Exclusive breast-feeding is recommended for up to 6 months of age and supports the health of the mother and the child(1,Reference Victora, Bahl and Barros2) . In particular, there is a consensus about the beneficial effects of breast-feeding on child neurodevelopmental outcomes, which is supported by well-established associations(Reference Horta, Loret de Mola and Victora3,Reference Horta, de Sousa and de Mola4) . While an overwhelming 90 % of infants still receive breast milk at age 6 months globally, one out of two infants from high-income countries do not receive any breast milk at this age, with some Western countries showing even lower rates(Reference Neves, Vaz and Maia5). This is evident in France, a country with traditionally low rates of breast-feeding, where the corresponding rate of formula-fed infants (not necessarily exclusively) is high(Reference Wagner, Kersuzan and Gojard6) and parents use a broad range of infant formulas(Reference de Lauzon-Guillain, Davisse-Paturet and Lioret7,Reference Davisse-Paturet, Raherison and Adel-Patient8) . Comprehensive nationwide data demonstrate that the use of formulas with varying degrees of protein hydrolysates ranges between 2 and 7 % for extensively and partially hydrolysed forms, respectively(Reference de Lauzon-Guillain, Davisse-Paturet and Lioret7). According to European regulations, it is required to compare hydrolysed formulas against an approved control formula (hydrolysed or non-hydrolysed) in order to establish adequate growth among infants(Reference Turck and Heinonen9). However, there is scarce evidence of possible effects of hydrolysed formulas on neurodevelopmental outcomes among formula-fed infants(Reference Mennella, Trabulsi and Papas10).
Hydrolysed formulas include proteins that have been broken down partially or extensively in order to facilitate easier transition through the gut and decrease the likelihood of an immune reaction(Reference Vandenplas, Latiff and Fleischer11,Reference D’Auria, Salvatore and Acunzo12) . However, the efficacy of partially hydrolysed formulas in primary prevention of allergies is still debated(Reference Davisse-Paturet, Raherison and Adel-Patient8,Reference Vandenplas, Meyer and Chouraqui13–Reference Halken, Muraro and de Silva15) . According to the European regulatory framework, each new hydrolysed formula needs to be evaluated on an individual basis to ensure its safety and suitability, in addition to meeting the nutritional requirements of the infant(16). Different randomised controlled trials have reported adequate physical growth among children who consumed hydrolysed formulas(Reference Rzehak, Sausenthaler and Koletzko17–Reference Mennella, Inamdar and Pressman21). Moreover, some randomised controlled trials have shown growth among children consuming a non-hydrolysed formula to be accelerated compared with children consuming an extensively hydrolysed formula(Reference Rzehak, Sausenthaler and Koletzko17,Reference Mennella, Ventura and Beauchamp20,Reference Mennella, Inamdar and Pressman21) . However, only one randomised controlled trial has examined the influence of protein hydrolysed infant formula on neurodevelopmental outcomes. In this trial, infants fed with extensively hydrolysed formula up to 8·5 months of age had more favourable cognitive outcomes during the first year of life, compared with those fed with regular cows’ milk formula(Reference Mennella, Trabulsi and Papas10).
Research that has distinguished between different types of formulas according to their content in protein hydrolysates has oftentimes drawn links between hydrolysed formulas and breast milk – the gold standard for infant feeding – insofar as they both contain free amino acids. Human milk is characterised by a high content in free amino acids (in particular glutamate), which is seven times higher in extensively hydrolysed formulas(Reference Agostoni, Carratù and Boniglia22,Reference Ventura, San Gabriel and Hirota23) . However, free amino acids are rare in non-hydrolysed formula(Reference Agostoni, Carratu and Boniglia24). As outlined above, there is preliminary evidence from a randomised controlled trial involving term infants, which has implicated extensively hydrolysed formulas (with a high ratio of free amino acids) in favourable developmental outcomes, compared with regular formula(Reference Mennella, Trabulsi and Papas10). The same line of research has previously shown higher satiation among infants fed regular formula with added glutamate compared with regular formula. The authors have clearly framed their observation in light of the role of glutamate (and other free amino acids for that matter) in signalling satiation to the central nervous system(Reference Ventura, Beauchamp and Mennella25). Interestingly, the satiation was equally high when infants were fed extensively hydrolysed formula, which also contains high levels of glutamate and other free amino acids(Reference Ventura, Beauchamp and Mennella25). While dietary glutamate is not considered to enter the brain in relation to the blood–brain barrier, it may indirectly activate brain areas (and conceivably influence brain functions, including ingestive behaviours) since it is sensed in the oral cavity and the intestine(Reference Torii, Uneyama and Nakamura26,Reference Kondoh, Mallick and Torii27) . Thus, it may have the capacity to transfer information to the central nervous system through the vagal afferent system(Reference Torii, Uneyama and Nakamura26,Reference Kondoh, Mallick and Torii27) . Moreover, animal studies suggest that administration of monosodium glutamate directly affects several behavioural aspects and cognitive capacities, though findings are mixed, also depending on the age of assessment(Reference Kiss, Hauser and Tamás28,Reference Onaolapo, Odetunde and Akintola29) . Thus, the free amino acid content, in particular glutamate, of hydrolysed formulas (extensive and partial forms) compared with regular ones may provide a possible mechanistic explanation of their associations with neurodevelopment among a small sample of term infants(Reference Mennella, Trabulsi and Papas10).
