The Allergy Reduction Trial (A.R.T.) is a multicenter, double-blinded, parallel, randomized controlled study assessing differences in the incidence of CMPA [presented as part of milk related allergic manifestations (MRAM) in the original study protocol] and AD within the first 6 months of life in apparently healthy term infants at high-risk of developing allergy (family history of allergy), conducted in 6 centers in 3 countries; Bulgaria (1), Cyprus (1) and Greece (4) between 2017–2019. Two intervention formulas were provided, either to be fed exclusively or supplementary to breastfeeding: a) a partially hydrolyzed whey-based formula (pHF) or b) an intact protein (standard) cow's milk formula (SF) containing both whey and casein fractions in a 60:40 ratio. Exclusively breastfed infants were followed as a parallel observational group.

Only term infants (≥37 weeks), with birth weight ≥2500g, postnatal age <5 days, apparently healthy with no signs of allergy, who were exclusively breastfed or fed with an extensively hydrolyzed infant formula (eHF) since birth were recruited. The inclusion criteria are detailed in the Supplementary Material (SM).

The study protocol, information letter to parents/legal guardians, and written informed consent form were approved by the appropriate independent ethics committee in each center as detailed in the SM. The study was conducted in accordance with the guidelines of the Declaration of Helsinki and the International Conference on Harmonization guidelines on Good Clinical Practice and registered in the Netherlands Trial Registry [Identifier: Trial NL6120 (NTR6259)].

During the 7^th-9^th month of gestation (or shortly after delivery), families attending public and private maternity clinics, were interviewed regarding family history of allergy (Atopic Dermatitis/Eczema, Allergic Asthma, Allergic Rhinitis-Conjunctivitis, Urticaria and Food Allergy), using an enriched validated questionnaire (10) to identify infants at high-risk of developing allergy (at least one parent or sibling with doctor-diagnosed allergy). Within 4 days after delivery, parents willing to participate in the study signed the informed consent form.

All mothers were strongly encouraged to breastfeed and inclusion in the study was not dependent on mothers' choice of feeding regimen. In case of non-exclusive breastfeeding, infants were randomly allocated to the pHF or the SF based on computer-generated randomization tables with stratification for gender, type of feeding [exclusively formula-fed or mixed-fed (breastfeeding combined with infant formula)] and presence of AD in the family. Upon parents' request, infants from the exclusively breastfed group could be allocated to one of the two formula groups within 10 weeks of life.

The intervention formulas were a partially hydrolyzed whey-based formula (pHF; Frisolac Gold preventive HA) and an intact protein/standard formula (SF; Frisolac Gold) nutritionally suitable for the first 6 months of life (Supplementary Table S1). Both formulas were produced in the Netherlands by FrieslandCampina and packed in identical white unlabeled 400g tins with the code-name of each formula group (FCA and FCB) at the bottom. This was performed by the packaging department at FrieslandCampina not involved in the study. All study personnel, FrieslandCampina employees involved in the study and parents/legal guardians were blinded to the study formulas until the whole study was completed and database was locked.

All adverse events and actions taken were recorded throughout the study and monitored by an independent Data Safety Monitoring Board (DSMB).

After allocation, infants visited the study centers for bi-monthly follow-up assessments (at 2^nd, 4^th, and 6^th month) during the first 6 months of life. Additional visits were performed if needed, such as in case of development of any signs of allergy or adverse events. Infants had not consumed any other formula with intact or partially hydrolyzed protein prior to allocation and solid foods were allowed after the age of 4 months. No dietary restrictions were advised to breastfeeding mothers. Formula intake compliance was evaluated using a 7-day milk diary completed the week preceding the 1^st, 2^nd, 4^th, and 6^th month of age. If formula consumption was <40 mL/kg body weight/day during the 1^st month or <60 mL/kg body weight/day during the 2^nd month (and thereafter), the infant was considered as a drop-out (22).

