Effect of nutritionally modified infant formula on academic performance: linkage of seven dormant randomised controlled trials to national education data

Abstract Objective To compare differences in academic performance between adolescents who were randomised in infancy to modified or standard infant formula. Design Linkage of seven dormant randomised controlled trials to national education data. Setting Five hospitals in England, 11 August 1993 to 29 October 2001, and schools in England, September 2002 to August 2016. Participants 1763 adolescents (425 born preterm, 299 born at term and small for gestational age, 1039 born at term) who took part in one of seven randomised controlled trials of infant formula in infancy. Interventions Nutrient enriched versus standard term formula (two trials), long chain polyunsaturated fatty acid (LCPUFA) supplemented versus unsupplemented formula (two trials), high versus low iron follow-on formula (one trial), high versus low sn-2 palmitate formula (one trial), and nucleotide supplemented versus unsupplemented formula (one trial). Main outcome measures The primary outcome, determined by linkage of trial data to school data, was the mean difference in standard deviation scores for mandated examinations in mathematics at age 16 years. Secondary outcomes included differences in standard deviation scores in English (16 and 11 years) and mathematics (11 years). Analysis was by intention to treat with multiple imputation for participants missing the primary outcome. Results 1607 (91.2%) participants were linked to school records. No benefit was found for performance in mathematics examinations at age 16 years for any modified formula: nutrient enriched in preterm infants after discharge from hospital, standard deviation score 0.02 (95% confidence interval −0.22 to 0.27), and nutrient enriched in small for gestational age term infants −0.11 (−0.33 to 0.12); LCPUFA supplemented in preterm infants −0.19 (−0.46 to 0.08) and in term infants −0.14 (−0.36 to 0.08); iron follow-on formula in term infants −0.12 (−0.31 to 0.07); and sn-2 palmitate supplemented formula in term infants −0.09 (−0.37 to 0.19). Participants from the nucleotide trial were too young to have sat their General Certificate of Secondary Education (GCSE) examinations at the time of linkage to school data. Secondary outcomes did not differ for nutrient enriched, high iron, sn-2 palmitate, or nucleotide supplemented formulas, but at 11 years, preterm and term participants randomised to LCPUFA supplemented formula scored lower in English and mathematics. Conclusions Evidence from these randomised controlled trials indicated that the infant formula modifications did not promote long term cognitive benefit compared with standard infant formulas.

exp fatty acids, omega-3/ or fatty acids, essential/ or Dietary Fats, Unsaturated/ or linolenic acids/ or exp fish oils/ or (n 3 fatty acid$ or omega 3).tw. or docosahexa?noic.tw,hw,rw. or eicosapenta?noic.tw,hw,rw. or alpha linolenic.tw,hw,rw. or (linolenate or cervonic or timnodonic).tw,hw,rw. or menhaden oil$.tw,hw,rw. or (mediterranean adj diet$).tw. or ((flax or flaxseed or flax seed or linseed or rape seed or rapeseed or canola or soy or soybean or walnut or mustard seed) adj2 oil$).tw. or (walnut$ or butternut$ or soybean$ or pumpkin seed$).tw. or (fish adj2 oil$).tw. or (cod liver oil$ or marine oil$ or marine fat$).tw. or (salmon or mackerel or herring or tuna or halibut or seal or seaweed or anchov$).tw. or (fish consumption or fish intake or (fish adj2 diet$) exp fatty acids, omega-3/ or fatty acids, essential/ or Dietary Fats, Unsaturated/ or linolenic acids/ or exp fish oils/ or (n 3 fatty acid$ or omega 3).tw. or docosahexa?noic.tw,hw,rw. or eicosapenta?noic.tw,hw,rw. or alpha linolenic.tw,hw,rw. or (linolenate or cervonic or timnodonic).tw,hw,rw. or menhaden oil$.tw,hw,rw. or (mediterranean adj diet$).tw. or ((flax or flaxseed or flax seed or linseed or rape seed or rapeseed or canola or soy or soybean or walnut or mustard seed) adj2 oil$).tw. or (walnut$ or butternut$ or soybean$ or pumpkin seed$).tw. or (fish adj2 oil$).tw. or (cod liver oil$ or marine oil$ or marine fat$).tw. or (salmon or mackerel or herring or tuna or halibut or seal or seaweed or anchov$).tw. or (fish consumption or fish intake or (fish adj2 diet$) The target sample size per randomized group (113 infants) was calculated to permit detection of a 300-g difference in weight between groups at 9 months and a 400-g difference at 18 months (in each case the difference between the 10th and 25th centiles) at 5% significance and 70% to 80% power. This sample size would also permit detection of a 4-point (approximately 0.3 SD) difference in Bayley developmental indices.