Aim
The aim of the present study is to examine the associations between the degree of protein hydrolysis in infant formula and child neurodevelopment up to 3·5 years of age, among formula-fed infants. We hypothesise that the early consumption of hydrolysed formulas predicts more favourable neurodevelopmental outcomes within the first 4 years of life, with stronger associations for formulas with extensive hydrolysates than for those with partial hydrolysates.
Materials and methods
Study population
The present analyses were based on data from the French Étude Longitudinale Française depuis l’Enfance (ELFE) study, a nationwide birth cohort, which included 18 329 infants born in 2011 in a random sample of 349 maternity wards around metropolitan France across four recruitment waves(Reference Charles, Thierry and Lanoe30). Inclusion criteria included singleton or twin births, term and moderate to late preterm births (≥33 gestational weeks), mother aged ≥18 years old and no plan to move outside metropolitan France within the next 3 years.
Ethical approval
This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving human subjects/patients were approved by the Advisory Committee for the Treatment of Information on Health Research (Comité Consultatif sur le Traitement des Informations pour la Recherche en Santé/File number 10.623), the National Agency Regulating Data Protection (Commission Nationale Informatique et Libertés/File number 910504) and the National Statistics Council (File number 2011X716AU). Written informed consent was obtained from all subjects/patients. Mothers provided written informed consent for themselves and their children(Reference Charles, Thierry and Lanoe30). Fathers could also provide consent if present at the maternity ward; otherwise, they were informed afterwards and they could object to their child’s participation.
Infant feeding
Data on milk feeding practices were collected monthly from 2 to 10 months and then at 12 and 24 months. From these data, any breast-feeding duration was calculated as previously described(Reference Wagner, Kersuzan and Gojard31). From 2 to 10 months, the name and brand of the formula were reported for formula-fed infants(Reference de Lauzon-Guillain, Davisse-Paturet and Lioret7). According to the label on the formulas defining the degree of protein hydrolysis, formulas were classified as containing non-hydrolysed proteins (nHF), partially hydrolysed proteins (pHF), extensively hydrolysed proteins (eHF) and amino acid mixture (AA)(Reference Davisse-Paturet, Raherison and Adel-Patient8). While eHF and AA explicitly address cows’ milk protein allergy(Reference Koletzko, Niggemann and Arato32), formulas with HA label are, by definition, based on pHF. In France, some infant formulas without the HA label (pHF/non-HA) also contained partially hydrolysed proteins, probably with a lower level of hydrolysis. Thus, we decided to consider them separately. Terms on the label of formulas (ingredients), which facilitated the classification of formulas in terms of protein hydrolysates in the present paper, are shown on online Supplementary Table S1.
Child neurodevelopment
The French version of the Child Development Inventory (CDI) was administered during the phone interviews with parents at 1 and 3·5 years post-partum(Reference Charles, Thierry and Lanoe30,Reference Ireton33,Reference Duyme and Capron34) . At age 1 year, items adapted to the developmental age were selected from the full version of the CDI, while at age 3·5 years, the brief version was used – in particular the two parts highlighting developmental milestones at this age(Reference Duyme and Capron34). The CDI-1 assesses six developmental domains (social skills, self-help, gross motor skills, fine motor skills, language expression and language comprehension), and the CDI-3·5 assesses two additional domains (characters and numbers). Response options for each item were yes (1) if the child had achieved the described ability and no (0) if not. The summary score of items in CDI-1 and CDI-3·5 was used to assess overall child neurodevelopment, ranging from 0 to 50 at 1 year and 17 to 62 at 3·5 years (under the assumption that earlier milestones had been reached; thus, the minimum score at 3·5 years corresponds to the maximum score of the CDI items for younger ages)(Reference Duyme and Capron34).
