Study interventions and comparators

Studies of total breastfeeding duration, exclusive breastfeeding duration [14], and timing of solid food introduction were included in Review A ‘milk feeding’. Studies of any other nutritional exposure, ranging from dietary pattern to specific micronutrient supplementation and from single to multifaceted interventions, were included in Review C ‘maternal and infant diet’. For observational studies, data on dietary exposures were acquired from interviews, health records, diaries, and questionnaires. With the exception of blood vitamin D level, we did not include assessments of nutrient status as an exposure, and we did not include measures of dietary diversity as an exposure.

Study designs and populations

We included all intervention trials, and observational studies described as cohort, case control or cross-sectional analytic studies. Intervention trials were classified as randomised controlled trials (RCTs), where the method of treatment allocation was random, and as controlled clinical trials (CCTs), where treatment allocation was nonrandom and likely to lead to significant imbalance between treatment groups. CCTs were analysed separately from RCTs. Prospective and retrospective observational studies were analysed separately. We included studies of diet during pregnancy or lactation and of infant feeding between birth and 12 months of age. We excluded studies in which participants or their mothers were defined by the presence of a preexisting disease state, including very low birth weight or very premature infants.

Study outcomes

Allergic and autoimmune outcomes were selected on the basis of their population prevalence in children and young adults [15]. We included diseases with a prevalence of at least 1 in 1,000 in children/adolescents or young adults (aged <40 years) but did not include rarer diseases. We did not include pernicious anaemia or adult-onset rheumatoid arthritis despite a high prevalence in middle-aged and elderly people because their prevalence in young people is lower than 1 in 1,000, and prospective studies of infant feeding in relation to diseases of older adults were thought unlikely to have been reported. Allergic outcomes that met our inclusion criteria were asthma or wheeze, eczema, allergic rhinitis and/or conjunctivitis, food allergy (defined by food challenge, medical diagnosis, or self/parent report), allergic sensitisation, i.e., skin prick or specific immunoglobulin E (sIgE) assessment, and total immunoglobin E (IgE) level. We categorised asthma as wheeze, recurrent wheeze, or atopic wheeze depending on the definition used in the original publication, and we included measures of lung function, specifically bronchial hyper-reactivity, forced vital capacity, peak expiratory flow rate, and forced expiratory volume in 1 second. Physician-diagnosed asthma was included within the category ‘recurrent wheeze’, unless defined as a single episode of wheeze. Autoimmune diseases that met our inclusion criteria were type 1 diabetes mellitus (TIDM) (doctor diagnosed or serological diagnosis), coeliac disease defined serologically or clinically (positive Immunoglobulin A tissue Transglutaminase antibodies on one or more visits or a positive small bowel biopsy), inflammatory bowel disease, autoimmune thyroid disease, juvenile rheumatoid arthritis, vitiligo, and psoriasis, all as doctor diagnosis. For clinical allergic outcomes, we grouped age at assessment as 0–4 years, 5–14 years, and ≥15 years. Allergic sensitisation and autoimmune diseases were not stratified by age at outcome assessment.

Data sources

We searched The Cochrane Library (2013, Issue 7), Excerpta Medica dataBASE (EMBASE) (1947 to July 2013), Literatura Latino Americana em Ciências da Saúde (LILACS) (1982 to July 2013), Medical Literature Analysis and Retrieval System Online (MEDLINE) (1946 to July 2013), and Web of Science (1970 to July 2013) on 25 July 2013. We reran the searches for intervention trials and systematic reviews on 26 February 2017 and for intervention trials again on 15 December 2017. We included all studies published up to that date and studies in progress or completed but unpublished studies identified through http://apps.who.int/trialsearch/. Both peer-reviewed publications and abstract publications were included. We reviewed the bibliography of eligible studies for possible additional publications and included all eligible publications, regardless of the language. Study authors were not contacted for original data. The search strategies were extensively piloted and refined to optimise sensitivity, comparing search results with those of other systematic reviews. The search strategies are shown in S1 Text.

