Allergic diseases are among the most prevalent chronic health conditions globally, affecting an estimated 20–30% of the population [1]. The World Allergy Organization (WAO) reports a rising trend in the prevalence of allergic diseases, including asthma, allergic rhinitis, food allergy, eczema, and drug allergies, particularly in industrialised nations and urbanising low- and middle-income countries [1, 2]. The burden of these conditions extends beyond morbidity and mortality in asthma, contributing to significant reductions in quality of life, school and work absenteeism, and sleep disturbance.

From a health systems perspective, allergic diseases account for substantial economic costs [1, 3]. Allergic diseases represent a major public health issue in the UK, with nearly 40% of children and 30% of adults diagnosed with at least one atopic condition such as asthma, eczema, and allergic rhinitis [3]. Asthma, often co-occurring with other atopic diseases, is responsible for the majority of emergency admissions, and food allergies are associated with growing healthcare utilisation for anaphylaxis and specialist referrals [4].

Clinical allergy practice in the UK is hampered by a critical shortage of specialist allergists and inadequate allergy training for general practitioners, which potentially increases underdiagnosis and misdiagnosis and inconsistent patient care [5]. Access to expert services is highly unequal, creating a postcode lottery for patients and exacerbating health inequities—especially in underserved areas where poor health outcomes are further compounded by social determinants such as inadequate housing. As allergy prevalence and severity continue to rise [6], particularly among children, the growing gap between the availability of specialist paediatric allergists and the increasing number of cases has rendered current workforce and training provisions increasingly unsustainable [5]. The absence of national coordination and underinvestment in allergy services have further compounded these issues, leading to preventable health risks, inefficiency, and unnecessary National Health Service (NHS) costs [5]. The growing NHS budget deficit has further compounded the issue, leading to longer waiting times and a rise in preventable, worsening health outcomes. Urgent, coordinated national action is needed to address these systemic challenges.

Artificial intelligence (AI), including machine learning (ML), natural language processing (NLP), deep learning (DL), and related technologies, has emerged as a potentially transformative tool in addressing these challenges in healthcare, where there have been traditional barriers to implementing comprehensive health care screening programmes [7, 8]. In allergy and immunology, AI is increasingly being used to predict oral food challenge (OFC) outcomes, analyse large-scale electronic health record (EHR) data, personalise immunotherapy protocols, and automate clinical documentation [9–12]. There is also growing potential for AI to support asynchronous patient communication and tailored advice, improving access to specialist input and streamlining follow-up care [11]. These tools offer potential pathways to enhance efficiency, reduce diagnostic delays, improve patient safety, and enable proactive disease monitoring [10, 11].

However, the integration of AI into clinical allergy care remains limited. Most published AI medical devices are proof-of-concept or retrospective validations, with few tools successfully implemented in real-world settings [13]. Key barriers include the limited availability of high-quality, diverse datasets (particularly paediatric-specific data), hindering the robust development, validation, and generalisability of AI tools in allergy and immunology. The “black box” nature of many AI models raises concerns about transparency, interpretability, and explainability, leading to clinician hesitation [13]. Ethical and regulatory issues, such as patient privacy, data security, and algorithmic bias, are significant, compounded by the lack of clear governance frameworks [13, 14]. There are also concerns about how AI-enabled medical tools may influence clinician behaviour, including the risk of algorithmic aversion, highlighting the need for thoughtful integration that supports, rather than undermines, clinical judgment [13, 14].

This review aims to provide a comprehensive assessment of current AI medical devices in clinical allergy practice, including outpatient workflows, diagnostics, immunotherapy, and allergy research. We highlight the opportunities and limitations of current tools and highlight areas for future development and innovation. By synthesising evidence from multiple domains, we provide insights into how AI can support a more personalised, efficient, and sustainable approach to allergy care.

This section outlines the applications of AI in clinical allergy practice, with Figure 1 providing a summary of the key themes.

AI is reshaping diagnostic strategies in allergy by enabling early recognition, enhanced phenotyping, and non-invasive evaluation, all of which contribute to improved safety and efficiency in clinical decision-making (Figure 2) [28].

Despite its promise, the integration of AI into allergy care faces substantial technical, ethical, and regulatory issues (Figure 3). A primary limitation is the lack of interoperability between AI tools and EHR systems, many of which lack standardised data structures or application programming interfaces (APIs) to support seamless exchange. This is compounded by the prevalence of bias in training datasets, often drawn from homogeneous or incomplete populations, where inequity in data can only perpetuate bias and disparity, particularly for underrepresented demographic groups [13].

For instance, models trained primarily on European paediatric cohorts may underperform in predicting food allergy risk in Asian or African populations, where sensitisation patterns differ. Similarly, asthma phenotyping algorithms built on adult datasets often fail to capture unique paediatric endotypes, leading to reduced accuracy and potential misclassification.

