence in the estimate of effect;

• Moderate certainty: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate;

• Low certainty: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate;

• Very low certainty: we are very uncertain about the estimate.