Our data provide evidence against the preventing role of prolonged exclusive (but not partial) breastfeeding in AD occurrence and confirm recent results indicating a beneficial role of early weaning in AD.

Early weaning, defined as the introduction of solid foods at 4 or 5 months of age, was inversely related to the risk of AD, with children weaned at 4 months having lower AD risk (OR = 0.41, 95% CI, 0.20–0.87) compared to those exclusively breastfed. Similar results were observed for weaning started at 5 months of age (OR = 0.39, 95% CI, 0.18–0.83). This association persisted when children with and without family history of allergy were considered separately. Prolonged partial breastfeeding (breastmilk plus milk formulas) was not associated with AD. Consistently, the introduction of a high number of different solid foods reduced the risk of AD ( P trend = 0.02 at 4 months of age and P trend = 0.04 at 5 months).

We conducted a matched case–control study on incident physician‐diagnosed AD in early childhood including 451 cases and 451 controls. Data on several factors, including feeding practices, were collected through an interviewer‐administered questionnaire. Odds ratios (OR) and the corresponding 95% confidence intervals (CIs) were estimated through logistic regression models, conditioned on study center, age, sex, and period of interview, and adjusted for potential confounders.

Epidemiological data on infant feeding practices and allergic diseases are controversial. The purpose of this study was to explore the association of early weaning with the occurrence of atopic dermatitis (AD).

Atopic dermatitis (AD) is a major public health problem, given its high and increasing prevalence in higher‐income countries 1, the substantial impact on the quality of life of both patients and their families 2, and the associated economic burden 3. Moreover, increasing evidence from cross‐sectional and longitudinal studies and experimental data suggests that AD may progress to other allergic phenotypes (‘atopic march’) 4. Thus, any intervention preventing AD would have appreciable public health relevance. Breastfeeding has been considered an important strategy to reduce atopic diseases for many years, the first epidemiological evidence for a protective effect appearing in the mid‐1930s 5. Current prophylactic feeding guidelines to aid allergy prevention recommend exclusive breastfeeding for 4–6 months of life and the introduction of solid food thereafter 6-9. These guidelines were based on results from earlier observational studies, which reported a protective effect of prolonged exclusive breastfeeding on AD and increased risks among infants with early introduction of solid food 10-13, and derived from the theoretical concern that presenting environmental stimulations through the introduction of solid food to an immature mucosal immune system may result in IgE‐mediated sensitization to allergens 14, 15. However, recent studies did not support a favorable role of prolonged breastfeeding in AD 16-21, and some of them suggested that delayed introduction of solid food may even increase the risk of AD 22-24. Most of these recent studies had larger sample size and a more careful control for confounding than previous ones. Another possible explanation of the conflicting findings between earlier and recent studies is related to the decrease in the overall exposure to microbial stimulation in high‐income countries over the last decades, which followed the improvements in household amenities and higher standards of personal hygiene. According to the ‘hygiene hypothesis’, this may increase the risk of atopic diseases and swamp any protective effect of breastfeeding 25. The aim of the present multicenter investigation was to explore the association of different patterns of feeding practice with the occurrence of AD, using data from a case–control study of incident AD in early childhood in Italy, a country from where scanty data are available on this atopic disease 26.

