With approval and oversight from the Institutional Review Board at Penn State Hershey Medical Center (PSHMC) (IRB # 24958EP), a retrospective chart review was undertaken to identify children visiting either (a) a PSHMC primary care practice (controls) or (b) allergy specialty clinic for a food allergy-related concern (cases) who were also born at the institution’s medical center. Penn State Hershey Children’s Hospital has more than a million children in the referral area and offers the only Level I pediatric trauma center in more than 70,000 square miles. Similarly, the allergy clinic is a physician-referral clinic that sees complex patients; the catchment area for the allergy clinic covers more than 85,000 square miles of central and eastern Pennsylvania and accepts both private and public insurances.

To identify allergic children, the ICD-9-CM coding system was used to identify clinic visits that were coded in a manner consistent with food-related allergic reactions (Table 1). Those codes were subsequently matched to children who were born at PSHMC between July 1, 2000, and June 30, 2005. The children identified as a result of this match were considered to be “potential cases”, and their medical charts were retrieved for review. However, only children whose medical charts confirmed the presence of food allergies, documented by either a positive serum specific IgE test, (specific IgE antibodies >0.35 kU/L) or positive skin prick test (SPT) (≥ 3 mm compared with negative control), were included in our data analysis as allergic “cases”. Children were excluded from our dataset if miscoding errors were identified (e.g., latex allergies instead of food allergies) or if parent reported “food allergy” that was not confirmed by laboratory testing (e.g., parent thinks child is “allergic” to milk because s/he experiences abdominal discomfort). In an attempt to increase our number of allergic cases, a protocol modification was made near the conclusion of our chart review period that also permitted us to accept direct referrals from pediatricians familiar with our study objectives if they were aware of patients with food allergies that had not been identified via billing records (e.g., possibly missed by initial search of billing records due to ICD-9 mis-coding).

Age- and sex-matched controls for this analysis included children born at PSHMC, followed by our pediatric clinics for well-visit check-ups, but without evidence of food-mediated reactions; this cohort of children was also identified from billing records. For both control children and case children, data abstracted from birth records of the child and labor/delivery records of the mother were used to investigate perinatal factors known to impact gut flora acquisition. Data retrieved from charts included: delivery type, time spent in the neonatal intensive care unit, evidence of atopy for child and mother (including anaphylaxis, atopic dermatitis, hives, respiratory allergies), maternal group B Streptococcus status, maternal receipt of antibiotics during delivery, maternal intent to breastfeed, and age and parity of birth mother.

Chart review analyses were performed using a matched case–control design where nearly every case was matched with 2 controls. A sample size of 200 cases and 400 controls was planned. It was expected that such a sample would provide >90% power to detect differences between cohorts if the odds ratio was 1.8 or higher for variables such as intrapartum antibiotic exposure where use in the control population approximates 30%. Chi square and Fisher exact tests were used for data analysis using SAS version 9.2 (SAS Inc, Cary, NC). Descriptive statistics, including mean, median, standard deviation and quartiles for continuous measures, and frequencies for categorical measures were used to characterize the case and control populations. Logistic regression analysis was used to assess associations between variables of primary interest and food allergy. Strength of associations is expressed using odds ratios with 95% confidence intervals.

Initial billing records identified 6160 children with ICD-9-CM codes consistent with medical visits related to food allergic reactions between 2000 and 2005 (Figure 2). However, only 235 (3.8%) of those children were born at PSHMC, to permit direct examination of birth records. Of those 235 children (n=125 for males; n=110 for females), a thorough review of medical records confirmed food allergies (positive IgE or SPT) in only 82 (35%) children. In an effort to increase the number of food allergic cases, direct referrals from pediatricians were subsequently accepted; as a result, we expanded the age of children who could be considered as “cases” to 18 years of age based on these direct physician referrals. This protocol modification increased the number of food allergic children (by 17) for whom we could evaluate birth records to 99; three of these children were born in 1998 or 1999. Medical records from 192 non-food allergic children born at the same hospital were also examined as controls. Overall, the mean age of children at the time that charts were reviewed was 5.0 ± 1.5 years, with a mean age of 4.8 ± 1.9 years in the allergic cohort and 5.1 ± 1.3 years (p=0.12) in the non-allergic children.