Second, the brief French version of the MacArthur-Bates Communicative Development Inventory was used during phone interviews with parents at 2 years (MB-2) to assess children’s early language development on a 100-point continuous score (each point corresponds to a word expressed by the child)(Reference Kern, Langue and Zesiger35).
Last, the Pictures Similarities sub-scale from the British Ability Scale was administered by trained research assistants during home visits at 3·5 years (PS-3·5) to assess child cognitive development in terms of their pictorial reasoning ability(Reference Elliott, Smith and McCulloch36). The score ranged from 10 to 119.
Perinatal, family and feeding characteristics
Data on family background characteristics were collected by trained interviewers at the maternity ward, and they were complemented by data on the newborn according to medical records(Reference Charles, Thierry and Lanoe30). Complementary information regarding the families was obtained during phone interviews at 2 months and 1 year post-partum.
As regards family background characteristics, the following information was of interest: mother’s age (<25 years, 25–29 years, 30–34 years, ≥35 years), education (upper secondary or lower, high school diploma, 3-year university education, at least 5-year university education), employment (employed, unemployed, out of the labour force – i.e. housewife, retired, students) and migration status (migrant/not born to French parents, descendant of at least one migrant parent, majority population/born to French parents), household income/consumption unit (≤1111 €/month, 1112–1500 €/month, 1501–1944 €/month, 1945–2500 €/month, >2500 €/month), maternal smoking during pregnancy (never smoker, smoker only before pregnancy, smoker only in early pregnancy, smoker throughout pregnancy), parental history of allergy (no parent with allergy, at least one parent with allergy), sibling history of allergy (no sibling, no sibling with allergy, at least one sibling with allergy) and mother’s diet quality during the last trimester of pregnancy using the Probability of Adequate Nutrient intake-based Diet quality index (PANDiet) score (adapted for pregnancy, which reflects nutrient-based reference guidelines; total scores range from 0 to 100)(Reference Kadawathagedara, Ahluwalia and Dufourg37,Reference Bianchi, Mariotti and Verger38) . The region of residence (Paris region, North, East, Paris Basin – East, Paris Basin – West, West, Southwest, Southeast, Mediterranean) of the family, as well as the urban/rural area of living, was determined from the postal code of residence. At the 1-year interview, the mother indicated the frequency (rarely/never/sometimes, often) of some activities with their child: playing, reading books, drawing, speaking and tickling/massage(Reference Charles, Thierry and Lanoe30,Reference Martinot, Adjibade and Taine39) . The modal value of these activities was used to estimate a maternal stimulation score indicating a family environment that is conducive to favourable child development(Reference Tucker-Drob and Harden40,Reference Klasen and Crombag41) .
Characteristics related to the infant include the following: sex (boy, girl), gestational age (in weeks), physician consulted between hospital discharge and 2 months post-partum (general practitioner, paediatrician, another child doctor, none/other) and allergy to cows’ milk (yes/no). Birth weight was classified into three categories (small/adequate/large for gestational age) according to the French Audipog reference curves(Reference Mamelle, Munoz and Grandjean42).
Sample selection
The ELFE sample consisted of 18 329 infants and their families who fulfilled the inclusion criteria and consented to participate, at least in the beginning (Fig. 1). Fig. 1 shows the consecutive steps leading to the analytical sample, which was used for the main analyses (complete-case). Additional (sensitivity) analyses accounted for missing data through multiple imputations of the confounding variables. The analyses and the rationale thereof are described in the next section. Varying sample sizes across analyses are due to missing data in the respective neurodevelopmental scores (Fig. 1).
Fig. 1. Flow chart for the analyses. *Varying sample sizes due to missing data in the respective neurodevelopmental score. CDI, Child Development Inventory; BAS, British Ability Scales.