Methods Setting and study population We conducted a case–control study on incident AD between March 2011 and April 2014. Cases and controls were recruited in 10 Italian hospital centers in the northern and central Italy (Table 1). Table 1. Distribution of 451 incident cases of atopic dermatitis and 451 matched controls according to study design and children major characteristics. Italy, 2011–2014 Total N = 902 N (%) Cases N = 451 N (%) Controls N = 451 N (%) Study design characteristics Study center Milano 326 (36.1) 163 (36.1) 163 (36.1) Bergamo 304 (33.7) 152 (33.7) 152 (33.7) Bologna 124 (13.7) 62 (13.7) 62 (13.7) Othera 148 (16.4) 74 (16.4) 74 (16.4) Distance in months between first symptoms and interview 0 63 (14.0) NA 1 93 (20.6) 2 114 (25.3) 3 102 (22.6) 4–5 79 (14.0) Child's characteristics Sex Male 604 (67.0) 302 (67.0) 302 (67.0) Female 298 (33.0) 149 (33.0) 149 (33.0) Age (months) 3 108 (12.0) 51 (11.3) 57 (12.6) 4 128 (14.2) 67 (14.9) 61 (13.5) 5 129 (14.3) 64 (14.2) 65 (14.4) 6 88 (9.8) 44 (9.8) 44 (9.8) 7 66 (7.3) 36 (8.0) 30 (6.7) 8 63 (7.0) 30 (6.7) 33 (7.3) 9 45 (5.0) 21 (4.7) 24 (5.3) 10–11 71 (7.9) 37 (8.2) 34 (7.5) 12–14 73 (8.1) 36 (8.0) 37 (8.2) 15–17 46 (5.1) 24 (5.3) 22 (4.9) 18–19 27 (3.0) 13 (2.9) 14 (3.1) 20–24 58 (6.4) 28 (6.2) 30 (6.7) SCORAD (classification of severity)b Mild (<25) – (Median = 20.1) 31 (6.9) NA Moderate (25–50) – (Median = 39.7) 198 (44.0) Severe (>50) – (Median = 61.7) 221 (49.1) Controls diagnosis Check‐up growth NA 236 (52.3) Dermatologic diagnosisc 154 (34.1) Eye check‐upd 24 (5.3) Vaccinations 17 (3.8) Pelvic echography 1 (0.2) Minor congenital malformations 9 (2.0) Trauma/ingestion of foreign bodies/others 10 (2.2) Cases Of the 504 cases originally included, 10 did not meet inclusion criteria, and 43 were excluded from analysis for matching reasons (absence of controls with same center, period of enrollment, sex, and age). Thus, cases considered were 451 outpatient children aged 3–24 months (302 boys and 149 girls; median age 5 months) for whom a first diagnosis of AD was established by a trained dermatologist during the inclusion visit. In particular, the dermatologist asked parents about children AD symptoms and performed a physical examination. The same diagnostic criteria, previously defined during a preliminary consensus study carried out on 100 AD cases not considered in the present study, were adopted by all the dermatologists. The diagnostic criteria of incident AD were based on the following eight mandatory conditions: ‘no previous diagnosis of AD’, ‘first symptoms occurring no longer than 5 months before AD diagnosis’, ‘symptoms occurring also in the last 4 weeks’, ‘the child is suffering from itching’, ‘the child has eczematous lesions’, ‘age‐specific affected areas’, ‘flexural involvement’, and ‘the groin and armpits areas are not affected, unless a diagnosis of inverted psoriasis’. In 406 of the 494 cases, all the criteria were met. The dermatologists confirmed a diagnosis of incident AD for the remaining 88 children, for whom at most 2 criteria were not met at the time of the inclusion visit but their occurrence in the last 4 weeks was reported by parents (see Table S1). During the inclusion visit, the dermatologists accurately recorded the date of the occurrence of the first symptoms and assessed the severity of AD through the SCORAD index 27, 28. Only children whose parents reported that symptoms had started no longer than 5 months before were included in the study (incident AD). Controls Of the 505 controls originally included, seven did not meet inclusion criteria, and 47 were excluded from analysis for matching reasons. Thus, controls considered were 451 outpatients attending for a pediatric/dermatological visit, with no history of AD. They were matched 1 : 1 with cases by centers, sex, age, and period of interview. Two hundred and thirty‐six controls (52%) were recruited during a check‐up growth visit, 154 (34%) had a nonatopic dermatologic diagnosis (i.e. angioma (37%), hemangioma (25%), nevi (21%)), and the remaining 17% had other dermatologic conditions presented by no more than two children each. The remaining 14% (61 children) attended the hospital for other care needs (e.g. eye check‐up, mandatory vaccinations, minor congenital malformations, traumas). Less than 3% of both cases and controls approached refused to participate. Ethical issues All study centers obtained local ethics committee approval. Written consent was obtained from parents for the child, based on the recommendations of the ethical committees of the study hospitals. Data collection Trained personnel administered face‐to‐face to children parents a structured questionnaire, including information on a variety of factors of interest, for example, family socioeconomic context, maternal pregnancy exposures, birth factors (e.g. weeks of gestation, type of delivery, birth weight, and breastfeeding), factors favoring contact with infectious agents (e.g. the number of siblings, presence of pets, daycare, and contact with other children), child history of any manifestation of atopy, childhood diseases, dietary habits of the child from birth, and history of atopic conditions in first‐degree relatives. For more details, see ‘Data S1 in the Supplementary material’. A