Study selection

Title and abstract screening was undertaken by a team of 9 trained researchers (RJB, VGL, DI, NG, KJ, JC, ZR, AR, PD). Two researchers undertook title screening independently and met to agree upon included and excluded titles. Their screening was checked by a third person, and uncertainties were resolved at a weekly team meeting between February and April 2014, 2015, and 2017 (RB, SC, VGL). The full texts of all potentially eligible studies were reviewed.

Risk of bias assessment

Risk of bias assessment was undertaken in duplicate, using modified versions of the Cochrane Collaboration Risk of Bias tool for intervention trials and the National Institute for Clinical Excellence methodological checklist for cohort and case–control studies [16]. Risk of bias domains for intervention trials were Selection bias (sequence generation and allocation concealment), Assessment bias (blinding of outcome assessors and validity of the outcome assessment tool), and Attrition bias (considered high when <70% of randomised participants had outcome data). Risk of bias domains for observational studies were Selection bias (low if cases and controls came from similar populations and participation rate ≥80%), Assessment bias (the validity of exposure and outcome assessment tools), and Confounding bias (whether study design and analysis accounted for most potential confounders). We separately included an assessment of Conflict of Interest for each study, judged as low where there was no evidence of industry involvement in study design, analysis, interpretation, or publication and no evidence that study authors received remuneration from relevant industry partners for other activities.

Data extraction

Data extraction was undertaken in duplicate. Disagreements and uncertainties about data coding and risk of bias were discussed within the team. For foreign language studies, data were extracted by VGL together with a native speaker of the relevant language (see Acknowledgemnts section). We extracted all relevant data from included studies, including data not able to be meta-analysed. Data were extracted using either raw frequencies or crude or adjusted effect estimates. Random effects meta-analysis was undertaken, and where this was not possible, study results were summarised in a narrative table.

Data selection for analysis

For intervention trials, we extracted outcome data that adhered to the intention-to-treat principle in preference to data based on per protocol analyses. Where studies included multiple intervention groups, we performed pairwise comparisons where we split the number of events and no events in the unexposed/control group to prevent double counting. Where studies reported data at multiple timepoints, we extracted the most complete dataset available beyond the intervention period (i.e., from 1 year of age onwards); this is the dataset with the largest denominator or, where the denominator is identical for multiple time points, the largest numerator (number of events). Where studies reported multiple assessments of the same outcome at the same timepoint, clinical assessments were selected in preference to serological assessments, and skin prick in preference to sIgE assessment of allergic sensitisation. For observational studies of breastfeeding or timing of solid or other food introduction to the infant diet, data were only included in meta-analysis where the reference group was complete, i.e., ‘less than’ a certain duration or, in some cases, ‘never’. Where more than one exposure group was compared with the reference group, we analysed data for the latest exposure group. For all observational studies, adjusted effect estimates were used in preference to unadjusted estimates for meta-analysis if both were reported.

Breastfeeding duration and timing of solid food introduction

One intervention trial (17,046 participants) and 259 observational studies (947,097 participants) reported the association between total breastfeeding, exclusive breastfeeding or timing of solid food introduction, and allergic or autoimmune diseases. The intervention trial found reduced eczema in the first year of life in centres allocated to deliver a breastfeeding promotion intervention. Meta-analyses of 24 observational studies showed an association between total breastfeeding duration and reduced recurrent wheeze at 5 to 14 years (e.g., Fig 2), but we also found evidence of publication bias. Meta-analyses of 31 observational studies showed an association between total and exclusive breastfeeding duration and reduced TIDM (e.g. Fig 3). Associations were statistically significant for retrospective studies but not for prospective studies. The GRADE certainty of these breastfeeding findings was low to very low. There was no association found for other allergic outcomes or for the other autoimmune diseases studied (coeliac disease, 12 studies; inflammatory bowel disease, 13 studies; juvenile rheumatoid arthritis, 3 studies; autoimmune thyroid disease, 1 study).