Families who withdrew consent (n 57) were excluded from the analyses. For families with twins, we proceeded to a random selection of one twin to avoid clustered data (n 287). Further exclusions were performed in relation to the exposure, that is, no follow-up at 2 months (n 1696), no formula feeding at 2 months (n 5054) and no information on the degree of protein hydrolysis in formula (n 658). From the remaining sample, infants with missing data across all neurodevelopmental outcomes were further excluded (n 1205) along with those who had missing data in adjustment variables (n 2393), leading to the analytical sample (n 6979). The analytical sample provided the basis for the complete-case analyses (main models) according to the availability of neurodevelopment data, that is, at 1 year for CDI (n 6977), at 2 years for MB (n 6145) and at 3·5 years for CDI (n 5696) and for PS (n 4511). Multiple imputations of the missing confounding variables will be described in the next section. These analyses were based on the analytical sample including missing data on confounding variables (n 9372), and they were performed according to data availability on the neurodevelopmental outcomes, on 8980 at 1 year for CDI, 7962 children at 2 years for MB and 7334 and 5695 at 3·5 years for CDI and for PS, respectively.
As compared with children and their families who were included in the analyses, those who were excluded (apart from those who had withdrawn consent and the selection of twins, n 11 006) were characterised by slightly lower income levels (mean €1600 v. €1675 per consumption unit, P < 0·001), more mothers with a migration history (24·5 % v. 15·7 %, P < 0·001) and with a higher education level (22·5 % v. 17·6 % at least 5-year university education, P < 0·001) and higher rates of mothers’ never smoking (59·4 % v. 53·7 %, P < 0·001). On the other hand, excluded sample was similar to the included sample in terms of child sex (girls: 48·6 % v. 48·4 %, P = 0·80), mean gestational age (39·2 v. 39·2 weeks, P = 0·06) and mean maternal age (30·7 v. 30·9 years old, P = 0·07).
Statistical analyses
For the total analytical sample, frequencies (n) and mean values and standard deviations were computed.
We considered the following neurodevelopmental outcomes: one summary and six domain-specific scores for child motor and cognitive development were based on the CDI at 1 and 3·5 years post-partum, a score for early language development was based on the brief French MacArthur-Bates Inventory at 2 years post-partum (MB-2) and a score for pictorial reasoning ability was additionally assessed according to the PS sub-scale at 3·5 years post-partum (PS-3·5).
The six domain-specific sub-scores of the CDI at 1 and 3·5 years did not follow a normal distribution; thus, they were divided into quartiles. Children within the lowest quartile were considered as having a poor developmental sub-score and they were compared with children from the three upper quartiles (reference group). The overall CDI scores as well as the MacArthur-Bates scores and the Picture Similarities score were considered as continuous variables.
Binary logistic and linear regression models were used to conduct the unadjusted analyses between the degree of protein hydrolysis in infant formula and neurodevelopmental outcomes. Multivariable logistic and linear regression models were run to account for confounding factors. These were identified from the literature and selected using the directed acyclic graph method(Reference Ferguson, McCann and Katikireddi43). Then multivariable models were adjusted for: study design variables (maternity size and recruitment wave), socio-demographic and family characteristics (parental stimulation, maternal age, maternal employment, maternal educational attainment, migration history, household income, region of residence, urban/rural area), infant characteristics (child sex), perinatal and health-related factors (gestational age in weeks, gestational age, parents’ and siblings’ history of allergies, type of physician consulted between hospital discharge and 2 months of age, any breast-feeding duration, cows’ milk protein allergy reported at the 2-month interview) and lifestyle factors (maternal smoking during pregnancy, dietary quality using a validated scoring system adapted for the French population and to nutritional needs during pregnancy). In addition, all models were adjusted for the child’s age (in months) at the time of the respective neurodevelopmental assessments.
The main analyses were conducted on the complete-case sample. Sensitivity analyses were performed using additional models for sub-samples of infants without any congenital malformations (n 6713) and term infants (n 6604). These sub-samples were excluded because of the clear links between these birth outcomes and later neurodevelopmental outcomes(Reference Villar, Restrepo-Mendez and McGready44,Reference Cheong, Doyle and Burnett45) . Additional sensitivity analyses included sub-samples of infants who did not change formula over the first 2 months’ follow-up (n 3680) and those who did not change formula between 2 and 6 months (n 3970). Based on infants with complete data on infant formula consumption between 2 and 6 months (at 2 months n 4063 for nHF; n 118 for pHF/non-HA; n 190 for pHF/HA; n 42 for eHF/AA), an overwhelming 93 % of infants consuming nHF at 2 months showed a consistent consumption of this formula between 2 and 6 months. By contrast, one out of two infants consuming a hydrolysed formula (of any type according to the classification in the present paper) at 2 months showed an inconsistent use of it between 2 and 6 months (45·8 %, 51 % and 45·2 % for pHF/non-HA, pHF/HA and eHF/AA, respectively).