reproducibility study including 190 patients was also conducted. Parents were re‐contacted by phone by a trained interviewer, between December 2011 and November 2012, in order to quantify differences among answers gave to the same question during the two successive interviews for 100 items. The proportion of observed agreement for answers related to ever breastfeeding and month of stopping breastfeeding was evaluated by Cohen's kappa statistic (K). For more details, see ‘Data S2 in the Supplementary material’. Data on feeding practices A specific section of the questionnaire collected childhood feeding data from birth. The questionnaire enquired about the pattern of breastfeeding (current breastfeeding and time when breastfeeding had stopped) and the child age (in weeks) when infant formulas (e.g. partially and extensively hydrolyzed formulas, plant‐based milk, humanized milk) and cow milk were introduced. Parents were also asked to indicate the age at the introduction of 22 foods or food groups representing the most common food in the diet of Italian infants in early childhood. These foods/food groups were grouped as vegetables, legumes, or roots (potatoes, carrots, tomatoes, and beans), fruit (apples/pears, peaches/apricots/plums, citrus fruits, and red fruits), cereals (maize/tapioca, rice, pasta, and gluten‐free pasta), meat (poultry, pork, and beef), dairy products (cheese, and other dairies), fish, eggs, and nuts/cacao/chocolate. Supplementations with vitamin D, K, fluorine, and lactic ferments and the child age (in weeks) when these supplementations were introduced and stopped were also collected. The availability of a detailed schedule of infant feeding practices, as well as of the date of the occurrence of the first AD symptoms, allowed us to exclude exposures occurring after first AD symptoms. Dietary exposure occurring after AD onset was not considered in the analysis, which was focused on the influence of diet on AD occurrence rather than progression. We evaluated breastfeeding pattern at 1, 2, 3, 4 and 5 months of age. Breastfeeding pattern was categorized as ‘no breastfeeding’ (i.e. children received only milk formulas and/or solid food), ‘exclusive breastfeeding’ if she had given only breastmilk, while milk formulas and solid food had not been introduced, ‘introduction of other milks’ if the mother had given breastmilk and milk formulas but solid food had not been introduced, and ‘starting weaning’ if solid food had been introduced. We also assessed the total number of food items included in the child's diet at 4 and 5 months of age, as well as specific food groups’ introduction. Statistical analysis Data were presented as frequency/percent distribution in the overall study sample and separately for cases and controls. Because we were interested in exposures influencing AD incidence, and AD diagnostic criteria include symptoms recurrence, we adopted a conditional approach in order to consider the same time window of exposure from birth for a case and his matched control. As an example, for a case of 6 months of age at inclusion with AD first symptoms occurring at the age of 4 months (i.e. 2 months before), we excluded from the analyses those exposures (e.g. the number of solid foods introduced), occurring between 4 and 6 months of age. Thus, for the matched control, we excluded exposures occurring in the same period (i.e. the number of solid foods introduced between 4 and 6 months). In this way, we ensured that a control has the same time window to experience an event as his matched case (Fig. 1). Figure 1 Open in figure viewerPowerPoint Diagram of the exposures included and excluded from the analysis. For cases, we included only exposures that occurred before first AD symptoms. For each control, the same ‘time window’ of observation of his matched case was considered. Odds ratios (ORs) and the corresponding 95% confidence intervals (CIs) of AD for infant feeding practices were computed using conditional logistic regression models, matching 1 : 1 cases and controls according to center, sex, age, and period of enrollment. Multivariable models including terms for history of allergy in mother and/or father, ethnicity, maternal age, maternal education, number of siblings, weeks of gestation, type of delivery, vitamin D and vitamin K supplementations were performed in order to adjust for potential confounding effects. Conditional models were also applied to perform analises on two subgroups of interest (i.e. controls with dermatologic diagnosis and controls with other diagnosis). Moreover, stratified analyses considering the type of milk and presence of atopy in parents were performed applying unmatched logistic regression models adjusted by center, sex, age, and period of enrollment (i.e. matching variables). The unmatched logistic model was used, instead of the matched one, in order to avoid the elimination of those matched pairs (case–control) presenting ‘discordant values’ for the factor of stratification. Q test was used to assess heterogeneity between strata (considering two strata heterogeneous when P was lower than 0.1). Statistical analyses were performed using SAS v. 9.4 (SAS Institute, Cary, NC, USA). We followed the STROBE guidelines on reporting of epidemiological studies throughout 29.

