1 INTRODUCTION

Food allergy (FA) is one of the most common chronic conditions in children.¹ The prevalence, incidence, persistence, and severity of FA are on the rise, as recently suggested by epidemiologic studies.^2-4 This epidemiologic pattern is associated with an increase in the economic burden related to the management of pediatric FA, presently estimated to be € 3.820/year/child.⁵ The diagnostic procedures contribute substantially to the financial and psychological burden of pediatric FA, and in many cases, the FA diagnosis remains presumptive or delayed.⁵ The overdiagnosis of FA is a substantial concern, as evidenced by an up to 15-fold disparity between self-reported and challenge-verified prevalence rates in the pediatric age.⁶ Overdiagnosis increases the need for tertiary center access, with a negative impact on the waiting lists. The waiting lists are one of the main criticisms of the National Healthcare System in Italy, as they dramatically compromise accessibility and availability of healthcare services.⁷ Finally, diagnostic delay remains a significant challenge in pediatric FA, particularly for non-IgE-mediated FA, in which it is estimated to be up to 6 months. This delay leads to additional psychological and economic burdens for both patients and the healthcare system.^{8, 9} The available scores or questionnaires for the diagnosis of pediatric FA are mainly focused only on cow's milk allergy or on IgE-mediated FA and mostly require the use of allergy tests.^10-14 Therefore, tools for facilitating the diagnostic approach for pediatric FA without excessive reliance on testing are urgently needed.

The goal of the Naples Pediatric Food Allergy (NAPFA) project was to develop a standardized scoring system that incorporates anamnestic and clinical data, as well as allergy test results, to facilitate the diagnostic process for children with suspected FA.

2 METHODS

2.4 Study procedures

The study procedures are depicted in Figure 1.

During the initial visit, the multidisciplinary team operating at the Center performed a complete anamnestic and clinical evaluation to assess the eligibility of subjects, collected the written consent from the parents of each subject, and performed FA screening tests, (skin prick tests, SPT; atopy patch tests, APT; and/or measurement of food-specific serum IgE, sIgE levels). At the end of the initial visit, certified dietitians provided the parents with written instructions for a four-week elimination diet, which was tailored to medical history. During the second visit, children who did not respond to the elimination diet were deemed non-allergic and underwent a comprehensive diagnostic work-up. In patients who were responsive to the elimination diet and had complete disappearance of FA-related signs and symptoms, a diagnostic oral food challenge (OFC) was planned. In patients with suspected multiple FA, the OFC was planned with one food at a time. In patients with anaphylaxis or food protein-induced enterocolitis syndrome (FPIES) induced by the ingestion of a single food, the OFC test was not performed, as suggested by other authors.^{16, 17} The OFC was performed at the Hospital within 7 days from the second visit. Children with OFC-proven FA were categorized as having IgE or non-IgE FA on the basis of their clinical features.^{16, 18, 19} Children who were negative at OFC underwent an extensive diagnostic work-up, and alternative diagnoses were obtained.

3 STUDY AIM

The aim of the study was to develop and internally validate a clinical scoring system for facilitating the diagnostic approach in pediatric patients with suspected FA.