To deal with selection and attrition bias, a sensitivity analysis was conducted with weighted data on the complete-case sample. Weighting was calculated to take into account the inclusion procedure and biases related to non-consent or attrition and also included calibration on margins from the state register’s statistical data and the 2010 French National Perinatal study(Reference Blondel, Lelong and Kermarrec46) on the following variables: age, region, marital status, migration status, level of education and primiparity(Reference Siméon47). A specific weighting was calculated for the sub-samples included in the complete-case analyses at 1- 2-, and 3·5-year follow-ups, respectively.
Finally, a sensitivity analysis was performed with multiple imputation of confounding variables to deal with missing data(Reference Sterne, White and Carlin48). This approach has been integral to the analytical plan of the ELFE study – a nationwide birth cohort with long follow-up(Reference Charles, Thierry and Lanoe30) – and it has been applied in multiple analyses in order to address the bias introduced due to missing data(Reference Martinot, Adjibade and Taine39,Reference de Lauzon-Guillain, Thierry and Bois49–Reference Schmengler, El-Khoury Lesueur and Yermachenko51) . Based on the assumption that confounding variables were missing at random and using the fully conditional specification method, the procedure of multiple imputations generated five independent and complete data sets (SAS software: MI procedure, FCS statement, NIMPUTE option). Pooled effect estimates were then calculated for each outcome of interest (SAS software: MIANALYSE procedure). For significance testing of categorical variables, the median of the P values from the imputed data analyses in each data set was used(Reference Eekhout, van de Wiel and Heymans52).
All analyses were carried out using SAS v9.4 (SAS Institute Inc.). Significance was set at P < 0·05.
Results
Table 1 summarises infant and family characteristics according to the degree of protein hydrolysis of the formula in the analytical sample (n 6979). The majority of infants (n 6432) consumed nHF, and the rest consumed increasingly hydrolysed formulas as follows: pHF/non-HA (n 189), pHF/HA (n 288) and eHF/AA (n 70). The majority of infants consuming eHF/AA had cows’ milk protein allergy at 2 months. By contrast, infants fed nHF had higher rates of no parental family history for allergies. Overall, summary and sub-domain neurodevelopmental scores were found to be similar across formula groups with an increasing degree of protein hydrolysis (Table 2).
Table 1. Sample characteristics according to the degree of protein hydrolysis in infant formula consumed at 2 months (n 6979)
(Numbers and percentages; mean values and standard deviations)
HA label, hypoallergenic label; GA, gestational age; CMPA, cows’ milk protein allergy.
* Parental stimulation was defined according to the frequency of activities (e.g. drawing, playing) with the child, as reported by mothers at 1-year follow-up.
† Size at gestational age is classified according to birth weight.
‡ From maternity unit or from child and maternal protection centres.
Table 2. Neurodevelopmental scores across infant formulas with an increasing degree of protein hydrolysis (n 6979)
(Mean values and standard deviations; numbers and percentage)
* At 1 year <6 and at 3·5 years <9.
† At 1 year <4 and at 3·5 years <7.
‡ At 1 year <3 and at 3·5 years <8.
§ At 1 year <7 and at 3.5 years <6.
‖ At 1 year <5 and at 3·5 years <9.
¶ At 1 year <7 and at 3·5 years <8.
a Varying sample sizes due to missing data in the respective neurodevelopmental score.
b High risk score (<25th percentile).
The degree of protein hydrolysis of the formula fed at 2 months was not related to the overall neurodevelopmental scores from 1 to 3·5 years, in the main analyses (complete-case adjusted) and those adjusted after multiple imputations (Table 3 and online Supplementary Table S4). When the specific weighting was applied to account for selection and attrition bias, compared with infants having consumed nHF at 2 months, infants having consumed pHF/non-HA had lower CDI-1 score, whereas infants having consumed eHF/AA had higher CDI-3·5 score (Table 3).