Of the 291 children (99 cases; 192 controls) included in our analysis, 167 were male and 124 were female (Table 2). There was no difference in gestational age for children with allergies versus non-allergic children (38.3 weeks ± 2.39 weeks; range 27 to 41 weeks versus 38.0 weeks ± 2.86 weeks; range 25 to 42 weeks). There was also no difference in birth weight between the two groups (3.36 kg ± 0.67 kg; range 0.81 to 4.61 kg versus 3.24 kg ± 0.76 kg, range 0.66 to 5.07 kg).

Ethnicity data were available for 282 mothers: 87.6% of mothers self-identified as white (non-Hispanic), whereas the remaining women identified as African American (4.3%), Asian (4.6%) or Hispanic (3.5%). For 45% of the mothers, the children represented by these data were firstborn. In all, 14% of mothers reported a personal history of asthma and 4% of mothers reported food allergies. The likelihood of having a child diagnosed with food allergies was slightly higher for women with a history of atopy compared to women that did not report atopic conditions (43% versus 32%), but this did not reach significance (OR, 1.60; 95% CI, 0.861-2.996; p=0.14). Maternal parity was not associated with an increased incidence of food allergies in offspring, neither did mean maternal age at time of delivery differ significantly between mothers whose children developed food allergies (30.68 ±5.06 years) compared to those that did not (28.55 ± 6.48 years). However, an association was identified between the odds of food allergy diagnosis in childhood and maternal age at delivery (OR =1.05; 95% CI, 1.017 to 1.105; p=0.005). For each advancing year of maternal age at time of delivery, the odds of food allergies being diagnosed in the child increased by 6.0%. This relationship was linear for mothers whose age ranged from under 20 years through 35 years of age (Figure 3).

In the cohort of 99 children with allergies, the mean age at the time of allergy diagnosis was 1.5 years of age and males were significantly more likely to be diagnosed with a food allergy than females (OR = 1.80; 95% CI, 1.088-2.985; p=0.02). Allergies to peanuts, eggs, milk, and other nuts were the most common, impacting 53%, 49%, 43%, and 18% of the food allergic children, respectively. Forty-five percent of children with food allergies were allergic to just one food item, with the remaining children allergic to two or more foods. When these children were exposed to allergenic food(s), they most often experienced skin eruptions, including dermatitis and hives (69%). Anaphylaxis was uncommon (n=10; 5.2%).

No differences were observed with respect to development of food allergies and method of delivery (allergy diagnosis in 34% delivered vaginally and 33% delivered via cesarean; OR=0.93; 95% CI, 0.557-1.564; p=0.79), postnatal antibiotics (40% who developed allergies received antibiotics versus 33% who did not receive antimicrobials; OR=1.35; 95% CI, 0.680-2.677; p=0.39), intrapartum maternal antibiotic exposure (allergy diagnosis in 34% of children whose mothers did or did not receive intrapartum antibiotics; OR=1.00; 95% CI, 0.587-1.715; p=0.998), or time spent in the NICU (allergy diagnosis in 34% and 35% of children who did and did not immediately move to the well child nursery; OR=1.04; 95% CI, 0.544-1.977; p=0.91). Vaginal group B Streptococcus (GBS) status was unknown in half the women; for those mothers in whom GBS status was known to be positive, an increased likelihood of developing food allergies was not observed (OR=1.14; 95% CI, 0.556-2.348; p=0.72). Most women (79%) expressed intent to breastfeed their infants; however, intent to breastfeed did not vary between mothers with a positive or negative personal history of atopy (84% versus 78%; p=0.21). A positive correlation between maternal intent to breastfeed and subsequent food allergy development was observed (p<0.005) (Table 2).

Previously, associations between disrupted gut flora and atopic dermatitis have been identified 11 12 35 36 37. Furthermore, physiologic evidence links atopic dermatitis with food allergies 38. Therefore, we wished to determine if an association exists between factors that disrupt perinatal gut flora acquisition and subsequent food allergy diagnosis.