Open in a separate window

Fig 3

Observational study findings for a relationship between breastfeeding ever (A) or exclusive breastfeeding for ≥3–4 months (B) and type 1 diabetes mellitus. CI, confidence interval; OR, odds ratio; W, weight.

Intervention trials of other maternal or infant dietary exposures

Probiotic and prebiotic supplementation

Twenty-eight intervention trials (6,705 participants) evaluated probiotic supplements. These were either single or multiple organisms, given as capsules, powder, or part of a drink or infant formula milk, at a dose of 1 to 10 billion colony-forming units per day. Meta-analysis showed an association between probiotic supplementation and reduced eczema (Fig 4; 19 studies; RR 0.78; 95% CI 0.68–0.90; I² = 61%) or ‘atopic’ eczema (Fig 4; 11 studies; RR 0.78; 95% CI 0.65–0.92; I² = 0%) at age ≤ 4 years and reduced allergic sensitisation to cow’s milk at age 1 to 2 years (Fig 5; 97 studies). No significant association was seen with eczema at age 5 to 14 years, nor with allergic sensitisation to other allergens. The GRADE certainty of these findings was moderate for eczema and low for cow’s milk sensitisation. Subgroup analysis for eczema showed a significant difference between supplementing mothers during the postnatal period (9 interventions, RR 0.64; 95% CI 0.51–0.80; I² = 59%) and studies that just supplemented infants during the postnatal period (11 interventions, RR 0.93; 95% CI 0.81–1.06; I² = 31%; test for subgroup difference P = 0.016). Ten intervention trials (4,242 participants) evaluated prebiotic supplements (always a galacto-oligosaccharide) either alone or combined with other prebiotics or with probiotics. Nine studies involved comparison of an infant formula milk with versus without prebiotic. Meta-analysis showed no clear evidence that prebiotic supplementation reduces eczema at age ≤ 4 years (Fig 6; 7 studies; RR 0.75; 95% CI 0.56–1.01; I² = 57%) and no association at age 5 to 14 years. There was no association between probiotics or prebiotics and other allergic outcomes. In our updated search on 15 December 2017, we identified one publication reporting no effect of a perinatal probiotic supplement on risk of coeliac disease at age 13 years from the trial of Kukkonen [20]. No other studies reported autoimmune outcomes.

Open in a separate window

Fig 4

RCT findings for probiotic supplementation compared with no probiotics and risk of eczema (A) or atopic eczema (B) at age ≤4 years. CI, confidence interval; RCT, randomised controlled trial; RR, risk ratio; W, weight.

Open in a separate window

Fig 5

RCT findings for probiotic supplementation compared with no probiotics and risk of allergic sensitisation to any allergen (A), any inhalant allergen (B), any food allergen (C), egg (D), milk (E), or peanut (F). CI, confidence interval; RCT, randomised controlled trial; RR, risk ratio; W, weight.

Open in a separate window

Fig 6

RCT findings for prebiotic supplementation compared with no prebiotics and risk of eczema at age ≤4 years.

CI, confidence interval; RCT, randomised controlled trial; RR, risk ratio; W, weight.

Polyunsaturated fatty acid supplementation

Nineteen intervention trials (14,479 participants) evaluated fatty acid supplements. These were given as fish oil supplements in 12 trials, advice to eat oily fish in 1 trial, supplementation of infant formula milk with omega-3 fatty acids in 3 trials, and other oils in 3 trials. Comparators were corn, soya, canola, olive, sunflower, or vegetable oil, or no supplement. Meta-analysis showed an association between omega-3 polyunsaturated fatty acid supplementation using fish oil during pregnancy and lactation and reduced allergic sensitisation to egg at 1 year (Fig 7; 6 studies; RR 0.69 95% CI 0.53–0.90; I² = 15%). No significant association was seen with allergic sensitisation to other allergens. The GRADE certainty of this finding was moderate. Subgroup analysis showed a significant difference between studies which supplemented mothers during pregnancy (4 studies; RR 0.55 95% CI 0.40–0.76; I² = 0%) and studies which supplemented during lactation only (2 studies; RR 0.92 95% CI 0.65–1.28; I² = 0%; test for subgroup difference P = 0.032). Similarly, for allergic sensitisation to peanut, there was reduced risk in the subgroup of studies which supplemented mothers during pregnancy (2 studies; RR 0.62 95% CI 0.40–0.96 I² = 0%). There was no association between omega-3 polyunsaturated fatty acid supplementation and other allergic outcomes. Data were sparse and inconclusive for food allergy and for omega-6 supplementation and allergic outcomes. No autoimmune outcomes were reported, and no new trials were identified in our update search on 15 December 2017.