At follow-up visits, infants were clinically examined by experienced pediatricians and nurses for the presence of CMPA and AD. Suggestive CMPA and AD symptoms were objectively scored using the Scoring for Atopic Dermatitis [SCORAD (23)] tool and the awareness Cow's Milk-related Symptom Score [CoMiSS (24, 25)] tool. Furthermore, the “Screening for IgE- and non-IgE-mediated food allergy symptoms questionnaire” was completed (26).

Anthropometric measurements (weight, length, and head circumference) were performed by the same two well-trained research team members at each center, using calibrated digital infant scales (SECA 354), infantometers (SECA 210) and non-elastic tapes (SECA 211) respectively. All measurements were performed in triplicates and averaged.

The sample size calculation was based on two studies: the German Infant Nutritional Intervention (GINI) study (2003) (10), and the study performed by Halken et al. (2020) (28, 30). Assuming a drop-out rate of 30%, 158 infants had to be included per treatment arm. The detailed sample size estimation is presented in the SM.

Statistical analysis was carried out by an independent statistical company (OCS Life Sciences, The Netherlands) using the SAS software version 9.4 (SAS Institute, Cary, NC). Correction for multiple testing was done for the primary parameters by correcting the threshold using the Hochberg procedure (31).

Next to primary outcomes, post-hoc analysis was performed for CMPA defined as: (1) CMPA+_SPT including also IgE-mediated CMPA cases considered as CMPA based on clinical history and positive Skin Prick Test (SPT) to cow's milk without performing OFC and (2) CMPA+_susp including confirmed CMPA and CMPA+_SPT cases plus the clinically suspected cases of CMPA in which symptoms resolved after introducing an extensively hydrolyzed formula, but parents did not consent for OFC.

CMPA (confirmed by OFC) and AD outcomes were analyzed in the Intention-to-Treat (ITT) and Per-Protocol (PP) population. The ITT analysis set includes all study participants who were randomized to one of the study formulas (pHF or SF). The PP analysis set includes only the subjects in which no protocol deviation was recorded until the age of 6 months. Missing data were not imputed. For the ITT analysis the last available follow-up data were taken into account.

The incidence of CMPA and AD within the first 6 months of life were calculated and Relative Risks, 95% Confidence Interval (RR, 95% CI) were estimated separately by a Poisson General Equation Estimation (GEE) regression analysis (with a log link) applying two models. The first model adjusted for study formula and stratification factors (center, gender and family history of AD). The second model additionally adjusted for the interaction between family history of AD and study formula. A two-sided statistical significance level of p < 0.05 was accepted. The Number Needed to Treat (NNT) for AD as described by Sackett and Haynes (32) was also calculated.

Of 650 subjects eligible for participating in the study, 99 were excluded before being assigned to any group, 331 were randomized to one of the two study formula groups and 220 were exclusively breastfed and followed as an observational group. The data for the exclusively breastfed infants will be presented in detail in another publication. The whole study population and reasons for early termination (dropouts) are presented in Figure 1. Of those randomized (ITT), 160 (48%) infants were allocated to the pHF (17 exclusively formula-fed, 143 mixed-fed) and 171 (52%) to the SF group (19 exclusively formula-fed, 154 mixed-fed). The PP analysis set consisted of 105 (66%) infants in the pHF group and 120 (70%) in SF group that strictly adhered to the study protocol.

The effect of pHF on the incidence and relative risk for CMPA and AD compared to SF is presented in 3 population groups: a) all formula-fed (exclusively and mixed-fed) b) mixed-fed and c) exclusively formula-fed infants.

The baseline characteristics (including mode of delivery, birth weight, gender and pet ownership) of study participants for the pHF and SF groups are presented in Table 1. There were no significant differences in any of the characteristics between the two groups in both the ITT and PP sets, except from the residence in urban areas (ITT). In the ITT analysis set, for each participating country, the percentage of infants receiving the intervention and control formulas was almost identical (Bulgaria; 47.5 vs. 46.8%, Cyprus; 34.4 vs. 35.7%, Greece; 18.1 vs. 16.4% p = 0.73). Approximately 1 in 4 infants in each formula group had a family history of AD (pHF: 28.8% vs. SF: 25.7% p = 0.54).

