The Influence of DHA on Language Development: A Review of Randomized Controlled Trials of DHA Supplementation in Pregnancy, the Neonatal Period, and Infancy

This review summarizes randomized controlled trials (RCTs) assessing the effect of docosahexaenoic acid (DHA) supplementation in the first 1000 days on child language. Six databases were searched and RCTs were included if they involved supplementation with DHA during pregnancy, to preterm infants, or during the postpartum period, included a placebo group with less or no DHA, and reported a language outcome. We included 29 RCTs involving n = 10,405 participants from 49 publications. There was a total of 84 language measures at ages ranging from 3 months to 12 years. Of the 84 assessments, there were 4 instances where the DHA group had improved scores, and 2 instances where the DHA group had worse scores (with the majority of these significant effects found within one RCT). The remaining comparisons were null. A few RCTs that included subgroup analyses reported (inconsistent) effects. There was limited evidence that DHA supplementation had any effect on language development, although there were some rare instances of both possible positive and adverse effects, particularly within population subgroups. It is important that any subgroup effects are verified in future trials that are adequately powered to confirm such effects.


Introduction
The first 1000 days of life is a period of rapid brain development where embryonic stem cells develop into a functioning brain that is 80% of the size of an adult brain [1]. Although the brain continues to develop well into early adulthood, the foundations are laid for later development during this critical period [2]. Appropriate nutrition in early life is considered to be one of the most important (non-genetic) influences on brain development [2][3][4][5][6][7]. The omega-3 long chain polyunsaturated fatty acid, docosahexaenoic acid (DHA, 22:6n−3) is one nutrient that is concentrated in neural tissues, and is actively accumulated in the brain during early development [8][9][10].
Observational studies have demonstrated associations between intake of foods rich in DHA (oily fish) with positive outcomes for child development [11][12][13][14][15][16][17][18]. A number of randomized controlled abstracts of all articles retrieved by the search were screened to assess eligibility. The reference lists of eligible articles identified by the search were also checked for other potentially relevant articles. In addition, we searched the reference lists of relevant reviews [19][20][21][22][23]40]. Search engines used were set up to email new publications identified by the search on a monthly basis, with new articles added to the review up until acceptance of the manuscript.

Inclusion and Exclusion Criteria
To be eligible for inclusion, a trial had to be in English, be conducted in humans, have a RCT design, include supplementation with DHA (where trials that included long-chain polyunsaturated fatty acids (LCPUFAs) in conjunction with DHA were also considered), include a placebo group without, or with less, DHA, with the intervention occurring during pregnancy, the postpartum period, or during infancy, and report a language outcome.

Data Extraction
Three authors were involved with reviewing search results and extracting data from included articles (J.F.G., N.R.G., and A.J.A.). Where relevant data or details were absent in an article, study authors were contacted. Study characteristics extracted included authors, publication year, sample information, intervention details, and language assessment and results. Possible sources of bias in each trial were noted.

Data Synthesis
Included studies were grouped according to the intervention period as (1) maternal prenatal interventions; (2) interventions for preterm infants; (3) postnatal interventions for breastfeeding mothers; and (4) postnatal interventions for infants. Although preterm neonates are typically supplemented during the same period of brain development as in trials where supplementation took place prenatally, they are best considered separately. Infants born preterm undergo their fetal brain growth spurt ex-utero and, hence, often have different developmental characteristics to term-born infants, such as poorer cognitive development and increased risk of behavioral problems [45][46][47][48][49][50]. Interventions that commenced either during pregnancy, or to preterm infants and continued through to infancy, were considered to be prenatal and preterm infant interventions, respectively.
Data were extracted into a table of characteristics of the included RCTs with the overall language assessment results. In order to comprehensively capture the language development of young children, language outcomes derived from multiple sources were included. Results of language assessments were categorized and discussed as: (a) Clinician-administered global language measures (including language-specific subscales of global developmental tests); (b) Parent-rated assessments of global language or domain specific language abilities (including language, communication or verbal scores from parent-completed assessments of non-language domains such as global development or behavior; (c) Assessments of language-based cognitive abilities; (d) Language-based academic abilities; (e) Other language measures not classified above, particularly experimental measures; (f) Subgroup effects (where reported) such as for sex, socio-economic status or birthweight.