Open in a separate window

Fig 7

RCT findings for omega-3 polyunsaturated fatty acid supplementation compared with no polyunsaturated fatty acids and risk of allergic sensitisation to any allergen (A), any inhalant allergen (B), any food allergen (C), milk (D), egg (E), or peanut (F). CI, confidence interval; RCT, randomised controlled trial; RR, risk ratio; W, weight.

Maternal allergenic food avoidance and multifaceted interventions

Twelve intervention trials (1,945 participants) evaluated maternal allergenic food avoidance during pregnancy (2 trials), lactation (6 trials), or both (4 trials). The foods excluded were milk alone (2 trials), egg alone (1 trial), milk and egg (3 trials), or multiple foods (6 trials). We did not find that maternal allergenic food avoidance during pregnancy or lactation reduces risk of allergic disease or TIDM development. Nine trials (2,317 participants) evaluated multifaceted interventions—defined as an intervention with more than one category of dietary intervention. Six trials included promotion of prolonged breastfeeding, avoidance of allergenic foods for mother or infant, and delayed solid food introduction; 6 included environmental control measures to reduce dust mite, pet, or tobacco smoke exposure; 7 used an alternative formula milk as part of the intervention. Meta-analysis shown in Fig 8 found an association between multifaceted interventions and reduced allergic rhinitis at age ≤ 4 years (2 studies; RR 0.61 95% CI 0.44–0.84; I² = 0%), but not at age 5 to 14 years (4 studies). We also found an association between multifaceted interventions and reduced wheeze (2 studies; RR 0.57 95% CI 0.41–0.81; I² = 0%) and recurrent wheeze (4 studies; RR 0.77 95% CI 0.61–0.97; I² = 0%) at age 5–14 years. There was no association at age ≤ 4 years (4 studies), nor with measures of lung function. The GRADE certainty of findings for multifaceted interventions and wheeze or rhinitis was low, there were no autoimmune outcomes reported, and no new trials were identified in our update search on 15 December 2017.

Open in a separate window

Fig 8

Randomised controlled trial findings for multifaceted dietary interventions compared with no multifaceted intervention and risk of allergic rhinitis at age ≤4 years (A) or 5–14 years (B), wheeze (C) or recurrent wheeze (D) at age 5–14 years, and wheeze (E) or recurrent wheeze (F) at age ≤4 years. CI, confidence interval; RCT, randomised controlled trial; RR, risk ratio; W, weight.

Observational studies of other maternal or infant dietary exposures

Ninety-two observational studies (485,733 participants) evaluated other maternal or infant dietary exposures, including fruit and vegetable intake (35 studies), classified as fruits, vegetables, citrus fruits, nuts, or sometimes as individual foods or food groups; fat and fatty acid intake (26 studies), classified as total fat intake, omega 3 or omega 6 poly-unsaturated fatty acids or other types of fatty acids, and specific fatty acids (we also included fish intake and fish oil supplementation in this group); vitamin and mineral intake (50 studies), including antioxidant vitamins A, C, E, and their precursors, vitamin D and folic acid, mineral intake, and blood vitamin D levels; and a wide range of ‘other dietary exposures’ (47 studies), including dietary pattern, maternal allergenic food avoidance, alcohol, tea or coffee intake, and maternal or infant meat, cereal, milk, or egg intake. Opportunities for meta-analysis of these observational studies were very limited due to heterogeneity of exposure assessment methods and reporting—all analyses are described in the full Food Standards Agency report (https://www.food.gov.uk/science/research/allergy-research/fs305005ac). Overall, there was no consistent association between these dietary exposures and risk of allergic outcomes or those autoimmune diseases studied (TIDM, 23 studies; coeliac disease, 2 studies; inflammatory bowel disease, 2 studies; juvenile idiopathic arthritis, 1 study). There was insufficient evidence to either support or refute the intervention trial findings for fish oil supplementation and allergic sensitization to foods, in the observational studies of fatty acid intake during pregnancy.