The mean daily amount of infant formula (mL) consumed was similar between the two study formula groups at all follow-up visits (2^nd, 4^th, and 6^th month of age) in both the Per-Protocol and Intention-to-Treat analyses (Supplementary Table S4).

Sixteen infants in the pHF group had symptoms/signs suggestive of CMPA and were offered an OFC. Six challenges were positive and 10 negative. In the SF group, 20 infants had suggestive CMPA symptoms/signs and were offered an OFC of which 12 challenges were positive, 5 negative and in 3 were never performed because of parental refusal.

The incidence and RR for CMPA and AD within the first 6 months of life in formula-fed infants (both exclusively formula-fed and mixed-fed) in both analyses are presented in Table 2.

In ITT analysis, 6 (3.8%) subjects in the pHF vs. 12 (7.0%) in the SF group developed CMPA within the first 6 months of life (p = 0.19). Adjusting for the interaction between FHAD and study formula did not provide any statistical significance. Similar results were observed for both Models in the PP analysis.

The incidence of AD was significantly lower in the pHF compared to SF group in both analyses (ITT: 10.6 vs. 18.7% p = 0.024, PP: 11.4 vs. 24.2% p = 0.021), with RR 0.54 (0.32, 0.92) and 0.49 (0.26, 0.90) respectively. The effect was stronger when adjusting for family history of AD (Model 2) as shown in ITT analysis [6.5 vs. 27.3%, RR 0.24 (0.07, 0.78), p = 0.018] representing a risk reduction effect of 76%. A trend was observed in PP analysis [7.7 vs. 29.4%, RR 0.28 (0.07, 1.19), p = 0.085].

The incidence of CMPA within the first 6 months of life in mixed-fed subjects for both analyses is depicted in Figure 2. The incidence of CMPA in the SF compared to the pHF group was higher in both analyses (ITT: 7.9 vs. 3.5%, p = 0.11 and PP: 9.4 vs. 4.5%, p = 0.29) although the difference did not reach statistical significance. Post-hoc analysis using CMPA+_SPT and CMPA+_susp showed a trend in the ITT analysis toward a preventive effect of pHF as compared to SF [CMPA+_SPT: 4.2 vs. 9.2%, RR 0.45 (0.18, 1.12), p = 0.084, CMPA+_susp: 4.9 vs. 10.5%, RR 0.46 (0.20, 1.11), p = 0.067].

Figure 3 illustrates the incidence and RR for AD in the same group. The incidence of AD was significantly lower in the pHF compared to the SF group in both analyses (ITT: 10.5 vs. 19.7% p = 0.016, PP: 11.2 vs. 25.5% p = 0.024), with RR 0.50 (0.29, 0.88) and 0.47 (0.24, 0.91), respectively. The effect was stronger when adjusting for family history of AD (Model 2) as shown in the ITT analysis [7.3 vs. 29.3%, RR 0.24 (0.08, 0.79), p = 0.019] representing a risk reduction effect of 76%. The calculated number needed to treat (NNT) (32) was 4.5, which means that the number of infants needed to be given the pHF in order to prevent the development of AD in 1 subject was 5. In the PP analysis, in Model 2 only a trend for AD was observed [9.1 vs. 32.3%, RR 0.31 (0.08, 1.25), p = 0.10].

The results for this group of infants are presented in detail in Supplementary Table S2.

Given the small number of subjects exclusively formula-fed (pHF n = 17, SF n = 19) no differences could be detected (Supplementary Table S3).