4 DEVELOPMENT AND INTERNAL VALIDATION OF MULTIVARIABLE REGRESSION MODELS

5 STATISTICAL ANALYSIS

Most continuous variables were not Gaussian-distributed, and all are reported as median (50th percentile) and interquartile interval (25th and 75th percentiles). Discrete variables are reported as the number and percentage of subjects with the characteristic of interest. We used logistic regression to develop a multivariable regression model (M1) using FA as outcome and the following predictors: (1) sex, (2) age at the onset of signs and symptoms possibly related to FA, (3) cesarean delivery, (4) occurrence of atopic dermatitis before the onset of signs and symptoms possibly related to FA, (5) first degree family member with allergy, (6) symptoms occurrence after ingestion of specific food, (7) skin symptoms, (8) gastrointestinal symptoms, (9) respiratory symptoms, and (10) systemic symptoms. Because the predictor “symptoms occurrence after ingestion of specific” food had two levels, the number of effective predictors was 11. A further multivariable logistic regression model (M2) was developed, replacing the symptoms after food ingestion with SPT/APT/sIgE results as the two predictors were collinear, resulting in a total of 10 effective predictors. Predictors were kept in the model regardless of whether they were statistically significant.³⁰ Overall fit was evaluated using the scaled Brier score, that is, the Brier score scaled by its maximum score (Briermax) according to the equation (1 − Brier score)/Briermax, with a higher score representing greater accuracy.³¹ Discrimination, that is, the ability to separate subjects with disease from those without disease, was evaluated using Harrell's C-statistic which, for the case of logistic regression, equals the area under the receiver-operating characteristic curve.³² Calibration, that is, the agreement between observed and predicted risk, was assessed by evaluating: (1) “mean calibration” or “calibration-in-the-large” (CITL), by comparing the observed event rate with the average predicted risk; (2) “weak calibration”, by performing a logistic analysis testing whether the calibration slope is 1; and (3) “moderate calibration”, by using a “calibration plot” to test whether the predicted risks correspond to the observed event rates. Such a graph plots the predicted (expected) outcome probabilities (x-axis) against the observed outcome frequencies (y-axis). We used a locally weighted scatterplot estimator with 95% CI to assess how well the model prediction lies around the 45-degree line of the calibration plot.¹⁵ All models were internally validated by calculating the scaled Brier score, C-statistic, CITL, and calibration slope on 1000 bootstrap samples with replacement^{29, 32-34} and drawing a calibration plot with 95% confidence intervals. The linearity of the logit of age in all models was evaluated using multivariable fractional polynomials with bootstrap evaluation of stability.^{35, 36} Age was found to be linear in all models and was modeled as such. Collinearity among predictors was assessed by evaluating the condition matrix³⁷ and by using Spearman's rho.³⁸ Collinearity was detected between SPT/APT/sIgE results and the appearance of symptoms after the ingestion of specific food so they were not used together in the same model as already reported above (Section 5). Besides giving the regression equations of the models, we developed a nomogram to simplify their use in clinical practice.³⁹ Decision curve analysis (DCA) was also employed to evaluate clinical utility, which refers to the implications of model adoption in clinical practice.⁴⁰ Statistical analysis was performed using Stata 18.5 (Stata Corporation, College Station, TX, US) using the mfpboot,³⁶ bsvalidation,⁴¹ pmcalplot,⁴² pmsampsize,⁴³ nomolog,⁴⁴ and dca⁴⁵ community-contributed commands.

6 RESULTS

6.2 Development, internal validation, and decision curve analysis of the multivariable models

As described in detail under statistical analysis section, two models were developed that included or did not include APT, SPT, and sIgE. Table 2 reports such models, labeled M1 and M2, and their associated metrics of overall fit, calibration, and discrimination as determined by bootstrap on 1000 samples without replacement. The regression equations of such models are given in Appendix 1.

TABLE 2. Multivariable logistic regression models.

	Model M1 (without allergy tests)	Model M2 (with allergy tests)
Sex	0.255 [−0.200, 0.710]	0.012 [−0.663, 0.686]
Age at onset (months)	−0.053 [−0.069, -0.037]	−0.066 [−0.093, -0.039]
Cesarean delivery	0.421 [−0.038, 0.880]	0.938 [0.233, 1.642]
AD before FA	0.482 [−0.022, 0.986]	0.490 [−0.261, 1.241]
First degree family member with allergy	0.372 [−0.111, 0.854]	0.910 [0.184, 1.637]
Symptoms after specific food – 1 time	1.885 [1.260, 2.511]	–
Symptoms after specific food (≥2 times)	3.482 [2.830, 4.133]	–
Skin symptoms	0.875 [0.240, 1.509]	1.238 [0.375, 2.101]
Gastrointestinal symptoms	1.183 [0.556, 1.811]	1.999 [1.072, 2.927]
Respiratory symptoms	2.296 [1.366, 3.225]	1.699 [0.235, 3.164]
Systemic symptoms	1.454 [0.227, 2.682]	2.515 [0.912, 4.118]
Positivity of SPT/APT/sIgE	–	5.998 [4.867, 7.128]
Intercept	−3.205 [−3.434, -2.975]	−4.315 [−4.653, -3.978]
Brier scaled	52.5	79.0
C-statistic	0.915 [0.895, 0.937]	0.977 [0.969, 0.988]
Expected:Observed (E:O) ratio	0.995 [0.925, 1.050]	0.992 [0.922, 1.032]
Calibration in the large (CITL)	0.006 [−0.43, 0.260]	0.016 [−0.361, 0.415]
Calibration slope	0.941 [0.798, 1.081]	0.916 [0.715, 1.120]
N	627	627