Results
The initial search resulted in 1304 articles, 634 of which remained after duplicates were removed. Article abstracts were reviewed and 593 were excluded for the following reasons: not human trials (n = 63), did not involve omega-3 LCPUFA supplementation (n = 127), were not RCTs (n = 302) or did not have a language outcome (n = 101). The full text of 41 citations were then examined in detail and further exclusions were made primarily due to a focus on the effects of supplementation outside the first 1000 days (n = 16). Subsequent searches identified 13 subsequent eligible studies. A total of 29 studies were identified as eligible for inclusion, with relevant trial outcomes or details published across 49 articles.

Interventions for Preterm Infants
There were 8 RCTs that supplemented n = 2214 preterm infants [24,54,56,57,[64][65][66][67][68][69][70]84,85]. The intervention was administered to infants through supplements for breastfeeding mothers, supplemented infant formula, through enteral feeds, or as sprinkles or a dissoluble powder that could be added to milk or food. It is noteworthy that in some trials, interventions in exclusively formula-fed infants compared formulas containing some DHA, to formulas that contained no DHA [64][65][66][67]. Other trials compared a low or standard dose of DHA to a higher dose of DHA [54,56,57,[68][69][70]84]. Only one trial comparing low-dose DHA to high-dose DHA, included breastfed babies as well as formula fed babies [68][69][70]. All but one study [24] commenced the intervention within the first week of birth. Three of the trials finished the intervention by the time the infant was discharged from hospital (at around term-equivalent age) [54,56,57,65,[68][69][70]84]. Three trials continued the intervention past discharge, until infants were aged 9 [66,67], 12 [64], or 24 months of age [85]. One trial administered the intervention later in infancy, commencing at 10 to 16 months of age, once formula feeding and breastfeeding had ceased [24]. The doses administered were largely dependent on the amount of breastmilk or formula the infants were able to consume on a daily basis and hence varied within intervention groups.

Postnatal Interventions for Breastfeeding Mothers
Three trials assessed DHA interventions for breastfeeding women in n = 765 mother-infant pairs [32,53,62,63]. In one trial mothers were provided with capsules, and in another trial mothers had the option of capsules, muesli bars, or cookies with the supplementation period commencing within 5 days of birth and lasing up until 4 months of age [32,62,63]. The dosage of daily DHA was 200 mg in one of these trials [62,63], and was not reported in the other trial [32]. One trial targeting women with habitual low fish intake (for the population) additionally included a reference group of non-randomized breastfeeding women with high habitual fish intake [32]. The third trial involved randomizing breastfed infants aged 6 to 12 months to 4 intervention groups for a 12-month intervention [53]. In this study, mothers were provided with capsules containing 215 mg DHA, and infants were provided with a corn-soy blend complimentary food supplement that included 19 micronutrients (with or without 285 mg DHA depending on randomization group) [53]. One group received study products containing DHA for mothers only, one group received DHA study products for both mothers and children, one group received DHA in the child supplement only and the fourth group received capsules and child complimentary food supplements devoid of DHA [53]. The four groups were compared individually, rather than combined. This trial was conducted in Ethiopia where dietary DHA intake is reported to be habitually low [53].

Possible Sources of Bias
Trials generally had adequate sequence generation, treatment allocation and blinding of participant processes. One trial did not appear to adequately blind participants as 92.9% of participants correctly guessed their infant's group allocation [74]. In this trial as well as one other there was higher attrition from the treatment group compared with the control group [74,75]. There was some evidence of reporting bias, where one trial with a relevant assessment was reported in an abstract in 2010 and is yet to be published in full [55], although other outcomes from this trials have been published [91,92].

Assessments of Language Abilities
Child language was assessed with a total of 83 assessments at a variety of different ages between 3 months [82] and 12 years [76].
(a) Clinician-administered global language measures (including language-specific subscales of global developmental tests).
The Peabody Picture Vocabulary Test (PPVT), as used in five trials, [28,29,31,73,75,80] is designed to measure receptive vocabulary for individuals 2.5 years an older. The Clinical Evaluation of Language Fundamentals (CELF) assesses general language ability and included studies used editions for preschool children [77] as well as school aged-children [78,89]. The Clinical Linguistic and Auditory Milestone Scale (CLAMS) provides a language development quotient that was used in one trial at 2 ages, but with limited details available about the assessment [62].
