Docosahexaenoic acid for reading, working memory and behavior in UK children aged 7-9: A randomized controlled trial for replication (the DOLAB II study)

376 children randomized from 1,230 invited (618 consented and screened), across 84 mainstream primary schools and academies in five UK local/unitary authorities (Oxfordshire, Northamptonshire, Buckinghamshire, Milton Keynes, Swindon); recruitment January 2013 to March 2015, follow-up to July 2015; 372 of 376 reassessed at 16 weeks. Planned n was 400, so the trial fell 24 short.

Healthy UK children aged 7-9 underperforming in reading (below the 20th centile on the recalibrated New BAS II or on BAS 3) - deliberately the exact subgroup in which DOLAB I found its effect. Mean baseline word reading standard score 79.6 (sd 6.5), about 1.3 sd below the norm, roughly 27 months behind chronological age; 62.5% male.

Direct, pre-registered, adequately powered replication of DOLAB I by the ORIGINAL AUTHORS, with a published protocol (ISRCTN48803273; protocols.io), CONSORT-compliant reporting, and anonymised data plus Stata syntax posted on the Open Science Framework (osf.io/9ynjf). Same dose (600 mg/day algal DHA), same duration (16 weeks), same taste- and colour-matched corn/soybean placebo. Randomization by Sealed Envelope Ltd with minimization on school and sex plus a 30% random element; blinding verified post hoc and intact. It recruited MORE children (376 v 362) in MORE schools (84 v 74) and targeted the responsive subgroup directly. Powered at 80% for the d = 0.28 observed in DOLAB I subgroup. This is the strongest design in the omega-3-and-learning literature and it is what a replication should look like. FOUR DEVIATIONS the record must carry, because the authors themselves offer them as reasons the studies disagree. (1) The reading instrument was NOT identical: the 2011 phonics reforms had shifted decoding ability, so DOLAB II used a RECALIBRATED New BAS II plus BAS 3 rather than DOLAB I uncalibrated BAS II, and the authors speculate the recalibrated version may be less sensitive to change. (2) Teacher-rated behaviour was demoted from primary (DOLAB I) to SECONDARY outcome. (3) The pre-planned subgroup here was the sub-10th-centile band (n = 213); the sub-20th-centile band was the whole sample. (4) Capsule shells (colorant and gelatine) were changed in January 2014 mid-trial under a protocol amendment. Two further weaknesses: behaviour questionnaires were the only measures with >15% missing data and lost about half the sample (parent change scores n = 187, teacher n = 196, imputed with treatment-group medians); and post-intervention DHA reached 2.9% of blood fatty acids versus 3.8% in DOLAB I, i.e. lower uptake - though the active group did move from 1.6% to 2.9% while placebo did not (p < 0.001), so the intervention demonstrably reached the bloodstream and the null is not an adherence artefact. Industry ties: funded by DSM Nutritional Products, which also supplied product and placebo and is stated to have had no role in design, data collection, analysis, decision to publish or manuscript preparation ("conducted without direct influence of its funder by way of a robust contract"); PM and AJR declare occasional paid consultancy for companies producing or promoting omega-3. Industry-funded and industry-supplied, but independently executed and analysed - hence developer-involved, not developer-led.

The replication FAILED. Reading, working memory and behaviour change scores showed no consistent differences between the DHA and placebo groups; primary-outcome reading change was Active +0.64 (sd 3.7) v Placebo +0.83 (sd 3.6), p = 0.616, and in the pre-planned sub-10th-centile subgroup +1.4 (3.6) v +1.4 (3.7), p = 0.938. The authors report the observed effect size on reading as d = 0.05, implying achieved power of 8% and a required sample of more than 11,500 to detect it. The direction of the residual noise is worth recording because our earlier extraction did not: EVERY statistically significant change-score difference on the behaviour scales favoured PLACEBO, not DHA (parent-rated Anxiety -1.0 v -3.8, p = 0.002; parent-rated Global Emotional Lability -0.9 v -3.2, p = 0.019; teacher-rated Anxiety -0.5 v -3.7, p = 0.005). Conversely, the one signal favouring active was on post-intervention working-memory LEVELS, not change (Recall of Digits Backward difference 1.774 points, 95% CI 0.045 to 3.503) - but Digits Forward was already significantly higher in the active arm AT BASELINE (42.9 v 41.1, p = 0.048), so the post-intervention gap is a baseline-imbalance artefact and the change scores are flat. Under the replication rule this caps any omega-3-for-learning verdict at mixed at best, and given that the target population was the very subgroup DOLAB I identified, the honest reading is that the original subgroup finding was noise.

None - randomized, double-blind, placebo-controlled.

This IS the replication - of Richardson et al. 2012 (DOLAB I, nutr-richardson-2012-dolab-dha-rct) - and it failed on every primary outcome. Two honest caveats travel with that. (1) It is a SELF-replication: same senior authors, same funder line (Martek, later DSM Nutritional Products), same research group. Under the Replication rule that is not an independent replication, so it does not license a stronger verdict than mixed on its own. (2) Cutting the other way, a self-replication has every incentive - reputational and financial - to succeed, and this one was pre-registered with a published protocol, was larger than the original, targeted the original subgroup directly, verified biological uptake by fingerstick blood DHA, and still found nothing. The authors state plainly in print that the trial did not replicate DOLAB I. No independent group has run the same trial.