• Note: Values are logistic regression coefficients with 95% confidence intervals in square brackets for variables from sex to intercept. 95% confidence intervals are given in brackets for measures of discrimination and calibration.
• Abbreviations: AD before FA, Occurrence of atopic dermatitis before the onset of the signs and symptoms possibly related to food allergy; Age at the onset, age at onset of the signs and symptoms possibly related to food allergy; APT, atopy patch test; sIgE, food-specific serum IgE; SPT, skin prick test; Symptoms after food ingestion, Symptoms occurrence after ingestion of a specific food 1 time; Symptoms occurrence after ingestion of particular food ≥2 times.

The discrimination made by M1 (optimism-corrected c-statistic = 0.915, 95% CI 0.895–0.937) and M2 (optimism-corrected c-statistic = 0.977, 95% CI 0.969–0.992) models was good and larger than the hypothesized one (0.810, see Section 5).

Figure 2 (left quadrants) gives the calibration plots of M1 and M2. At logistic calibration, the average calibration slope was 0.941 for M1 and 0.916 for M2, showing a satisfactory weak calibration. The examination of calibration plots showed an acceptable profile of moderate calibration (Figure 2). We also developed two nomograms corresponding to M1 and M2 (Figures 3 and 4).

Tables S1 and S2 provide the sensitivity, specificity, positive likelihood ratio (LR+), negative likelihood ratio (LR-), positive predictive value, and negative predictive value for each 10% increment in the probability estimated by Models M1 and M2. LR+ and LR- can be useful to estimate post-test probability from pre-test probability. Based on LR+ and LR-, thresholds of 10% and 90% have excellent ability in ruling out and ruling in allergy.

7 DISCUSSION

We have developed a scoring system, the NAPFA score, that may help the diagnostic approach in children with suspected FA by providing the probability to be affected by these conditions.

The NAPFA score consists of two multivariable models that can be applied based on the availability of FA screening tests. Its feasibility even without allergy test results allows its application in various healthcare settings, with the potential to reduce overdiagnosis, waiting lists, and associated economic burdens.

Overdiagnosis is a major challenge in pediatric FA that is driven by the reliance on tertiary centers for the confirmation of FA diagnosis, the risks associated with OFC procedures, the psychological stress, and the associated financial costs.⁵ A systematic review conducted in Europe revealed a significant disparity between self-reported (17.3%) and OFC-verified (0.9%) FA,⁶ as also observed in our study, where, at the end of the diagnostic process, FA was confirmed in about half of the subjects visiting the center for suspected FA. In addition, diagnostic delay remains a significant problem in pediatric FA. It has a negative impact of disease natural course, and it leads to additional psychological and economic burdens for both families, patients, and healthcare systems.8,9

A tool for facilitating the FA diagnostic approach is urgently needed. Ideally, this diagnostic tool should be able to efficiently address the major limitations that could negatively impact the initial diagnostic approach: definition of the main anamnestic and clinical data that could raise the suspicion of FA; the needs of screening tests to identify the culprit food and of OFC to confirm the diagnosis.

As far as anamnestic data are concerned, we have identified the main variables that have been associated with the occurrence of FA: first-degree family member with allergies, male sex, born by cesarean delivery, presence of AD before the FA symptoms onset, and young age.^{1, 16-19, 22-25, 46} Despite the acknowledged significance of these data, the questionnaires available in the literature primarily focus on clinical symptoms.^10-13 Notably, Galvin et al.¹⁰ were the only authors that developed a model to predict OFC, incorporating sex and age as anamnestic variables. The diagnostic accuracy of Galvin's model was demonstrated by identifying 97% of cases as true positives and 94% as true negatives. However, it was developed specifically for children with IgE-mediated FA, including only hen's egg, peanuts, or cow's milk allergy. Furthermore, the FA diagnosis was not confirmed by OFC in all study subjects.¹⁰ The identification of predictors that can be readily gathered in every healthcare setting, such as sex, delivery method, age at onset of signs and symptoms potentially associated with FA, occurrence of AD prior to the onset of signs and symptoms potentially associated with FA, and family allergy risk, makes the NAPFA score an easily and readily accessible tool, but it needs to be verified with further study.