As language is a key ability that emerges in early childhood, many general developmental tests designed to detect developmental delays in infants and toddlers specifically include a measure of early overall language development. The Bayley Scales of Infant Development (Bayley) is one of the most widely used development assessments for young children. It was developed for infants and toddlers (up to~3.5 years of age) and has multiple editions. The Bayley-III was used in nine studies [24,25,28,54,74,82,83,85,90] and includes a specifically designed, standardized global Language Scale score made up of a receptive and expressive language score. Bayley editions I and II combine cognitive and language development into one mental scale although authors in 2 trials calculated a language score from this (which has been included in this review) [30,87]. The Griffiths Mental Development Scale (GMDS) [75] is a global developmental test from birth to 8 years that includes a Speech and Hearing subscale that captures expressive and receptive language. The Knobloch, Passamanick and Sherrard's Developmental Screening Inventory (KPSDSI) is a global developmental assessment for infants and young children with a language subscale that was used in three trials for infants aged 9 months [65,66,86]. The Denver Developmental Screening Test (Denver) is a global developmental test that includes a language subscale. The Denver test was culturally adapted for use in Ethiopia and included items that were clinician administered as well as parent-report items [53].
(b) Parent-rated assessments of global language or domain-specific language abilities (including language, communication or verbal scores from parent-completed assessments of non-language domains, such as global development or behavior).
There were two parent-rated measures designed specifically to measure language abilities. The MacArthur-Bates Communicative Development Inventories (MCDI) are age-standardized forms to capture developmentally appropriate comprehension, non-verbal communication, vocabulary and early emergence of grammar depending on the age of the child. The MCDI was used in nine studies and involves a Words and Gestures form for children 8-18 months and a Words and Sentences form for children age 16-30 months [28,30,32,64,69,71,72,74,80]. The Children's Communication Checklist (CCC) likewise specifically targets language and was developed as a screen for communication difficulties (both expressive and receptive) and pragmatic impairments in children aged 4 to 16 years and was used in two trials [76,89].
There were two general developmental questionnaires that included a language outcome, and one behavior questionnaire that had a language survey. The Child Development Inventory (CDI) is a parent-rated measure of global child development between the ages of 1 and 6 years. Along with an overall developmental score, there are subscales for Expressive Language, Language Comprehension, and Letters. A German-translation was used in one of the included trials [58]. The Ages and Stages Questionnaire (ASQ) is also designed as a global measure of general development for children aged 1 to 66 months. A subscale for Communication is included and was reported in two trials [26,57]. The Child Behavior Checklist (CBCL) is a measure of behavioral development and behavior problems with editions for young children (1.5 to 5 years) and older children (6-18 years). The questionnaire for young children includes a Language Development Survey where parents indicate the number words in their child's vocabulary and was used in 2 of the include trials [74,75].
(c) Assessments of language-based cognitive abilities.
The majority of trials that evaluate the effect of a DHA intervention on neurodevelopment include a test of cognition, and many of these tests include an assessment of language-based cognitive abilities. We identified 12 trials that reported the results of a language-based cognitive ability; five prenatal trials [52,55,[76][77][78]80], two trials in preterm infants [67,70], 1 in breastfeeding mothers [63], and in four postnatal interventions for infants [29,60,61,88].
Three IQ tests commonly used in child and young-adult populations were outcome assessments in the included trials; the Wechsler Abbreviated Scale of Intelligence (WASI) was used in five trials [60,67,70,78,84], the Wechsler Preschool and Primary Scale of Intelligence (WPPSI) was administered in six trials [29,55,61,63,80,88], and the Wechsler Intelligence Scale for Children (WISC) was conducted in one trial [76]. All include a Verbal Comprehension Index (an equivalent of a VIQ score) capturing crystallized or verbal abilities. The Differential Ability Scales (DAS) is an assessment of global cognitive functioning ability, similar to an IQ test. The DAS includes a Verbal Scale Score that measures verbal-based cognitive abilities [77]. The McCarthy Scales of Children's Abilities (MSCA) is a cognitive development assessment for young children with a Verbal subscale that was used at 5 years in one prenatal trial [52]. The Developmental NEuroPSYchological Assessment (NEPSY) is a general cognition test battery for school-age children with a language domain that was used in two trials [60,67].