Results The distribution of 451 incident cases of AD and 451 matched controls according to the study design and selected children characteristics is presented in Table 1. Six hundred and four enrolled children were males (67%). About 50% of children were 3–6 months of age (108 were three months old, 128 four months old, 129 five months old, and 88 six months old), and ~22–23% were older than 12 months of age. SCORAD in cases ranged from 5.2 to 102.8 (mean = 49.4, SD = 17.1). After the exclusion of children with missing values in the variables of interest, and of 59 children (22 cases and 37 controls) not breastfed since birth, 329 case–control pairs (73%) were available for subsequent analyses. Compared to exclusive breastfeeding, the ORs for the introduction of other milks (i.e. partial breastfeeding) were, respectively, 0.70 (95% CI, 0.43–1.14) at 1 month of age, 0.91 (95% CI, 0.54–1.52), 0.73 (95% CI, 0.42–1.27), 0.77 (95% CI, 0.42–1.42), and 0.58 (95% CI, 0.25–1.33) at 2, 3, 4, and 5 months of age, respectively (Table 2). Although not significant, this inverse association observed for partial breastfeeding compared to exclusive breastfeeding persisted even when children fed with intact protein cow milk were considered separately (Table 3). OR estimates for children fed with hypoallergenic milks during the first 5 months of life were not reliable, because they were related to less than five exposed cases and five exposed controls. No major heterogeneity was also found when partial breastfeeding was considered in strata of matched case–control pairs with dermatologic or nondermatologic controls (i.e. 111 and 218 matched pairs, respectively) and history of allergy in mother and/or father (211 (64%) cases and 177 (54%) controls) (Table 3). Table 2. Odds ratios (OR) and 95% confidence intervals (CIs) of incident atopic dermatitis according to feeding practices before atopic dermatitis (AD) occurrence. Italy, 2011–2014 Total N = 774 N (%) Cases N = 387 N (%) Controls N = 387 N (%) ORa (95% CI) ORb (95% CI) Breastfeeding Ever 715 (92.4) 365 (94.3) 350 (90.4) 1.00c 1.00c Never 59 (7.6) 22 (5.7) 37 (9.6) 0.58 (0.34–1.00) 0.68 (0.38–1.24) Breastfed children Total N = 658 N (%) Cases N = 329 N (%) Controls N = 329 N (%) ORa (95% CI) ORb (95% CI) Feeding pattern at 1 month Exclusive breastfeeding 435 (78.8) 229 (83.0) 206 (74.6) 1.00c 1.00c Introduction of other milks 117 (21.2) 47 (17.0) 70 (25.4) 0.57 (0.36–0.89) 0.70 (0.43–1.14) Matched pairs for which cases had AD symptoms occurred at 1 months or before 106 53 53 Feeding pattern at 2 months Exclusive breastfeeding 326 (71.5) 171 (75.0) 155 (68.0) 1.00c 1.00c Introduction of other milks 130 (19.4) 57 (25.0) 73 (32.0) 0.68 (0.44–1.05) 0.91 (0.54–1.52) Matched pairs for which cases had AD symptoms occurred at 2 months or before. 202 101 101 Feeding pattern at 3 months Exclusive breastfeeding 242 (63.0) 132 (68.8) 110 (57.3) 1.00c 1.00c Introduction of other milks 142 (37.0) 60 (31.2) 82 (42.7) 0.56 (0.35–0.89) 0.73 (0.42–1.27) Matched pairs for which cases had AD symptoms occurred at 3 months or before 274 137 137 Feeding pattern at 4 months Exclusive breastfeeding 141 (45.2) 80 (51.3) 61 (39.1) 1.00c 1.00c Introduction of other milks 107 (11.5) 51 (32.7) 56 (35.9) 0.66 (0.35–1.15) 0.77 (0.42–1.42) Starting weaning 64 (20.5) 25 (16.0) 39 (25.0) 0.42 (0.21–0.84) 0.41 (0.20–0.87) Matched pairs for which cases had AD symptoms occurring at 4 months or before 346 173 173 Feeding pattern at 5 months Exclusive breastfeeding 72 (28.1) 44 (34.4) 28 (21.9) 1.00c 1.00c Introduction of other milks 52 (20.3) 25 (19.5) 27 (21.1) 0.57 (0.27–1.19) 0.58 (0.25–1.33) Weaning 132 (51.6) 59 (46.1) 73 (57.0) 0.43 (0.22–0.86) 0.39 (0.18–0.83) Matched pairs for which cases had AD symptoms occurring at 5 months or before 402 201 201 Solids diversity at 4 monthsd No solid food 248 (79.5) 131 (84.0) 117 (75.0) 1.00c 1.00c 1–2 foods 31 (9.9) 15 (9.6) 16 (10.3) 0.79 (0.34–1.83) 0.71 (0.29–1.74) 3–22 foods 33 (10.6) 10 (6.4) 23 (14.7) 0.31 (0.12–0.77) 0.30 (0.11–0.81) chi‐trend (P‐value) 5.1 (0.019) 5.7 (0.017) Matched pairs for which cases had AD symptoms occurring at 4 months or before 348 174 174 Solids diversity at 5 monthsd No solid food 124 (48.4) 69 (53.9) 55 (43.0) 1.00c 1.00c 1–7 foods 75 (29.3) 37 (28.9) 38 (29.7) 0.69 (0.34–1.41) 0.59 (0.27–1.26) 8–22 foods 57 (22.3) 22 (17.2) 35 (27.3) 0.48 (0.25–0.94) 0.44 (0.21–0.91) chi‐trend (P‐value) 4.6 (0.032) 5.1 (0.025) Matched pairs for which cases had AD symptoms occurring at 5 months or before. 