Phelps 2016.

Methods	Prospective, parallel, randomised controlled trial. Infants enrolled in 14 centres in the Eunice Kennedy Shriver NICHD Neonatal Research Network.
Participants	Infants ≤ 29 weeks' PMA (23 0/7 to 29 6/7 weeks' PMA), who weighed at least 400 G, and could receive study drug by 72 H after birth.
Interventions	Myo‐inositol provided by Abbott Nutrition, Columbus, Ohio, USA as an isotonic, preservative and pyrogen‐free, sterile, 5% solution at 10, 40 or 80 mg/kg/day. Intravenous administration converted to enteral when feedings were established, and continued to the first of 10 weeks, 34 weeks' PMA, death or discharge. Total number randomised: 10 mg/kg N = 29; 40 mg/kg N = 30; 80 mg/kg N = 28 Placebo: 5% glucose. Total number randomised: N = 35.
Outcomes	Adverse events were prospectively monitored from 24 hours prior to study drug until 7 days following the final dose (unless discharged sooner), and judged according to a neonatal toxicity table developed for the study. An unfavourable outcome was defined as either type 1 ROP or worse, in either eye, or surgical intervention for severe ROP in either eye. A favourable ROP outcome was assigned if the retinal vessels progressed to full vascularization in both eyes without meeting criteria for severe ROP, or if on 2 consecutive examinations the retinal vessels were in zone III. Infants who did not meet either criterion had all available examinations reviewed by an adjudication committee. Adjudication was conducted by a committee of 3 experienced ophthalmologists not involved with the study and masked to study group assignment. The final ROP status was judged separately in each eye as 'probably favourable', 'probably unfavourable' or 'cannot be determined', and the majority classification was assigned as the adjudicated outcome. At 18 to 22 months' corrected age, infants received a set of standardized examinations of neurologic function and development according to the NRN Follow‐Up Protocol (to be reported separately).
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer generated.
Allocation concealment (selection bias)	Low risk	Computer generated and communicated to research pharmacist.
Blinding (performance bias and detection bias) All outcomes	Low risk	Personal communication from the first author indicates blinded performance and detection bias for all outcomes. Ophthalmologists were blinded during the adjudication process.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Severe ROP data presented for 106 surviving infants and there were 15 deaths, which adds up to 121 infants not 122 infants, which was the number enrolled.
Selective reporting (reporting bias)	Low risk	The protocol was available to us and we did not notice any major deviations from the planned study. The study was registered as: NCT01030575.
Other bias	Low risk	Appears free of other bias.