Table 3. Adjusted estimates of summary developmental scores at 1, 2 and 3·5 years across formulas with increasing degree of protein hydrolysis consumed at 2 months, complete-case analyses
(Numbers; estimates and 95 % confidence intervals)
CDI-1, 1-year Child Development Inventory; MB-2, 2-year MacArthur-Bates Communicative Development Inventory; CDI-3·5, 3·5-year Child Development Inventory; PS-3·5, 3·5-year Picture Similarities ability score from the British Ability Scale; nHF, non-hydrolysed formula; pHF/non-HA, partially hydrolysed formula without any hypoallergenic label; pHF/HA, partially hydrolysed formula with a hypoallergenic label; eHF/AA, extensively hydrolysed formula, or formula based on amino acids.
a Values are estimates (95 % CI) from linear regression models adjusted for covariates: child age at each assessment, mother’s age, education, employment, and migration status, household income, maternal smoking during pregnancy, parental and sibling history of allergy, mother’s dietary quality during pregnancy, urban/rural are of living, region of residence, parental stimulation, child sex, gestational age and size according to gestational age, physician consulted between hospital discharge and 2 months post-partum, allergy to cows’ milk any breast-feeding duration.
b Varying sample sizes due to missing data in the respective neurodevelopmental score.
* Estimates are adjusted for covariates and are weighted in order to account for the inclusion procedure and biases related to non-consent or attrition.
When considering the specific developmental domains separately, the degree of protein hydrolysis in the 2-month infant formula was not related to the risk of having a poor score on social skills, self-help, fine motor skills, language expression and language comprehension, at the ages of 1 or 3·5 years, in the main analyses (complete-case adjusted), shown in Table 4.
Table 4. Adjusted OR of having a poor developmental sub-score across formulas with increasing degree of protein hydrolysis consumed at 2 months, complete-case analyses at 1-year follow-up (n 6977) and at 3·5-year follow-up (n 5696)
(Odds ratios and 95 % confidence intervals)
nHF, non-hydrolysed formula; pHF/non-HA, partially hydrolysed formula without any hypoallergenic label; pHF/HA, partially hydrolysed formula with a hypoallergenic label; eHF/AA, extensively hydrolysed formula, or formula based on amino acids.
a Values are odds ratios (95 % CI) from logistic regression models adjusted for covariates: child age at each assessment, mother’s age, education, employment, and migration status, household income, maternal smoking during pregnancy, parental and sibling history of allergy, mother’s dietary quality during pregnancy, urban/rural area of living, region of residence, parental stimulation, child sex, gestational age and size according to gestational age, physician consulted between hospital discharge and 2 months post-partum, allergy to cows’ milk and any breast-feeding duration.
b Varying sample sizes due to missing data in the respective neurodevelopmental score.
* Estimates are adjusted for covariates and are weighted in order to account for the inclusion procedure and biases related to non-consent or attrition.
At the age of 1 year, compared with children having consumed a non-hydrolysed formula, those having consumed pHF/HA were more likely to have a poor score on the gross motor sub-scale (Table 4). While these findings at the 1-year follow-up were consistent for the adjusted main analyses and in the analyses after multiple imputations, they did not reach significance after weighting, though the trend for effects remained the same (Table 4 and online Supplementary Table S5).
At the age of 3·5 years, early consumption of pHF/non-HA at 2 months was associated with a lower risk of having poor social skills, compared with having consumed non-hydrolysed formula (Table 4). This finding was also shown in the analyses with multiple imputations and in the weighted analyses, while children having consumed eHF/AA were less likely to have a poor score on fine motor skills in the weighted analyses only (Table 4).
The findings of the main analyses (complete-case) were in line with the unadjusted analyses (online Supplementary Tables S2 and S3) and the sensitivity analyses including specific sub-samples, except for the analyses at 3·5 years including infants without congenital malformations which were in line with the weighted analyses (online Supplementary Tables S4 and S5).
Discussion
The present study is the first to examine the effects of the use of formula with varying degrees of protein hydrolysis on child neurodevelopment up to 3·5 years of age in a birth cohort. The degree of protein hydrolysis in infant formula consumed at 2 months of age was not related to overall neurodevelopmental scores up to 3·5 years. Some associations were found with the gross motor skills CDI sub-domain, but they were not consistent at 1 and 3·5 years as well as across all sensitivity analyses (including specific sub-samples and also accounting for attrition and selection bias through weighted data as well as addressing missing data through multiple imputation procedures). Nonetheless, associations were only shown for formulas with partial hydrolysates, and they were not extended to those with extensive hydrolysates.