404 202 202 Table 3. Odds ratios (ORs) of atopic dermatitis according to feeding patterns before first atopic dermatitis (AD) symptoms at 1, 2, 3, 4, and 5 months of age in strata of selected subgroups, in the 658 matched pairs of breastfed children. Italy, 2011–2014 Feeding pattern before first AD symptoms in cases at 1, 2, 3, 4, and 5 months of age (compared to exclusive breastfeeding) Introduction of other milks Start weaning OR (95% CI) OR (95% CI) 1 month 2 months 3 months 4 months 5 months 4 months 5 months Dermatologic controlsa 0.52 (0.25–1.09) 0.68 (0.35–1.31) 0.65 (0.34–1.25) 0.55 (0.26–1.18) 0.55 (0.22–1.34) 0.31 (0.10–0.96) 0.40 (0.16–1.02) Nondermatologic controlsa 0.59 (0.34–1.05) 0.68 (0.38–1.21) 0.48 (0.25–0.93) 0.81 (0.34–1.80) 0.61 (0.16–2.31) 0.52 (0.22–1.24) 0.47 (0.17–1.31) Type of other milksb Intact protein cow's milk 0.62 (0.40–0.96) 0.64 (0.41–1.00) 0.57 (0.37–0.89) 0.64 (0.38–1.09) 0.61 (0.28–1.30) 0.48 (0.22–1.02) 0.46 (0.23–0.91) Hypoallergenic milksc 0.94 (0.23–3.96)d 3.69 (0.75–18.20)d 1.50 (0.41–5.47)d 2.30 (0.44–11.93)d 0.58 (0.09–3.64)d 0.89 (0.16–5.14)d 13.31 (1.38–128.35)d History of allergy in parentsb No 0.45 (0.23–0.91) 0.54 (0.27–1.06) 0.37 (0.18–0.76) 0.53 (0.23–1.22) 0.76 (0.23–2.49) 0.45 (0.14–1.44) 0.88 (0.30–2.41) Yes 0.85 (0.45–1.45) 0.93 (0.53–1.63) 0.89 (0.51–1.56) 0.85 (0.43–1.69) 0.63 (0.23–1.73) 0.40 (0.17–0.93) 0.35 (0.15–0.82) Early weaning, defined as the introduction of solid foods, at 4 and/or 5 months of age, was inversely related to the risk of AD, with children weaned at 4 months having significantly lower risk of AD (OR = 0.41, 95% CI, 0.20–0.87) compared to those exclusively breastfed ones. Similar results were observed in children for whom weaning started at 5 months of age (OR = 0.39, 95% CI, 0.18–0.83). These associations were consistent with results obtained when food diversity at 4 or 5 months of age was considered (Table 2). Compared to no solid food intake, the ORs for the introduction of a high number of different solid food at 4 and 5 months of age were, respectively, 0.30 (95% CI, 0.11–0.81) and 0.44 (95% CI, 0.21–0.91). Those results were also confirmed when age of introduction of single food groups, in weaned children, was investigated (Fig. 2). In particular, positive time–risk associations were found for age at the introduction of vegetables, legumes, or roots (P trend = 0.06) and fruit (P trend = 0.02). No major heterogeneity was found when weaning patterns at 4 and 5 months of age were investigated in strata of control diagnoses, type of milk added as a supplement to breastmilk, and history of allergy in mother and/or father (Table 3). All children were weaned at 6 months of age. Figure 2 Open in figure viewerPowerPoint Odds ratios (a) and 95% confidence intervals of atopic dermatitis according to the age of introduction of selected food groups (considering the period before first atopic dermatitis (AD) symptoms only) in the 207 matched pairs of weaned children. Italy, 2011–2014. (a) Matched by center, sex, age, and period of enrollment and adjusted for history of allergy in mother and/or father, maternal ethnicity, maternal age, maternal education, the number of older siblings, and contact with pets. Due to the small numbers in the analyses on eggs and nuts/cacao/chocolate, models were matched for age, sex, center, and period of enrollment and adjusted only for history of allergy in mother and/or father. Cutoffs for compared categories were defined according to the control distribution (i.e. approximate tertiles or median). (b) The age at the introduction of cow's milk was not considered as age at the introduction of a dairy product for 8 children (5 cases and 3 controls) for whom cow's milk was introduced before weaning.