Interpretation

Our key findings suggest that a mother’s diet during pregnancy and lactation may influence the risk that her child develops allergic disease, and this has implications for future research in this field. There are two specific implications for pregnant women. First, a daily probiotic supplement such as L. rhamnosus, taken from around 36 to 38 weeks gestation through the first 3 to 6 months of lactation, may reduce risk of eczema in the child. Although probiotics are generally considered safe, their pro-inflammatory effects may have potential adverse consequences when used earlier in pregnancy, and serious adverse effects in people with intestinal disorders or immune deficiency have been documented [30][31]. Second, a fish oil supplement, taken from around 20 weeks gestation through the first 3 to 4 months of lactation, may reduce risk of allergic sensitisation to egg or peanut in the child. In one included trial, the fish oil intervention was administered via advice to consume at least 2 servings per week of low mercury oily fish (farmed salmon), which is consistent with current United States Food and Drug Administration guidance [32]. It was not possible to reliably assess whether the findings in this small trial differed significantly from the findings in higher dose fish oil supplementation trials.

Strengths and limitations of the study

This comprehensive and wide-ranging systematic review has highlighted key dietary interventions that may reduce risk of allergic outcomes, information which will be useful for pregnant women and their healthcare providers. However, for many dietary exposures, data were inconclusive or inconsistent, such that we are unable to exclude the possibility of beneficial or harmful effects. One limitation of this study is the 2013 search date for observational studies, such that review findings that are derived from these studies may benefit from an update in due course. We included abstract publications, since their exclusion can lead to publication bias; however, their inclusion carries a risk of methodological bias due to the lack of peer review. While the list of dietary exposures included was comprehensive, we did not evaluate food diversity during infancy, which has been suggested as an important factor in relation to allergic outcomes, and we did not evaluate objective markers of nutritional status, with the exception of blood vitamin D level [33]. This means that we may have missed important relationships for food diversity or nutritional markers. Given the volume of titles included in this systematic review, we were unable to contact the corresponding authors of included studies for data verification, as has been recommended by some authors [34]. Finally, some outcome measures were reported using a wide variety of assessment tools—for example, wheeze and recurrent wheeze were defined in several different ways—and this may have contributed to some of the statistical heterogeneity seen in meta-analyses. There is a need for more well-powered intervention trials of dietary interventions, given the positive findings with some specific dietary supplements identified here.

We are grateful to the independent experts, Professor Graham Devereux and Dr Carina Venter, and to members of the UK Food Standards Agency, the Committee on Toxicity of Chemicals in Food, Consumer Products and the Environment (COT), and the UK Scientific Advisory Committee on Nutrition (SACN) for their comments on the study protocol and versions of the full report. We are grateful to Dr Yujie Zhao, Mr Szymon Mikolajewski, Dr Andre F Amaral, Dr Mari Kihara, Dr Christian Nielsen, Mr Radoslav Latinovic, Dr Stephanie MacNeill, Dr Andreas Forsters, Dr Daniel Munblit, Dr Sze-Chin Tan, and Dr Claudia Gore for their assistance with translation of foreign language reports. We are grateful to Tim Reeves for writing the search strategy and undertaking the initial literature search and to Jacqueline Cousins of the St. Mary’s library at Imperial College London for assistance with literature search training. We are grateful to Ulugbek Nurmatov and Nara Tagiyeva-Milne for assistance with data extraction, to Evangelia Andreou for assistance with creating tables and figures for the full report, and to Thomas Clark for assistance with manuscript submission.