There were five reports of a language-based academic assessment; two in prenatal trials [78,80], two in preterm infants [67,70], and one in postnatally supplemented infants [31]. All language-based academic assessments were administered by a clinician. The Wide Range Achievement Test (WRAT) includes a Word Reading task and a Spelling task and was performed in one prenatal trial [78] and one preterm infant trial [70]. The Wechsler Individual Achievement Test (WIAT) is an educational assessment with a test each of Word Reading, Spelling, and Pseudoword decoding that was administered in one preterm infant trial at 10 years [67]. The Test of Preschool Early Literacy (TOPEL) is designed to detect early literacy problems in preschool aged children. The TOPEL includes subtests of vocabulary, phonological awareness, and print knowledge (knowledge of written language) and was used in one trial to assess 42-month-old infants [80]. The revised version of the Bracken Basic Concept Scale (BBCS) similarly assesses language-based abilities associated with school-readiness [31]. The BBCS was administered to 2.5-year-old infants to capture acquisition of basic concepts and aspects of receptive language [31].
(e) Other language measures not classified above, particularly experimental measures.
There were four trials that included an alternate, or more experimental measure of language. One trial measured mean length of utterances (MLU) at 39 months of age [73]. A mother and child were recorded during~15-30 min of conversational free-play. Taped sessions were transcribed and coded to obtain the MLU score. One trial administered the Renfrew Bus Story at 6 years of age [89]. This task required the children to retell a narrative with correct sentence length, complexity and vocabulary. A prenatal trial assessed Sentence Repetition at 36, 42 and 48 months of age [80]. Children were required to repeat verbal sentences verbatim, despite increasing length and complexity [80]. One trial administered a Recognition task requiring the infants to react to English and non-English consonants at 9 months [27] and a word-object pairing task to the same infants at age 16 months of age [28] that were not otherwise described.
(a) Of the 10 trials with prenatal interventions, there were four that conducted an assessment of language through a clinician [28,75,77,78,80]. The PPVT was administered at 2.5 years [75], 60 months [80] and 5-6 years of age [28] with no evidence of an effect of DHA supplementation at any age. The CELF-P2 was conducted with 4-year-old children [77] and then the CELF-4 was used with the same children at 7-years of age, with no effect of DHA intervention detected [78].
There were an additional 5 trials that included a global developmental assessment that contained an assessment of overall language abilities [25,28,75,82,83]. The Bayley-III language subscale was administered at 3 months [82], 4 months [83], 12 months [83], and 18 months [25,28] of age, between four RCTs with null effects reported. The GMDS speech and hearing subscale revealed no group differences at 2.5 years of age [75].
(b) Of the 10 RCTs conducted during pregnancy, there were three that included a parent-rated measure of language [28,76,80]. The MCDI was completed three times by parents, once at 14 months [28] and in two trials at 18 months [28,80] of age with no evidence of a benefit of prenatal DHA supplementation. The CCC was administered at 12 years of age with no group difference [76].
There were four parent-rated measures of development in general [26,58], and one behavioral questionnaire that included a language subscale [75]. The CDI was completed by parents at 4 years and again at 5 years of age, with no group differences in expressive language or language comprehension [58]. The general development ASQ was completed by parents at 4 months and 6 months of age and similarly revealed no group effects [26]. A language survey included in a behavior questionnaire completed when children were 2.5 years old showed no effect of DHA intervention [75].