Discussion The present findings from a large case–control study on incident AD show that early introduction of solid food decreases the risk of AD occurrence. Early studies supported a protective role of breastfeeding in childhood AD 12, 13. A systematic review of prospective studies published until 2000 showed a protective effect of exclusive breastfeeding for at least 3 months on the development of AD 12, particularly in children with atopic heredity. These results were contradicted by a meta‐analysis updated to 2008 (21 studies) that concluded for a lack of a clear protection from AD by exclusive breastfeeding for at least 3 months (OR = 0.89, 95% CI, 0.76–1.04), even when stratifying for parental atopy 18. A review in 2012 (results from 24 studies) concluded that data did not support a beneficial role of breastfeeding in the development of AD 20, in line with a Cochrane review published in the same year 30. More recently, results from a birth cohort (more than 18 000 newborns and 1000 AD cases) reported that longer duration of breastfeeding increased the risk of AD up to the age of 18 months, with no significant effect with the time of introduction of solid food 31. Findings from the International Study of Asthma and Allergy in Childhood (ISAAC) Phase Two (over 51 000 randomly selected 8‐ to 12‐year‐old schoolchildren in 21 countries) found no evidence that exclusive breastfeeding protected against eczema 16. Duration of partial and exclusive breastfeeding did not influence AD risk by the age of 5 years in 3781 consecutively born children in Finland 17. In a nationally population‐based birth cohort study in Japan (more than 38 000 infants and 1402 young children with persistent AD and 8787 with episodic AD), exclusively breastfed infants during the first 6 months of life were at an increased risk of AD when compared to infants fed formula alone 23. Concerning the timing of introduction of solid foods, in the 1980s and 1990s, several studies reported a positive association between early introduction foods and AD 32, 33. However, as for breastfeeding, other studies did not reproduce these findings 17, 31, 34-37. Some of them even suggested, in line with our results, an inverse association between early introduction of solid food and the onset of AD 21, 24, 34, 38-40. The strengths of our study are as follows: (i) the inclusion of matched cases and controls, (ii) the inclusion only of children with a first diagnosis of AD (incident cases), and (iii) the very high rate of response to the questionnaire. The study also met 10 of the 12 biologic and methodological quality standards proposed by Kramer 41 for studies on breastfeeding and atopic diseases, that is, sufficient duration of breastfeeding (at least 2 months), assessment of exclusivity of breastfeeding for at least 3 months of life, strict diagnostic criteria, blinding of the investigators to the feeding pattern during their assessment of outcome, severity of outcome, assessment of age at onset of the disease, assessment of dose–response effect, adequate statistical power, evaluation of children at high risk of atopic diseases, and control for important confounding factors. The validity of maternally reported information on pregnancy and delivery was of high reliability 42, and it is unlikely that maternal recall of feeding practices has represented a bias. In fact, we focused on AD in early childhood with almost 80% of children being 3–12 months old, of whom around 52% of children were first born and most of the remaining ones (around 37%) were second born. Furthermore, results obtained from the reproducibility study indicated a good agreement for information on breastfeeding, in particular for month of stopping breastfeeding (K = 0.77; 95% CI: 0.67–0.88), with no differences when children younger or older than 12 months of age at the first