In this multicenter double-blinded randomized controlled trial of non-exclusively breastfed high-risk for allergy infants, nutritional intervention with a partially hydrolyzed whey-based formula compared to a standard formula resulted in a significant reduction in the cumulative incidence of atopic dermatitis within the first 6 months of life. The risk reduction effect was strongest (76%) in mixed-fed infants with a family history of AD. In this group, the number of subjects needed to receive the specific pHF in order to prevent the development of 1 case of AD was 5. Although the incidence of CMPA as confirmed by OFC in the SF compared to the pHF group was two times higher (12 vs. 6 cases; 7 vs. 3.8% in the whole study population), the difference did not reach statistical significance. However, clinical significance and a potential preventive role of pHF in CMPA may not be excluded.

Breastfeeding is undoubtfully the ideal way of feeding in early life and should be strongly encouraged (33–35). However, in real life, not all infants are exclusively breastfed and a number of recent studies have shown that mixed-feeding is a commonly practiced form of nutrition (36–38). In the present study, approximately 20% of the mothers who initially expressed their willingness to exclusively breastfeed, introduced another formula as supplementary feeding at some point during the study. The fact that the majority of infants (90%) in A.R.T. who were randomized to any of the 2 study formulas had also received breastmilk during the first 6 months of life, further supports this notion and reflects the successful efforts of breastfeeding campaigns (39–41). Mixed feeding may have an additional beneficial effect in the prevention of CMPA as shown in a large prospective USA study which demonstrated that the combination of breastmilk and infant formula feeding is associated with the lowest rate of food protein-induced allergic proctocolitis, the most common CMPA clinical presentation in the first year of life (42).

Partially hydrolyzed formulas have been used as a nutritional intervention to prevent the development of allergic manifestations in non-exclusively breastfed high-risk infants (7, 43–45). This recommendation was mainly based on the findings of studies reporting a reducing effect particularly on the incidence of AD (9, 10, 28, 46). The GINI study (10, 47–50), considered as the reference study in this research area, showed a preventive role of a specific pHF on AD, up to the age of 15 years, but not at all follow-up time points. In GINI, the significant reduction effect in AD was observed in infants without a family history of AD, whereas in A.R.T. the reduction effect was strongest in those infants with a family history of AD. This may be explained by methodological differences between the two studies including that in A.R.T. formula introduction was allowed until the age of 10 weeks and AD outcome reported at 6 months of age, whereas in GINI the formula was introduced at any time point during the first 6 months of life and AD outcome reported at the age of 18 months. There is some evidence that pHF may be effective in the prevention of allergic disease even in infants without a family history of atopy (51, 52). In contrast to A.R.T. and GINI, the large Melbourne Atopy Cohort Study (MACS) (53), found no evidence to support the use of a pHF on allergy prevention in high-risk infants.

Amidst the longstanding debate (7, 54–56), a number of scientific organizations have recently withdrawn or downgraded their recommendation for pHF in allergy prevention guidelines (57–62). The switch in recommendation was mainly based on the findings of recent systematic reviews and meta-analyses which pointed out inadequate scientific evidence supporting the preventive role of pHF and raised concerns for bias on several levels in most previous studies (including selection, assessment, attrition and conflict of interest) (13, 14). However, since it appears that effectiveness in risk reduction may be dependent of specific physicochemical properties of pHF used (as recognized by EFSA), the outcomes of meta-analyses should be handled with care since these analyses may do not make such a differentiation in pHFs. In 2016, EFSA noted that the safety, suitability and efficacy of each specific formula containing protein hydrolysates has to be established by clinical trials (21). Following this recommendation, and taking into consideration limitations of previous studies, A.R.T. aimed to assess the efficacy of a specific whey-based partially hydrolyzed formula by applying rigorous methodology. This study demonstrated a clear risk reducing effect of the pHF on AD development, whereas post-hoc analysis indicates a trend towards a preventive effect against CMPA.

The most important strength of the present study is the double-blinded randomized controlled design (63). Furthermore, only high-risk infants with a documented by a physician family history of allergy could participate. Since subjects were recruited from public and private maternity hospitals/clinics in 3 different European countries providing a representative high-risk population sample, results could be generalizable. Infants had not consumed any formula with intact or partially hydrolyzed proteins prior to allocation and solid foods were allowed after the age of 4 months.