NETSGA Bayley Scales at 18 months
The primary hypothesis in respect to development was that there would be a one-third SD difference in Bayley MDI and PDI scores at 18 months between the 2 randomized groups. We calculated that, with 144 infants per group, we had 80% power at 5% significance to detect this difference, and a similar one-third SD difference in developmental scores at 9 months.

LCPUFAP Bayley Scales at 18 months
The calculated sample size (100 infants per randomized group) permitted detection of a 0.4 standard deviation (SD) difference between diet groups with 80% power at 5% significance. Seventy-five subjects per randomized group (the approximate number seen at 18-month follow-up) would permit detection of a 0.46 SD difference between groups with 80% power at 5% significance.

LCPUFAT Bayley Scales at 18 months
A five-point difference in the primary outcome, Bayley MDI at 18 months, was judged clinically relevant and was also chosen based on comparative studies of formula-fed infants versus breastfed infants, from which it has been hypothesised that the difference in developmental scores could relate to LCPUFAs in breast milk. With a targeted maximum of 142 individuals in each group, we were 80% likely to detect such a difference at the 5% level of significance (minimum recruitment was set at 111 individuals per group, to detect the hypothesised difference at 70% power). At 18 months, we had developmental data on 125 individuals in each group, in fact yielding over 95% power to detect a five-point difference, given that the actual SD was lower than that estimated.

IRONT Bayley Scales at 18 months
The study was designed to have 80% power to detect an overall five-point (1/3 SD) difference in Bayley MDI and PDI at 5% significance between (1) the iron fortified follow-on formula and unfortified formula PALMT Stool hardness Sample size was calculated to detect plausible differences in stool hardness and constipation between the study groups at 5% significance and 80% power. This sample size would enable a plausible 0.365-SD difference in radial BMC between randomly assigned groups to be detected at 5% significance and 80% power. NUCLEOT Diarrheal episodes Sample size was initially calculated to detect a 0.5 SD difference in the number of diarrheal episodes between randomized formula-fed groups with 80% power at 5% significance. However, successful recruitment meant that the trial was continued beyond that originally planned to give a power of 0.4 SD difference in outcomes between the randomized groups at 80% power and P < 0.05.

Supplement 5: Robust standard errors used in this study
The trials showed heteroskedasticity and skewed residuals. While this does not bias regression coefficients, it makes the usual method to calculate standard errors (ordinary least squares, OLS) less efficient and can thereby affect the width of the confidence intervals. To address this issue we applied robust standard errors using Stata's vce(robust) option to all analyses for consistency: In the equation above, beta hatOLS is the sample estimator, and K are the cross-products of the variables X in the original regression equation. This equation describes that robust standard errors allow the variance of residuals to vary across the variables X. The standard errors are therefore consistent even if the residuals in the regression model do not have constant variance.

IRONT
High iron (12mg/dl) formula The English national average and standard deviation during the respective year was derived with data from: https://www.gov.uk/government/collections/statistics-gcses-key-stage-4  Secondary outcomes in the NETSGA trial showing mean differences in modified vs standard formula of GCSE English language exam as IRONT trial SD-scores, of Maths and English reading exams as withintrial SD-scores at age 11 years (KS2, final year of primary school) and odds ratios in modified vs standard formula of receiving five or more GCSE grades A* to C (including Maths and English) and of ever being eligible for special educational needs (SEN) support Secondary outcomes in the NETSGA trial showing mean differences in modified vs standard formula of GCSE English language exam as IRONT trial SD-scores, of Maths and English reading exams as withintrial SD-scores at age 11 years (KS2, final year of primary school) and odds ratios in modified vs standard formula of receiving five or more GCSE grades A* to C (including Maths and English) and of ever being eligible for special educational needs (SEN) support Secondary outcomes in the PALMT trial showing mean differences in modified vs standard formula of GCSE English language exam as PALMT trial SD-scores, of Maths and English reading exams as withintrial SD-scores at age 11 years (KS2, final year of primary school) and odds ratios in modified vs standard formula of receiving five or more GCSE grades A* to C (including Maths and English) and of ever being eligible for special educational needs (SEN) support participants in the standard palmitate formula group; SD standard deviation; SMD standardised mean difference; adjusted for: infant sex, birth weight, gestational age, recruitment centre, maternal smoking during pregnancy, and maternal education at birth; covariates and outcomes imputed for missing participants who have not died.

Secondary outcomes in the NUCLEOT trial showing mean differences in modified vs standard formula of
Maths and English reading exams as within-trial SD-scores at age 11 years (KS2, final year of primary school) and odds ratios in modified vs standard formula of ever being eligible for special educational needs (SEN) support NUCLEO NI=99, NC=97 Within-trial standardised grades: SMD 95% CI KS2 Maths (age 11) 0.17 -0.12, 0.46 KS2 English (age 11) 0.00 -0.28, 0.28 Other secondary outcomes: Odds ratio 95% CI Ever qualified for special educational needs 0.50 0.25, 1.01