As expected, the use of hydrolysed formulas, in particular the extensively hydrolysed ones, aligned with the presence of cow’s milk protein allergy and a family history of allergy. Such findings reflect current recommendations and/or common practices regarding the use of hydrolysed formulas(Reference Koletzko, Niggemann and Arato32,Reference Meyer, Smith and Sealy53,Reference Strozyk, Horvath and Meyer54) . There is scarce evidence regarding long-term effects (at least over 1 year of age) on child neurodevelopment of the use of infant formula in infancy. Therefore, we cannot directly compare our findings to previous studies. Our findings do not confirm the hypotheses by Mennella et al. (Reference Mennella, Trabulsi and Papas10), who have provided preliminary evidence on certain favourable effects of formula with eHF on motor skills and cognition among infants younger than the age of 1 year over 8 months follow-up. In addition, our findings do not support stronger associations according to an increasing degree of protein hydrolysis; we observed unfavourable associations of gross motor skills with partially hydrolysed formula only. Yet, the observed associations were transient, that is, they were present at 1 year of age but they were not significant anymore at the 3·5-year follow-up. This may support the argument of transient neurodevelopmental effects of formula in early life which was also presented by Mennella et al. (Reference Mennella, Trabulsi and Papas10) according to monthly assessments at a younger age than in the present study (i.e. between 5·5 and 8·5 months of age). Further evidence on transient effects may relate to the free amino acid content, especially glutamate, which marks an important difference between hydrolysed protein formulas and the regular ones(Reference Ventura, Beauchamp and Mennella25). In particular, an animal study involving rats fed monosodium glutamate during the neonatal period found transient effects of the use of monosodium glutamate on locomotor activity, whereby at 3 weeks of age there was an increase in locomotor activity which was followed by a marked hypoactivity the week after(Reference Kiss, Hauser and Tamás28). These follow-up times in the animal study roughly correspond to those examined in our study(Reference Wang, Lai and Deng55). Taken together, these findings highlight the relevance of length and timing of follow-up across studies due to the high neuroplasticity during the early life stages(Reference DeMaster, Bick and Johnson56).
It is conceivable that the literature on infant formula with hydrolysed proteins focuses on the free amino acid content of these formulas. Free amino acids are also present in human milk at higher concentrations than regular formula, and they could explain some of the differences in developmental indicators between breast-feeding and formula feeding(Reference Mennella, Inamdar and Pressman21,Reference Ventura, Beauchamp and Mennella25,Reference He, Sotelo-Orozco and Rudolph57,Reference Grote, Verduci and Scaglioni58) . However, drawing parallels between the high content in free amino acids (or any other biological component for that matter, such as biologically active molecules and microbiota) of hydrolysed formulas and breast milk may fail to account for other aspects of breast-feeding that may promote cognitive development among children, such as infant attachment, parenting practices and the home environment(Reference Gibbs and Forste59–Reference Azad, Vehling and Chan62). For example, McCormick et al. (Reference McCormick, Caulfield and Richard63) showed that while aspects of the home environment along with certain nutrients did differentiate between the identified child cognitive trajectories (i.e. consistently high scores, increasing scores, intermediate scores with early and late decline, and consistently low scores), exclusive breast-feeding had limited discriminatory power in relation to cognitive development.