(c) There were six IQ or cognitive tests with a VIQ or equivalent score reported between the 10 prenatal trials [55,[76][77][78]80]. The WPPSI-III was administered at 46 months of age in one study [55] and at 48 months of age in another [80], with no evidence of an effect found in either. The DAS-II was administered at 4 years of age [77], and the WASI-II was administered to these same children at 7 years of age [78] with no effect of prenatal DHA supplementation reported. The WISC-IV was conducted at 12 years of age with no group differences found [76]. The MSCA administered at 5 years of age likewise detected no difference in group scores [52]. (d) Two out of the 10 prenatal DHA trials assessed language-based academic abilities. One trial assessed preschool aged children [80], and the other trial assessed 7-year-old children [78]. No differences were found with these academic measures of reading, spelling, vocabulary, phonological awareness, or knowledge of written language [78,80]. (e) Two of the 10 prenatal studies included an alternate/experimental assessment of language [27,28,80]. In one trial, the assessment was repeated at three ages [80], whilst the other trial used differing alternative language measures at different ages [27,28]. Neither trial found a group difference on any of the measures at any of the ages [27,28,80]. (f) Of the 10 prenatal RCTs, two reported conducting a subgroup analysis [25,52,77,78,94]. Subgroup analyses were conducted for sex [25,77,78], maternal education [94], maternal smoking during pregnancy [94], and stimulation in the home environment [52]. Subgroup analyses for sex showed poorer language scores, and greater risk of delayed language (score < 85) within girls only at 18 months of age [25], although no sex by treatment effects were detected in the assessments at 4 and 7 years of age [77,78]. Subgroup analyses with maternal education in the same trial revealed no effect of DHA supplements amongst women who had not completed tertiary education. However, lower language scores at 18 months of age were observed with DHA supplementation in women who had completed tertiary education [94]. Trial authors likewise explored whether there was a smoking (during pregnancy) by treatment effect, but found none [94]. In one trial, authors explored an interaction effect for quality of stimulation in the home environment during childhood [52]. For children in the DHA group, the home environment appeared to have less influence on developmental outcomes than for children in the control group [52].
(a) Four of the eight trials conducted in preterm infants included a clinician-administered global assessment with an overall language subscale [24,54,65,66,85], although none assessed global language. The Bayley-III was administered at 16-22 months of age in one trial [24], at 2-3 years in one trial [54], and at both 12 and 24 months of age in another trial [85], with no differences detected in either study. Two studies used the KPSDSI at 9 months and likewise detected no effect of the DHA intervention [65,66]. (b) Three of the eight RCTs in preterm infants included a parent-rated measure of child language [57,64,69]. Two RCTs included a general language measure completed by parents, the MCDI, administered at 9 months [64], 14 months [64], and at 26 months [69] of age. No evidence of a benefit of DHA intervention was detected in either trial, at any of the ages administered [64,69]. Parents completed a global measure with a communication subscale in one trial at 6 and again at 20 months of age, with no hint of an effect of DHA intervention at either age [57]. (c) Between the eight RCTs conducted in preterm infants, there were three that included a language-based cognitive assessment [67,70,84]. All trials administered the WASI, at 7 years [70], 8 years [84] and 10 years of age [67], with no effect of DHA supplementation on VIQ. One trial also conducted the NEPSY at the 10-year follow-up and likewise detected no benefit of the intervention [67]. (d) Two of the trials in preterm infants included an assessment of language-based academic abilities [67,70]. Both measured word reading and spelling in school-aged children and found no effect of the intervention [67,70]. Nor was there an effect on Pseudoword decoding [67]. (e) There were no assessments of language abilities not otherwise classified in the trials with preterm infant interventions. (f) Of the eight preterm infant RCTs, 4 reported conducting a subgroup analysis [24,67,69,70,84] and two studies reported sensitivity analyses [64,67]. Subgroup and sensitivity analyses were conducted for sex [67,69,70,84], birthweight [24,69,70,84], and household income [24]. One of the earlier RCTs found a sex by treatment interaction where girls in the DHA group had better academic abilities, although there was no such effect on language-based cognitive abilities [67]. Performance of boys in this trial did not differ between groups [67]. In a larger trial including both breastfed and formula-fed preterm infants, subgroup analyses revealed no sex by treatment interaction on language development at 26 months [69], or 7 years [70] of age. Likewise, birthweight <1250 g or >1250 g did not appear to interact with language abilities in this same trial [69,70]. Another trial testing for interaction effects between sex and birthweight ≤1000 g found none [84]. A separate trial identified a negative effect of DHA supplementation on language at 16-22 months of age within infants with birthweight <1250 g [24]. This trial also explored household income and found no interaction effect with DHA supplementation on language outcomes [24]. Sensitivity analyses involving only the preterm infants who did not receive any breastmilk revealed benefits to language-based cognitive abilities in the intervention group at 10 years of age, but no effect on academic abilities [67]. Sensitivity analyses in another trial detected a benefit of the intervention to parent-rated vocabulary comprehension (although no effect on overall language scores, or language production) at 14 (but not 9) months when multiple births and non-English speaking families were excluded [64].