interview were considered separately (i.e. K = 0.72; 95% CI: 0.58–0.86 and K = 0.84; 95% CI: 0.68–1.00, respectively). These results indicate that maternal recall bias of feeding practices has not played a major role in our study. Another issue may arise from those mothers considering their infants at risk of allergy who may prolong exclusive breastfeeding and introduce solid food later on into the child's diet. This could mask a protective effect of breastfeeding or even result in false‐positive associations between late solid food introduction and the occurrence of allergy (i.e. reverse causation) 43, 44. It is, however, unlikely that reverse causation had a major impact on our results because the outcome of interest was AD, which commonly precedes other allergic manifestations and represents the earlier step of the ‘atopic march’ 4. Further, detailed information on the timing of the first symptoms of AD and on the timing of the infants feeding practices allowed us to consider only exposures occurring before disease onset. Moreover, when adjusting our analysis for family history of allergy, the protective effect of early weaning persisted, and no heterogeneity between strata was found when feeding pattern was considered separately in children with and without atopic heredity. One may mention as a limitation the lack of information on maternal diet during gestation and lactation, which may affect breastmilk fatty acid composition 45, 46. However, a recent systematic review did not find consistent associations between mothers’ dietary intake during pregnancy and lactation and atopic outcomes in their children 47. Results also suggest a possible inverse association between partial breastfeeding and AD occurrence. This may be at least in part attributable to a residual confounding effect of vitamin K supplementation, usually prescribed to exclusively breastfed children and which appears here as a risk factor for AD. A recent large Danish cohort study that enrolled 44 594 mother–child pairs concluded that maternal vitamin K intake increased the risk of allergy during the first 7 years of life (i.e. asthma and allergic rhinitis) 48. The hygiene hypothesis, according to which infections protect from allergic diseases by influencing the skew in the immune reactivity toward a Th2‐dominated profile 49, may provide a plausible explanation to our findings on an increased risk of AD associated with prolonged exclusive (but not partial) breastfeeding. In fact, human milk contains several antimicrobial, anti‐inflammatory, immunomodulatory, and bioactive molecules that contribute to protect infants from infections 14, 50-54. Our finding of the favorable role of food diversity introduced within 4–5 months of life supports the view that exposure to a high diversity of antigens during a ‘critical early window’ of development favors the maturation of the mucosal immune system and oral tolerance 55. This is in keeping with a recent report describing that increased food diversity within the first year of life is significantly associated with increased expression of the transcript for forkhead box protein 3, a lineage marker for FoxP3+ thymus‐derived regulatory T lymphocytes. These regulatory T cells could downmodulate the allergic response through an inhibitory effect of IgE isotype switching, fitting well with their observation of a decreased expression of the Cε germline transcript in the same patients 56. In conclusion, our present observations confirm and extend recent evidence indicating that delayed exposure to solid food increases the risk of AD occurrence. Therefore, far from contradicting the overwhelming evidence of the unique nutritional and health benefits conferred by breastmilk, including, among others, the reduced incidence of respiratory and gastrointestinal tract infections 57, 58, our results indicate that recommendations for prolonged exclusivity of breastfeeding should be reconsidered in favor to a more diversified diet since 4 months of age, integrating gradually solid foods to breastfeeding.

