Summary of findings for the main comparison.

Erythropoietin compared with placebo or no treatment for complications of preterm birth ‐ primary outcomes
Patient or population: preterm infants with low birth weight Settings: NICU Intervention: EPO Comparison: placebo or no treatment
Outcomes	Illustrative comparative risks* (95% CI)		Relative effect (95% CI)	No. of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Placebo or no treatment	EPO
Use of 1 or more red blood cell transfusions (low and high doses of EPO)	High‐risk population		RR: 0.79 (95% CI 0.74 to 0.85)	1750 (19)	⊕⊕⊝⊝ low	Bias: We had concerns about performance bias and detection bias in 10 of the studies. We downgraded the quality of the evidence by 1 step. Heterogeneity/Consistency: I2 for the typical RR was 69% and for the typical RD 62% (both moderate quality). We downgraded the quality of the evidence by 1 step. Directness of evidence: Studies were conducted in the target population. Precision: Because of the large sample size (n = 1750), the point estimate was precise with a narrow 95% CI. Presence of publication bias: The funnel plot was symmetrical for all larger studies.
	694 per 1000	522 per 1000 (0 to 1000)
Any neurodevelopmental impairment at 18 to 22 months' corrected age (in children examined)	High‐risk population		RR: 0.62 (95% CI 0.48 to 0.80)	1130 (4)	⊕⊕⊝⊝ low	Bias: We had concerns about performance bias and detection bias in 1 of the studies, the largest (n = 613) (Song 2016). This study carried a weight of 48.7% in the analysis. We downgraded the quality of the evidence by 1 step. Heterogeneity/Consistency: I2 for the typical RR was 76% (high) and for the typical RD 66% (moderate). We downgraded the quality of the evidence by 1 step. Directness of evidence: Studies were conducted in the target population. Precision: Because of the large sample size (n = 1130), the point estimate was precise with a narrow 95% CI. Presence of publication bias: Although only 4 studies were included in the funnel plot, the funnel plot was symmetrical.
	210 per 1000	128 per 1000 (71 to 438)
Bayley‐II MDI at 18 to 24 months Bayley Scales of Infant Development, Second Edition, yields 2 single age‐standardised composite scores (range 50 to 150): a Mental Development Index (MDI), which measures cognition through sensory perception, knowledge, memory, problem‐solving and early language abilities; and a Psychomotor Development Index (PDI), which assesses fine and gross motor skills.	Mean Bayley‐II MDI ranged across control groups from 84.1 to 94.5.	Mean Bayley‐II MDI at 18 to 24 months in the intervention groups was 8.22 higher (95% CI 6.52 to 9.92)	WMD: 8.22 (95% CI 6.52 to 9.92)	981 (3)	⊕⊕⊝⊝ low	Bias: We had concerns about performance bias and detection bias in one of the studies (Song 2016). We downgraded the quality of the evidence by 1 step. Song 2016 carried a weight in the analysis of 76.2%. Heterogeneity/Consistency: I2 for the WMD was 97% (high). We downgraded the quality of the evidence by 1 step. Directness of evidence: Studies were conducted in the target population. Precision: Because of the large sample size (n = 981), the point estimate was precise with a narrow 95% CI. Presence of publication bias: As only 3 studies were included, we did not prepare a funnel plot.
Necrotising enterocolitis (stage not reported)	High‐risk population		RR: 0.69 (95% CI 0.52 to 0.91)	2639 (15)	⊕⊕⊕⊝ moderate	Bias: We had concerns about performance bias and detection bias in 6 of the studies, especially for Song 2016, the only study that showed a significant reduction in NEC. It carried a weight in the analysis of 47.8%. We downgraded the quality of the evidence by 1 step. Heterogeneity/Consistency: I2 for the typical RR was 0% and for the typical RD 22% (both low). Directness of evidence: Studies were conducted in the target population. Precision: Because of the large sample size (n = 2639), the point estimate was precise with a narrow 95% CI. Presence of publication bias: The funnel plot was symmetrical.
	84 per 1000	57 per 1000 (0 to 143)
*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; EPO: erythropoietin; MDI: Mental Development Index; NICU: neonatal intensive care unit; NEC: necrotising enterocolitis; PDI: Psychomotor Development Index; RD: risk difference; RR: risk ratio; WMD: weighted mean difference.
GRADE Working Group grades of evidence. High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: 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 quality: We are very uncertain about the estimate.