Assessments of Language Abilities after Maternal Postnatal DHA Intervention
The three trials of postnatal maternal DHA supplementation primarily revealed null, but also some negative effects on seven comparisons of language development from 12 months to 5 years of age [32,53,62,63].
(a) One of the maternal postnatal trials conducted a clinician-assessment of language abilities [62], and one used a global development measure that included a language subscale [53]. No effect of the DHA intervention was detected with the CLAMS at 12 or 30 months of age [62]. Nor was an effect found on the Denver at 6-12 months, 12-18 months or at 18-24 months [53] of age. (b) One trial in breastfeeding mothers administered the parent-rated MCDI at 12 and 24 months of age and found no group differences [32]. (c) One postnatal RCT in mothers administered a WPPSI-R at 5 years of age and detected no effect of DHA supplementation [63]. (d) No trials of DHA supplementation in breastfeeding mothers assessed language-based academic abilities. (e) There were no assessments of language abilities not otherwise classified in the trials that supplemented breastfeeding mothers. (f) Only one of the three maternal postnatal RCTs reported conducting a subgroup analysis for sex [32]. Authors found poorer language scores within boys in the DHA group, when compared with the control group, but no effect within females [32].
(a) Of the eight trials conducted postnatally in infants, three included a clinician-administered assessment of global language abilities [29,31,73,89]. Null effects were reported for the CELF at 6 years [89] and the PPVT-R at 39 months [73] of age. However, in one trial with repeat language measures, DHA group children had a slightly worse PPVT-III score than control group children at 2 years, whilst there were no differences detected on the same assessment at 3.5 years [31] and at 5 years of age PPVT-4 scores were higher in the treatment group compared with the control group [29].
Five trials included overall language as a subscale of a global developmental test [29,30,74,86,87,90] and null effects were reported for the KPSDSI at 9 months [86], Bayley-III at 18 and 24 months [74,90], and Bayley-II at 18 months [87] of age, whilst another trial found that DHA group children had higher scores on the Bayley-II at 18 months of age [29,30].
(b) Of the eight RCTs in infants, only three included parent-rated measures of child language [29,71,72,74,89]. The MCDI was completed five times by parents at 9 months [72], 12 months [74], 14 months [71,72], and 18 months [29,74] of age with no hint of benefit of the DHA intervention at any age. The CCC was administered when children were 6 years of age [89] without any group differences. When the same children were 18 months of age, parents completed a language survey included in a behavior questionnaire [74] that likewise suggested no effect of DHA supplementation. (c) From the 8 infant intervention trials there were seven IQ or cognitive tests [29,60,61,88]. The WASI was conducted at 9 years of age along with the NESPY in one trial where authors were testing for effect interactions with smoking [60]. There were 3 trials that administered the WPPSI test [29,61,88]. Two conducted a WPPSI-R at 4 years [88] and 6 years [61] of age, with no evidence of an effect of the intervention. However, in a third trial the WPPSI-III at 6 years of age detected higher scores in children who received the DHA intervention compared with children who received no DHA [29]. (d) One trial of infant DHA supplementation included an assessment of language-based academic abilities and school readiness at 2.5 years of age and found no differences between the groups in either the overall score, or subscales [31]. (e) Two infant intervention studies included an alternate/experimental assessment of language, and neither found a difference in randomization group performance on the task [73,89].
(f) Of the eight infant RCTs, two reported conducting a subgroup analysis [31,60], for sex [31], and for maternal smoking during pregnancy [60]. There was no evidence for a sex by treatment interaction for clinician-assessed language abilities at 2 years of age [31]. Subgroup analyses for prenatal smoking revealed a benefit of the DHA intervention to VIQ among children whose mothers smoked in pregnancy, but no effect among non-smokers [60].

Discussion
This is the first systematic review of the effect of DHA supplementation in early life on child language abilities. After conducting an exhaustive search strategy, we identified 29 eligible studies. There was a total of 84 language outcomes compared between a DHA and a control group, with 78 comparisons revealing no effect of a DHA intervention. We conclude that the existing evidence does not conclusively support or refute the hypothesis that DHA supplementation in the first 1000 days of life improves children's language abilities. However, it is noteworthy that adequate DHA is likely to be important from conception through to 24 months of age, and that to date no trial has attempted to provide DHA across this entire period [2,7].