A.R.T. is one of the few studies reporting outcomes at the age of 6 months when CMPA and AD are the most common allergic manifestations (8, 9, 64). All researchers were appropriately trained prior to initiation of the study, and centers were compared during training, to get the most accurate and comparable results. Furthermore, statistical analysis was performed by an independent third party. All the data were prospectively collected and uploaded on the online Research Manager database and locked before analysis. Both the ITT and PP analyses are presented and even though the ITT results are considered more reliable (12), PP results are almost identical further supporting the ART study findings.

Importantly, CMPA was objectively confirmed by open oral food challenge after an elimination diet in accordance with current CMPA guidelines (6, 27, 45, 65–69). Although this could be considered as a strength of the study, one could see it as a limitation since the gold standard for food allergy diagnosis (double-blinded placebo-controlled food challenge) was not performed. It is generally accepted that the clinical diagnosis of AD in early life is challenging as the proposed criteria are difficult to apply (70, 71) given that they have been created for the assessment of older children. However, the AD outcome in ART was defined as a clinical diagnosis of AD by experienced clinicians supported by the objective SCORAD (23) tool and the awareness CoMiSS (24) tool.

There could be arguments that in order to assess the risk reduction effect of pHF the numbers of exclusively formula-fed infants should have been larger and that the protective effect observed in the A.R.T. study may be dependent only on the different maternal bioactive factors in breast milk (72). However, given that the number of mixed-fed infants and the amount of infant formula consumed (Supplementary Table S4) were similar between the two study arms (pHF and SF), the observed risk-reduction effect is likely due to the specific pHF used in the study. Moreover, since it is unethical to randomize breastfeeding and given that exclusive breastfeeding is strongly recommended by WHO (73), a study to enroll high numbers of exclusively formula-fed infants would have been difficult to gain approval by any bioethics committee.

Another limitation is the sample size and power estimation of the study that was based on limited data from older studies. The assumed incidence of CMPA was 20% (30) in the SF group, whereas in the actual study we found only 8%, which might suggest why the study did not detect full effectiveness of the pHF.

The number of subjects who strictly adhered to the study protocol (PP analysis) may sound relatively low [105/160 subjects in the pHF and 120/171 subjects in the SF group (drop-out rate of 34 and 30% respectively)]. However, it is close to what we have expected (estimated drop-out rate 30%) for this type of longitudinal intervention randomized controlled trial applying a demanding protocol. Interestingly, a review of 71 randomized controlled trials published in four prestigious medical journals revealed that 1 in 5 trials had drop-out rates of more than 20% (74).

The ITT analysis is considered more reliable acting as a critical safeguard against bias because it preserves the benefits of randomization which include balancing known and unknown factors and eliminating selection bias (12, 74). According to our estimations, 158 infants had to be included per treatment arm for the ITT analysis and in the A.R.T. study these numbers were achieved (160 in pHF and 171 in SF group) supporting valid results. We could assume that if we had the opportunity to recruit more subjects in each group, we would reasonably expect the observed risk-reduction effects to be even stronger in both the ITT and PP analyses.

Data on the possible mechanisms of action (i.e., immune profiling, Th1/Th2 cytokines, specific IgE measurements) by which a pHF may protect against the development of allergy would have been interesting. However, this was not within the scope of the A.R.T. study as our primary objective was to investigate the potential risk reduction effect with this specific pHF and not to explore the possible underline mechanisms of action which are amply discussed in other publications (75–77) and would be extremely important to incorporate in future studies within this research field.

Finally, true blinding for pHF is difficult due to its specific taste and smell as compared to SF (7, 12). Also working with FCA and FCB codes at the bottom of the tins was not ideal as it may lead to bias. However, products were prepared at home and FCA and FCB were nowhere explained during the study. Only a product developer at FrieslandCampina, not involved in the study, had access to decoding. Code-break or de-blinding did not take place until statistical analyses were completed.