Infant Food Allergy Algorithm Compared With Oral Food Challenges
Key Takeaways
- In Australian infant research cohorts, conclusive SPT-positive food-allergy classifications generally matched oral food challenges.
- Conclusive classifications covered 38% of cashew challenges versus 89.7% of cow’s milk challenges in SPT-positive infants.
- Zero-mm SPT-based peanut and egg tolerance assignments almost always matched challenges.
Odoi and colleagues developed the Vitality infant food-allergy classification algorithm through expert-panel review and validated it against OFCs in original research published in Pediatric Allergy and Immunology in 2026. The diagnostic-classification comparison included 587 infants and 740 OFCs from the Australian HealthNuts and EarlyNuts population cohorts and the Vitality vitamin D randomized trial; it did not test algorithm-directed care. Inputs were caregiver-reported tolerance, caregiver-reported reaction or anaphylaxis, SPT, and OFC result. Outputs were definite allergy, probable allergy, definite tolerance, probable tolerance, and inconclusive. When a challenge was unavailable, a zero-mm SPT yielded a tolerance assignment. For SPT-positive infants, the algorithm used allergen- and age-specific SPT thresholds alongside available ingestion or reaction history; high sensitization could also produce an allergy assignment without prior ingestion. Qualifying tolerance intake had to occur on at least two occasions. Investigators derived assignments before challenge results were revealed, combined definite and probable categories, and compared assignments with positive or negative OFCs. Inconclusive challenges and assignments were excluded from accuracy calculations. HealthNuts supplied the SPT-negative peanut and egg comparison; EarlyNuts and Vitality supplied the SPT-positive four-food comparison.
Among SPT-negative infants, peanut tolerance assignments matched every challenge; one egg assignment may have been falsely negative after a late reaction. Among SPT-positive food challenges, 378 of 549 (68.9%) received conclusive classifications. Of those, 329 of 378 (87%; 95% confidence interval, 83.2–90.3) agreed with the OFC. Inconclusive assignments did not enter the agreement calculation.
Agreement among conclusive SPT-positive classifications was lowest for peanut and highest for cow’s milk, with cashew and raw egg between them. At the infant level, most SPT-positive infants received a conclusive any-allergy classification; accuracy among those classified was 87.5%. At the food-challenge level, some negative challenges were classified allergic, predominantly for peanut, while some positive challenges were classified tolerant, predominantly for raw egg. Raw-egg tolerance assignments had a low negative predictive value. No severe reactions were observed among challenged infants with positive OFCs misclassified as tolerant; this observation does not establish the safety of algorithm-guided care.
Insufficient qualifying ingestion histories particularly limited cashew classification, and caregivers could misremember intake or reactions. Reported food portions were smaller than the cumulative amount needed to establish tolerance by OFC. Soft-cooked egg histories also informed classifications tested against raw-egg challenges, despite the difference in preparation. Validation concerned four foods in Australian infant research cohorts, not other allergens or populations outside Australia. Agreement against challenge outcomes does not establish the consequences of using these classifications for clinical decisions.
According to Odoi and colleagues, the algorithm can assign some infant research outcomes when OFCs are unavailable, although agreement varies by food. Whether it performs similarly for other allergens or populations beyond Australia remains untested. A research classification does not replace an OFC.
Clinician Questions
What food intake qualified as a tolerance history in the Vitality infant food-allergy algorithm?
For infants being classified for peanut, cashew, egg, or cow’s milk, qualifying caregiver-reported intake had to occur on at least two occasions. Examples included at least two bites of peanut butter or cashew-containing food, half a soft-cooked egg, or a sip of cow’s milk. These were research-history thresholds, not proof of challenge-defined tolerance.
How did reported anaphylaxis affect food-allergy classification in infants with positive skin-prick tests?
For infants with caregiver-reported anaphylaxis to the tested food, the Vitality infant research algorithm sent a 1-mm SPT result for expert-panel review and assigned definite allergy when the SPT wheal was at least 2 mm. Infants with a history of anaphylaxis to that food did not undergo OFC under the study safety protocol, so this rule was not directly checked against challenge outcomes in those infants. This was a research-classification rule, not a clinical diagnosis.
Was a serum-specific IgE pathway validated for the Vitality infant food-allergy algorithm?
No. Odoi and colleagues proposed an alternative research-classification pathway using serum-specific immunoglobulin E but did not validate it because optional blood collection was limited in the Australian EarlyNuts and Vitality infant cohorts. The validated comparison instead used SPT and history when challenge outcomes were unavailable for classification.