Clinical characteristics are the primary factors determining the suspicion of FA. Clinical features were considered in the Galvin et al. model, which included skin, respiratory, gastrointestinal, and cardiovascular symptoms, but also by the last EAACI guidelines for IgE-mediated FA; the Exposure, Allergen, Timing, Environment, Reproducible Symptoms (EATERS) method; the COMISS score, and by a Portuguese tool for children with suspected adverse food reactions.^10-14 To avoid possible errors in evaluating the clinical features, in the NAPFA score we included the signs and symptoms of both IgE- and non-IgE mediated FA, together with the “occurrence of symptoms after ingestion of specific food” as predictors.²⁵ The most recent EAACI guidelines on IgE-mediated FA highlight the relevance of an allergy-focused history, introducing the concept of a possible OFC-free FA diagnosis based on the presence of a suggestive clinical history for FA together with SPT or specific sIgE positivity.¹¹ The key questions for an allergy-focused history, as outlined in the EAACI guidelines, are undoubtedly useful in obtaining a focused history for IgE-mediated FA, and our findings align well with this approach, enhancing the most significant clinical features in children affected by IgE and by non-IgE-mediated FA.¹¹ In the EAACI guidelines, the time interval between food consumption and clinical symptoms occurrence was considered, as it plays a pivotal role in IgE-mediated FA, but not in delayed reactions such as non-IgE-mediated FA.²⁵ Consequently, it was unnecessary to include the timing of symptoms occurrence as a predictor of the NAPFA score, which has been designed for both IgE-mediated and non-IgE-mediated FA. In 2018, a research group developed the EATERS questionnaire. The authors asserted that the presence of several elements of an EATERS history should prompt clinicians to consider the possibility of FA, but the lack of a standardized method for interpreting the questionnaire hindered the comparison with the NAPFA score. In addition, EATERS is not validated yet.¹⁴ In contrast to the COMISS score, which was initially designed as an awareness tool for cow's milk allergy-related symptoms¹² and recently proposed as a diagnostic tool for cow's milk allergy,⁴⁷ the NAPFA score was developed to facilitate the diagnostic approach for children potentially affected by FA caused by any type of food antigens. Furthermore, the NAPFA score used logistic regression to assess the relative contribution of each item while in the COMISS score all items are considered to be equally relevant giving a score ranging from 0 to 6. Consequently, a direct comparison of NAPFA and COMISS is not possible.¹²

Considering the necessity of allergy screening tests to identify the culprit foods, it is important to consider that in certain healthcare settings, FA screening tests may not be readily available. Consequently, without a standardized allergy-focused medical history, it can be challenging to raise the suspicion of FA, particularly for non-IgE mediated FA. Notably, the NAPFA score demonstrated remarkable accuracy in diagnosing both IgE- and non-IgE mediated FA, irrespective of the availability of screening tests. Other available models, such as the one proposed by Galvin et al.,¹⁰ necessitate at least one allergy test between SPT and IgE, rendering it not applicable to non-IgE mediated FA. This aligns with the EAACI guidelines for IgE-mediated food allergies.¹¹ It appears that the EATERS method stands out as the most effective approach in identifying individuals with IgE- and non-IgE mediated FA without the assistance of FA screening tests. However, as previously mentioned, this method requires standardization and validation.¹⁴ Furthermore, the COMISS score does not necessitate allergy tests for its applications, but as previously stated, this tool serves as an awareness instrument for cow's milk allergy related symptoms rather than a diagnostic tool. Moreover, its applicability is limited to cow's milk allergy related symptoms, unlike the NAPFA score, which is applicable to a broader range of FA.^{12, 47}

Finally, while a positive OFC remains the gold standard for the diagnosis of FA; in some contexts OFC, it is not mandatory for FA diagnosis, as in the case of FPIES, anaphylaxis, or for typical IgE-mediated FA.^{11, 16} For the latter case, the combination of a positive allergy test with a suggestive clinical history could be sufficient to perform a diagnosis of IgE-mediated FA, as stated in the most recent EAACI guidelines.¹¹ However, OFC remains essential for the diagnosis of all other FA types.