It has previously been suggested that there may be participant characteristics that may modify the effect of a DHA intervention, and that RCTs should be encouraged to explore this possibility [94]. In particular, there is growing interest in the interaction effect of sex and early nutrition on outcomes [96]. However, there were only 9 of the 29 RCTs that explored the effect of DHA supplementation on language within a specific population subgroup. Evidence has pointed to the potential for inter-individual, biologically based, differences, such as child sex to moderate DHA synthesis [97][98][99][100][101]. There were six RCTs that reported subgroup analyses for sex [25,31,32,67,69,70,77,78,84], with no interaction detected in three trials [31,69,70,84], while two trials reported a difference within girls only [25,67] and one within boys only [32]. Exploration of birthweight by treatment interaction was considered in three trials of preterm infants, with no interaction effect of DHA reported in two [69,70,84], and a negative interaction effect in the other [24]. Trials that explored whether maternal education [94], maternal smoking during pregnancy [60,94], or home environment [24,52] had mixed findings. However, only three trials were likely powered to allow a meaningful subgroup analysis [25,51,52,68,70,77,78] and importantly, with all of these trials, the effects seen are generally in subgroup comparisons only, and could be Type I errors given that these are not the primary outcome and there are numerous comparisons. Further work is needed to determine whether there are any true characteristics that interact with DHA treatment, or whether there are subgroups of individuals more likely than others to respond positively to DHA supplementation.
DHA interventions are hypothesized to be beneficial for brain development, and at present there is no known mechanism for a negative effect of DHA. However, an optimal dose of DHA is yet to be identified, and it may be that exposure to excessive DHA is detrimental. In the formula trial testing three doses of DHA, the high-dose group performance was similar to or worse than the control group, while the low and mid-dose groups performed well [29][30][31]. In a trial of breastfeeding Danish women (who would naturally already be providing their infants with some DHA), increasing DHA intake led to some instances of worse language, mainly within boys, and infant DHA status at the end of the intervention period was negatively associated with vocabulary [32]. Of the studies that identified a possible negative effect of DHA on language, the doses were 800 mg/day or higher [25,32,77,78], and supplementation was in addition to DHA present in the diet of the participant. Additionally, two studies, one with subgroup analyses for maternal education and one exploring household income (both markers of socio-economic status likely positively associated with dietary DHA intake), indicate possible adverse effects of DHA supplementation in those who have completed higher levels of education or have higher household income [24,94]. Future research is needed to establish whether there is an upper limit for DHA intake, above which additional DHA is detrimental and whether DHA supplements should not be recommended for those already consuming sufficient dietary DHA.
Potential adverse effects of DHA supplementation are scarcely mentioned in other reviews and meta-analyses [19][20][21][22][23][40][41][42]. However, several intervention trials report negative effects of DHA, generally in subgroup analyses [24,25,70,78]. A prenatal trial found that parents of children in the DHA group perceived more behavioral problems and executive dysfunction at four and seven years of age than parents of children in the control group [25,77,78]. Similarly, in a RCT of DHA in neonates born <33 weeks' gestation parents of females in the high-DHA group rated their children as having poorer behavioral functioning than parents of females in the standard-DHA group [70]. A RCT in toddlerhood for infants born preterm conducted subgroup analyses and found that among children from higher income households there was a possible negative effect of the DHA intervention on effortful control [24]. An infant study testing three formulas with varying doses of DHA to a control formula reported a consistent benefit at the two middle doses and a decline in performance on cognitive, language and executive functioning tasks in the group that received the highest dose of DHA, suggesting a dose-effect [29]. Importantly, with all of these trials, the effects seen are generally in secondary or exploratory outcomes in subgroup comparisons and could be Type I errors. There are currently no known mechanisms for DHA supplementation to have an adverse effect on brain development or functioning.