We hypothesized that specific factors known to cause gastrointestinal dysbiosis in newborns -- namely perinatal antibiotic exposure, cesarean section delivery, and NICU admission -- are associated with subsequent food hyper-responsiveness. However, we did not find such correlations to be present in children with confirmed evidence of IgE-mediated food allergies.

In our cohort, we did find, however, that increasing maternal age at delivery is associated with food allergy diagnosis in children. A similar relationship was recently reported by Metsala and colleagues 39. Although the biologic mechanisms responsible for this association are unclear, it is possible that disrupted maternal normal flora may be involved. It is known that changes in normal flora occur across the lifespan from medications or changes in gastrointestinal tract function. Specifically, with advancing age, levels of gut bifidobacteria and lactobacilli decrease, whereas clostridia and yeast increase 40. These same changes in gut flora have been previously noted in children with atopy 2. Thus, while it has not yet been studied, there may be a critical time point during adult life in which mothers may begin to develop disruptions in gut flora, which may then be transferred to children during the delivery process.

We also observed that the number of allergic males in our cohort significantly exceeded the number of allergic females. We are not alone in identifying these gender disparities in atopic sensitization of children. Males have previously been found to have food allergies and other atopic diseases with a higher prevalence than females during childhood and early adolescence 41 42 43 44 45 46. Certainly, there may be genetic causes underlying this predisposition, but environmental factors may also play a role. It has been suggested that gender differences in atopic development may reflect a tendency for mothers to breastfeed male infants for a shorter period of time due to the misconception that male infants require a greater level of nutrition than can be provided by breast milk alone 47 48 49 50. Breastfeeding has been reported to provide protection against atopic dermatitis and/or food allergies 51 52 53 54. In addition to nutritive and immunological benefits, breast milk is also known to stimulate growth of bifidobacterial populations. Thus, it is conceivable that breastfeeding duration may play a role in gut flora acquisition and subsequent atopic development. In support of this theory, there is evidence that male infants are indeed more likely than female infants to experience benefits of early probiotic interventions 55.

In our study, we found that maternal breastfeeding intent at time of delivery was associated with increased risk of food allergies. Given the retrospective nature of our data collection and since we did not contact mothers directly, we do not know the extent to which maternal intent to breastfeed at time of delivery translated into initiation/duration of breastfeeding. With 79% of mothers in our cohort indicating initial intent to breastfeed, the observed correlation between breastfeeding and food allergies in our data may be attributed to reverse causation, as has been reported previously, since breastfeeding intent and exclusive breastfeeding after hospital discharge may differ considerably 56 57 58 59. Given the limitations imposed by retrospective data collection regarding breast-feeding decision and duration, we cannot draw definitive conclusions about its role in food-allergy development in our cohort.

Although we did not find an association between food allergies and type of delivery, maternal intrapartum receipt of antibiotics, or NICU placement, these relationships warrant further study. Technical limitations were imposed by our dependence upon the ICD-9-CM coding system to identify children with food allergies, as this system is an imperfect surrogate for allergy diagnosis. As a result, we were not able to identify the number of food allergic children that we had initially set out to include in our dataset. When initially established, ICD-9-CM codes were created for insurance reimbursement purposes, not for research purposes. Furthermore, there may be tendencies, on the part of clinicians, to assign specific ICD-9-CM codes to a particular group of signs and symptoms based on reimbursement levels. This may explain why we initially identified 235 children with food allergy-diagnosis-codes born at our institution, but whose diagnosis was only substantiated (by IgE or SPT) in 85 of them. Indeed, we observed that food allergy specific ICD-9-CM codes were used for parent-reported food intolerances, as well as allergies to non-food items (e.g., latex allergy). Recently, Clark et al. also observed that exclusive reliance upon food-allergy-specific ICD-9-CM codes to identify patients with food allergies, would have led to a ~50% discrepancy from the true number of patients with food hypersensitivities 60.

However, a particular strength of our data set is the rigor with which we identified children as “cases” only if an IgE-mediated food allergic reaction had been confirmed 61 62. By focusing exclusively on children with IgE-mediated food allergies, we may have failed to identify children with food allergic manifestations that are mediated by non-IgE mechanisms 63 64. Rather than relying upon retrospective review of medical and billing records, future investigations into the role of perinatal and immunological factors impacting food allergies should be initiated prospectively at birth.