The NAPFA score demonstrated satisfactory discrimination and calibration and exhibited clinical utility at DCA, potentially facilitating the diagnostic work-up for all types of FA. However, external validation is necessary to assess its role in clinical practice. As calibration is concerned, the internal validation of our models, performed with bootstrap,³⁴ revealed a mean calibration slope of 0.941 (95%CI 0.798 to 1.081) for Model M1 and 0.916 (95%CI 0.715 to 1.120) for Model 2. Consequently, higher predicted probabilities tend to overestimate the risk of FA, while lower predicted probabilities tend to underestimate it. The external validation of the proposed prediction algorithms will allow a potential benefit recalibration, which is a better benchmark of validity as compared to internal validation.³⁴

NAPFA may facilitate the early identification of FA children in primary care settings, emergency departments, and tertiary care facilities. It has the potential to effectively improve the circular continuum between these healthcare figures, as recently reported in the Italian Diagnostic Therapeutic Care Pathway (DTCP) for the management of pediatric FA.⁴⁸ The enhancement of this pathway could have a substantial impact on the FA diagnostic process, reducing diagnostic delays, errors, and FA-related costs.

The NAPFA score's strengths include prospective study design, rigorously determined sample size, well-defined predictors, and applicability to IgE and non-IgE-mediated FA caused by any type of food antigens. Limitations include age restriction to <14 years, ethnicity because the inclusion of Caucasian subjects from Southern Europe only, and the need for external validation. As the latter point is concerned, we plan to perform a multicenter validation study involving several Italian centers in collaboration with the Italian Society of Pediatric Allergy and Immunology (SIAIP) and the Italian Society of Pediatric Gastroenterology, Hepatology, and Nutrition (SIGENP).

8 CONCLUSION

The NAPFA score is the first scoring system to incorporate anamnestic and clinical features that facilitate the diagnostic approach of pediatric FA. It is an easily accessible score with good discrimination, calibration, and clinical utility, making it suitable for widespread use by healthcare professionals. Notably, its accuracy and feasibility even without allergy test results enable its application in various healthcare settings, with the potential to reduce healthcare costs and wait times, once externally validated. To facilitate and to speed-up the external validation of the NAPFA score, also across different populations and settings, we are developing a web app that will enhance the model's accessibility and implementation in clinical practice by easily running NAPFA on computers and mobile devices of interested practitioners.

AUTHOR CONTRIBUTIONS

Laura Carucci: Conceptualization; writing – original draft; investigation; methodology; data curation; visualization. Lorenza Biancardi: Investigation; writing – original draft; writing – review and editing. Rita Nocerino: Investigation; writing – original draft; writing – review and editing; resources. Letizia Ciliberti: Investigation; writing – original draft; writing – review and editing. Erika Caldaria: Investigation; writing – original draft; writing – review and editing. Giorgio Bedogni: Formal analysis; software; validation; writing – original draft; data curation; resources. Francesco Palmese: Formal analysis; software; writing – original draft; resources. Francesco Calabrò: Formal analysis; software; writing – original draft; resources; data curation; writing – review and editing. Roberto Berni Canani: Methodology; writing – review and editing; writing – original draft; conceptualization; project administration; supervision; data curation; visualization; investigation.

FUNDING INFORMATION

This research was funded under the National Recovery and Resilience Plan (NRRP), Mission 4 Component 2 Investment 1.3—Call for tender No. 341 of 15 March 2022 of Italian Ministry of University and Research funded by the European Union—NextGenerationEU. Project code PE00000003, Concession Decree No. 1550 of 11 October 2022, adopted by the Italian Ministry of University and Research, CUP E63C22002030007, Project title “ON Foods—Research and innovation network on food and nutrition Sustainability, Safety and Security—Working ON Foods” and from the Italian Ministry of Health-Health Operational Plan Trajectory 5-Line of action “Creation of an action program for the fight against malnutrition in all its forms and for the dissemination of the principles of the Mediterranean diet” (Mediterranean Diet for Human Health Lab, “MeDiHealthLab”, code T5-AN-07, CUP E63C22002570006).

CONFLICT OF INTEREST STATEMENT

The authors have no conflict of interest to declare.