The opaque, “black box” nature of many DL models can undermine clinical trust, particularly in high-stakes scenarios like predicting anaphylaxis, where a lack of explainability may perpetuate mistrust and contribute to algorithmic aversion. While interpretability tools like SHapley Additive exPlanation (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) offer partial insights, their regulatory acceptance remains limited [39].

Few AI models in allergy medicine have been prospectively validated in real-world settings, raising questions about generalisability and robustness. Implementation is further constrained by infrastructural and financial limitations, particularly in resource-poor settings. Effective deployment also requires clinician training, governance structures, and organisational change management, elements often overlooked.

However, one prospective study of an AI tool predicting peanut OFC outcomes showed reliable performance in a clinical setting with paediatric patients [9]. Similarly, CDSS platforms for asthma care integrated into primary care practices improved adherence to guideline-based treatment, although uptake by clinicians was variable [17]. These examples illustrate both the potential promise and the challenges of moving from proof-of-concept to real-world application.

There are significant ethical considerations at the core of deploying AI in healthcare. AI systems rely on access to large volumes of sensitive health data, raising concerns around consent, secondary data use, and privacy. While de-identification techniques, such as pseudonymisation and federated learning, help reduce data privacy risks, robust governance frameworks and transparent consent processes remain essential [14, 40].

Accountability and explainability are paramount. Clinicians must understand how AI generates recommendations, particularly for high-impact decisions like initiating immunotherapy or food challenges. Without transparency, clinical uptake will falter, and medicolegal liability may rise. Currently, there is a lack of clarity on whether liability rests with developers, institutions, or clinicians.

Bias and fairness continue to demand scrutiny. Unrepresentative training data can lead to inequitable outcomes, worsening existing health disparities. Regulatory agencies such as the FDA, MHRA, and NICE stress the need for subgroup performance benchmarks, diverse training datasets, and continuous post-deployment market surveillance [13, 14].

Foundational frameworks like the GDPR and NHS Digital guidance provide a starting point, while WHO’s ethical AI guidelines offer global direction [41–43]. However, allergy-specific policies and standards are notably absent. For example, the UK’s NHS Digital guidance requires explicit patient consent for secondary data use, while the FDA’s Good Machine Learning Practice framework mandates documentation of algorithmic performance across diverse patient groups [44]. In allergy practice, this could mean requiring subgroup analyses (e.g., paediatric vs. adult, different ethnic groups) before deployment.

To ensure ethical, safe, and effective AI adoption in allergy practice, equity, transparency, and accountability must be prioritised. This includes patient and public involvement in algorithm development, the establishment of ethical oversight boards, and continuous real-world performance monitoring [42]. Speciality-led advocacy is essential to develop tailored guidelines that safeguard equity and trust while enabling responsible innovation.

Several emerging areas are set to define the next phase of AI integration in allergy care. Foremost is the advancement of explainable AI (XAI), which will be crucial not only for fostering clinician trust but also for empowering patients and supporting broader adoption. Interpretability tools such as SHAP and LIME can provide transparent, clinician-friendly explanations of model outputs, facilitating their use in real-world decision-making and clinical workflows [39, 45].

Federated learning also holds significant promise, enabling collaborative AI model development across multiple healthcare institutions without compromising patient privacy [40]. This is particularly relevant in allergy research, where multi-centre data pooling is essential yet often constrained by privacy regulations. Moreover, federated learning can support the development of personalised digital twins by leveraging diverse, distributed datasets while maintaining data security.

The incorporation of multi-omics data, genomic, proteomic, and metabolomic profiles, alongside environmental exposure records, will enable more accurate allergy risk prediction and further individualise treatment strategies [10]. Simultaneously, AI-driven augmented and virtual reality technologies are poised to transform education and training by simulating high-risk scenarios such as anaphylaxis or complex immunotherapy protocols, offering immersive, hands-on learning experiences [46].

At the population level, AI will support allergy surveillance by analysing environmental data (e.g., pollen, pollution) and identifying emerging sensitisation trends. This capability will allow public health authorities to anticipate seasonal surges and deploy targeted interventions efficiently.

Collectively, these developments signal a shift toward more predictive, proactive, and precision-driven allergy care powered by AI innovation.

AI is reshaping clinical allergy practice, from diagnostics and documentation to personalised treatment and public health surveillance. Despite ongoing technical, ethical, and operational challenges, the potential for transformative impact is clear. Realising this potential will require multidisciplinary collaboration, robust governance, continuous evaluation, and inclusive design. With thoughtful implementation, the allergy community stands on the brink of a digital revolution that could redefine how allergic diseases are understood, treated, and prevented.