Acknowledgments We would like to thank Mrs. Ivana Garimoldi for editorial assistance, Miss. Chiara Conte for the graphical support, Mrs. Cornelia Zinetti for technical assistance, the association ‘Pediatri di Famiglia di Bergamo e Provincia’ and Allegria Onlus for assistance and co‐operation in recruiting and interviewing the study subjects. We would also thank Prof. Aurora Parodi, Dr. Stefano Cambiaghi, Dr. Riccardo Cavalli, and Dr. Lucia Restano for their suggestions and all the families who participated in the study. Hygiene study group Collaborators: The HYGIENE study group contributes substantially to atopic dermatitis diagnostic criteria definition, to conduction of the reproducibility study, and to the enrollment and the interview process of all the patients included in the study. Members of this group are listed as follows (in alphabetical order within centers): Bergamo Pediatric Center: Dr. Fabio Agostinis (Pediatric Private Practice, Bergamo, Italy); Mrs Stefania Carminati (Dr. Fabio Agostinis medical office, Via Filippo Corridoni, 27, Bergamo BG). Bologna Dermatological Center: Dr. Iria Neri, Prof. Annalisa Patrizi, and Dr. Michela Starace (Department of Experimental, Diagnostic and Specialty Medicine – Division of Dermatology, University of Bologna, Bologna, Italy). Firenze Dermatological Center: Dr. Samantha Berti (Dermatologia Pediatrica, UOS Dipartimentale Dermatologia, USL 11 Empoli [FI], Italy); Dr. Maria Francesca Gola (Azienda Sanitaria di Firenze, Pediatric Practice, Florence, Italy); Prof. Massimo Gola (Allergological Dermatology Unit, Azienda Sanitaria and University of Florence, Piero Palagi Hospital, Florence, Italy). Garbagnate Milanese Pediatric Center: Dr. Alberto Martelli, Dr. Daniela Origgi, and Dr. Lorenza Serradori (Department of Pediatrics, Santa Corona Hospital, Garbagnate Milanese [MI], Italy). Genova Dermatological Center: Dr. Anna Graziella Burroni and Dr. Giulia Ciccarese (IRCCS A.U.O. San Martino – IST DISSAL E.C. Section of Dermatology Genoa Italy). Milan Dermatological Center: Dr. Michela Brena, Dr. Cristiana Colonna, and Prof. Carlo Gelmetti (Pediatric Dermatology, Department of Pathophysiology and Transplantation, Università degli Studi di Milano, Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico, Milan, Italy). Milan Pediatric Center: Prof. Susanna Esposito, Dr. Valentina Montinaro, Dr. Maria Francesca Patria, Dr. Walter Peves Rios, and Dr. Claudia Tagliabue (Pediatric Highly Intensive Care Unit, Department of Pathophysiology and Transplantation, Università degli Studi di Milano, Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico, Milan, Italy). Padova Pediatric Center: Dr. Elena Fontana and Prof. Anna Belloni Fortina (Pediatric Dermatology Unit, Department of Medicine, University of Padua, Padua, Italy). Siena / Grosseto Dermatological Center: Dr. Camilla Peccianti (U.O. Dermatologia, USL 9 Grosseto, Italy); Dr. Michele Pellegrino and Dr. Emanuele Trovato (S.C. Dermatologia, University of Siena, Italy). Treviglio Dermatological Center: Dr. Anna Di Landro (USC Dermatologia, Azienda Ospedaliera Papa Giovanni XXIII, Bergamo, Italy); Dr. Massimo Menchini, Dr. Vera Quadri, and Dr. Oriana Testagrossa (S.C. di Pediatria e Neonatologia, Azienda Ospedaliera di Treviglio, Treviglio [Bergamo], Italy). Verona Pediatric Center: Prof. Attilio Boner, Dr. Pasquale Comberiati, and Prof. Diego Peroni (Paediatric Section, Department of Life and Reproduction Sciences, University of Verona, Verona, Italy).

Funding This work was supported by the European Research Council, grant agreement number 250290.

Conflict of interest The authors declare that they have no conflicts of interest.

Author contributions Carlo La Vecchia and Liliane Chatenoud designed the study and were together with Carlotta Galeone and Claudio Pelucchi responsible for the conduction of data collection. Federica Turati and Paola Bertuccio performed the data analyses, and all authors contributed substantially to the interpretation of data. The paper was written by Federica Turati and Liliane Chatenoud and was critically revised by Luigi Naldi, Carlo La Vecchia, Jean‐François Bach. All authors approved the final version of the manuscript before submission.

Supporting Information Filename Description all12864-sup-0001-TableS1.docxWord document, 12.8 KB Table S1. Distribution of unmet diagnostic criteria among the 76 cases who did not fulfill all the diagnostic criteria at inclusion (at most 2 diagnostic criteria were not met). all12864-sup-0002-DataS1-S2.docxWord document, 14.2 KB Data S1. Methods: data collection details. Data S2. Methods: reproducibility study. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.