Strengths and limitations
ELFE is a large birth cohort in France. Its prospective design limits recall bias for both exposure and outcome assessments. The very large sample and the collection of detailed socio-demographic or economic data ensure good statistical power and favour control for potential confounders, although residual confounding may remain. Of note, indicators for developmental delays in early infancy were not considered. Developmental outcomes may indirectly relate to the choice of infant formula since limited tolerance to standard feeds and regurgitation, which appear more common among children with developmental delays(Reference Chaidez, Hansen and Hertz-Picciotto64,Reference Schieve, Gonzalez and Boulet65) , may have prompted the use of hydrolysed formulas in infancy(Reference de Lauzon-Guillain, Davisse-Paturet and Lioret7,Reference Meyer, Smith and Sealy53) . Moreover, due to the small size (n 1) of the analytical sample consuming elemental formula, it was collapsed with the most similar category in terms of free amino acid content, namely extensively hydrolysed formula (n 69). In fact, the use of any type of formulas with protein hydrolysates was not very prevalent; the highest prevalence was registered for the use of partially hydrolysed forms with hypoallergenic label (just over 4 %). Although these findings are based on data from a nationwide cohort and they do map the use of formulas in France, large samples for the study of hydrolysed formulas have not been available(Reference de Lauzon-Guillain, Davisse-Paturet and Lioret7). Thus, our analyses did not have the capacity to distinguish between extensively hydrolysed and elemental formulas, and they were generally limited by the low statistical power as per the groups of high degree of protein hydrolysis. Finally, the sample considered for the present analyses was based on a higher rate of privileged families than the initial ELFE sample, which could limit the generalisation of our results(Reference Charles, Thierry and Lanoe30). However, sensitivity analyses based on weighted data, accounting for selection and attrition biases, gave similar findings, suggesting that this bias had limited impact on our conclusions. Similarly, missing data can introduce bias, yet analyses based on imputed data yielded similar findings. Moreover, diverse ranges of brands for the same type of hydrolysed infant formulas were used by families, and changes in infant formula were frequent. Sensitivity analyses including infants who had not changed infant formula for up to 2-month follow-up and those who did not change the type of formula between 2 and 6 months pointed to similar conclusions. Regarding glutamate, which is implicated in a mechanistic explanation of the initial hypothesis, the glutamate content of indicated formulas was not assessed. Using parental questionnaires may have introduced biases, including social desirability bias and imprecision, but parents completed a battery of valid and reliable instruments(Reference Duyme and Capron34–Reference Elliott, Smith and McCulloch36) to allow for international comparisons and reduce the above-mentioned biases. Still, the on-site assessment of the Picture Similarities test (as part of the British Ability Scales) by trained research assistants showed a similar pattern of (no) association.
Conclusion
In summary, among formula-fed infants, the degree of protein hydrolysis in infant formula fed at 2 months was not associated with overall neurodevelopmental scores up to 3·5 years of age. These findings are in favour of the safety of use of such formulas, beyond growth trajectories(Reference Strozyk, Horvath and Meyer54,Reference Rigo, Schoen and Verghote66,Reference Dupont, Hol and Nieuwenhuis67) . However, it would be important to replicate these analyses across settings with a different distribution of the studied formulas, as well as in more vulnerable populations.
Acknowledgements
The authors thank the scientific coordinators (B Geay, H Léridon, C Bois, JL Lanoé, X Thierry, C Zaros), IT and data managers, statisticians (M Cheminat, C Ricourt, A Candea, S de Visme), administrative and family communication staff, and study technicians (C Guevel, M Zoubiri, L G L Gravier, I, Milan, R Popa) of the ELFE coordination team as well as the families that gave their time for the study. We also thank the French Institute for Demographic Studies (INED) researchers for the sociodemographic variables created as part of the ANR-funded Veniromond project. The ELFE survey is a joint project between the French Institute for Demographic Studies (INED) and the National Institute of Health and Medical Research (INSERM), in partnership with the French blood transfusion service (Etablissement français du sang, EFS), Santé publique France, the National Institute for Statistics and Economic Studies (INSEE), the Direction générale de la santé (DGS, part of the Ministry of Health and Social Affairs), the Direction générale de la prévention des risques (DGPR, Ministry for the Environment), the Direction de la recherche, des études, de l’évaluation et des statistiques (DREES, Ministry of Health and Social Affairs), the Département des études, de la prospective et des statistiques (DEPS, Ministry of Culture) and the Caisse nationale des allocations familiales (CNAF), with the support of the Ministry of Higher Education and Research and the Institut national de la jeunesse et de l’éducation populaire (INJEP). Via the RECONAI platform, it receives a government grant managed by the National Research Agency under the ‘Investissements d’avenir’ programme (ANR-11-EQPX-0038, ANR-19-COHO-0001).
This study was funded by an ANR grant (InfaDiet project, no ANR-19-CE36-0008). The funders had no role in the study design, data collection and analysis, decision to publish or preparation of the manuscript.
M. S. conceptualised the study, conducted the formal analyses and drafted the manuscript. B. L.-G. and S. N. conceptualised and designed the study, designed instruments for nutritional data, supervised data collection and management, contributed to the interpretation of the findings and critically reviewed the manuscript. J. Y. B. and M. T. managed data on neurodevelopment and critically reviewed the manuscript. M.-A. C. coordinated the study and reviewed the manuscript.
The authors had no conflicts of interest relevant to this article to disclose.
Supplementary material
For supplementary material referred to in this article, please visit https://doi.org/10.1017/S0007114523000211