As a single nutrient intervention, DHA supplementation in nutritionally replete samples is likely to have a small-to-modest effect, requiring large samples in order to detect efficacy or adversity. This is particularly true if there are subpopulations that respond differently to increased DHA exposure. Furthermore, large samples increase the likelihood that characteristics conducive to optimal development, such as genetics and parental education, are balanced between the intervention and control groups, and hence will not confound group comparisons. Many of the trials included in this review had relatively small samples (n < 100 enrolled per group) [26,30,32,[54][55][56]65,75,76,81,83,85,87,90] and were powered to detect relatively large differences rather than the modest effects that might be expected from a single nutrient intervention. Few trials attempted to account for possible confounders such as environmental stimulation [25,[62][63][64]69,70,77,78], maternal intelligence quotient [63], paternal education [30,31,61,74] or maternal language [29], although all but one adjusted for maternal education [87,88]. Attrition was high (>20%) in many studies, and some attrition could be linked to post-randomization exclusions that could contribute to systematic loss to follow-up and attrition bias [93]. Given the already small and underpowered samples in many of the included trials, the chance of a Type I error may be increased.
Compounding the likelihood of a Type 1 error in the included studies is the fact that language was not a primary outcome of any trial, and all included trials compared multiple outcomes between the groups. A further important limitation is that few relevant DHA RCTs included a specific assessment of language abilities [24,[26][27][28][29]32,64,69,[71][72][73]75,76,82,83]. Several included trials only have a language score as part of an assessment of behavior [74,75], or general development [26,56,57,74], or from a cognitive assessment involving a measure of language-based cognitive abilities [52,55,58,[65][66][67]70,75,76,84]. Several RCTs assessed language development using the MCDI [28,29,32,64,69,[71][72][73][74]80], despite debate regarding its suitability for evaluating the effectiveness of interventions [102]. Two trials used the Bayley-I or II and attempted to calculate their own language score from the combined cognitive and language Mental Development Index [30,87]. Few trials provided scores for global language abilities with the majority using assessments that measured a specific domain, which for children under 24 months typically involves basic naming ability. Furthermore, there are numerous DHA trials that were not eligible for inclusion in this review as they had neither a language assessment nor assessment of language-based cognitive abilities, although they assessed an aspect of neurodevelopment such as cognition [51,. One trial was excluded as the form of supplementation was eggs (which contain DHA as well as other nutrients that contribute to neurodevelopment) and the control group received no intervention [126]. It is noteworthy that several ineligible RCTs used the Bayley-I or II (which combines cognition and language into a Mental Development Index) [29,51,57,59,62,[64][65][66]68,71,86,87,103,111,113,[118][119][120][121][127][128][129]. Hence, key aspects of language abilities may have been missed in many of RCTs assessing the effect of early DHA supplementation on child outcomes.
The effect of early DHA supplementation on language abilities as summarized here is comparable to previous reviews and meta-analyses of DHA interventions for child development outcomes [19-23, 40,41,43]. Only one other review to date has synthesized DHA trials conducted during pregnancy, as well as for preterm infants, and postnatally [42]. Authors of this study reported child outcomes for cognition, motor and visual development across 38 trials, DHA intervention enhanced infant cognition (but not child IQ) and visual acuity [42]. This differs to the findings of previous high-quality reviews and meta-analyses, that considered only one period of supplementation (for example, during pregnancy only) [19][20][21][22][23]40]. Unlike our own review, the authors of this review did not report or detect any subgroup effects (for world region, race, maternal education, age at assessment, intervention duration or trial quality) [42]. However, they did not explore the subgroups considered in our current review, with the exception of maternal education. Importantly, the review also missed several trials and eligible follow-ups [24,60,70,77,78,82,83,85,90], particularly some of the key larger ones with null findings [24,60,70,77,78].

Conclusions
The present study provides the first overview of the totality of evidence around the effect of early DHA supplementation on the development of language. There were 84 assessments of language reported between 29 trials of DHA supplementation in the first 1000 days, with only four findings of a positive effect on a language outcome (three from the same RCT). Substantial variation between the included trials, such as the timing, duration, and dosage of the intervention, the sample (preterm or full-term, or both, and high or middle-to low-income countries) as well as range of language outcomes and assessments, and absence of consistent subgroup comparisons make it difficult to reach a definitive conclusion regarding the efficacy of early DHA interventions to improve language. Whilst the vast majority of trials consistently reported a null effect, two studies detected potentially adverse effects, suggesting that it may be worth considering the possibility of a detrimental effect of DHA exposure on language, at high doses and/or in specific subpopulations. This review supports the need to consider that blanket DHA supplementation strategies may not be appropriate. Further work is needed to identify whether there is an upper-limit for safe DHA exposure, and which (if any) population subgroups may benefit or be adversely effected by DHA supplementation.