AI-assisted Diagnosis, Triage and Assessment of Hearing Loss and Tinnitus

BACKGROUND

Hearing loss affects approximately 11 million people in the UK, while tinnitus impacts around 7 million.1,2 Both conditions can significantly reduce quality of life and are linked to poorer mental health and employment challenges.3 Each year, tinnitus alone accounts for more than one million GP appointments, and patients referred to hospital ear, nose and throat (ENT) services often face long delays, sometimes exceeding a year, before their first assessment.4 This was exacerbated even further following the coronavirus (COVID-19) pandemic.5

To address this demand, the Royal Cornwall Hospitals NHS Trust has developed a virtual ENT clinic to manage patients with hearing loss and/or non-pulsatile tinnitus. Patients undergo a validated hearing test in person and complete online questionnaires. Clinicians then review these data to determine next steps, which may include discharge with advice, referral for imaging, or a face-to-face consultation. Inclusion criteria for the virtual clinic for hearing loss and non-pulsatile tinnitus were: ≥16 years of age; subjective bilateral hearing loss and/or tinnitus that did not fulfil the criteria for direct referral to audiology (they would see patients ≥50 years of age); subjective asymmetrical sensorineural hearing loss; and subjective unilateral non-pulsatile tinnitus. Exclusion criteria for the virtual clinic for hearing loss and non-pulsatile tinnitus were: cognitive, motor or visual impairment that may preclude use of an iPad®; patients who had been referred back to ENT having previously been fully medically assessed (redirected to audiology or hearing therapy where appropriate); and symptoms that required a face-to-face appointment (conductive hearing loss, pulsatile tinnitus, vertigo or dizziness, pain, discharge, infections, abnormal ear examination, and history of ear disease, surgery or ear injury).6

Published results demonstrated that most patients could be managed virtually, substantially reducing waiting times.6 However, clinicians must still review every case, limiting capacity for patients who require direct care. This project builds on the virtual clinic by introducing artificial intelligence (AI) to support the assessment process, as a clinical decision support system (CDSS). Using explainable AI methods, the system is trained to replicate clinician-level decision-making while providing transparent reasoning for its recommendations. This retrospective and prospective observational feasibility pilot study aims to evaluate how closely AI-generated outcomes align with clinician assessments, with all cases continuing to receive a clinician's final review. Clinicians will be blind to the recommendations produced by the AI tool, but the study aims to measure how concordant AI recommendations are with clinician assessments.

If the AI tool's clinical recommendations closely align with clinician recommendations (the gold standard for care), the AI tool could be introduced as a clinical recommendation assistant tool, streamlining the triage and management of hearing loss and tinnitus, enabling clinicians to focus on complex cases, accelerating access to care, and improving efficiency.

RATIONALE

The primary rationale for the study is to assess the concordance between AI-generated and clinician-generated outcomes within a real-world virtual hearing loss and tinnitus clinic environment. Demonstrating strong alignment would support the future integration of AI-assisted clinical decision support systems into routine clinical workflows, with the potential to improve service capacity, reduce waiting times, and allow clinicians to focus on more complex cases requiring direct specialist expertise.

In parallel with this work, the research team is undertaking a systematic review examining the use of artificial intelligence in the diagnosis, triage, and management of patients with hearing loss and tinnitus within otolaryngology settings. Although this review has not yet reached the stage of full-text screening, preliminary scoping suggests that, to date, there are no published studies directly evaluating an AI tool used to diagnose, triage, and manage hearing loss and tinnitus patients within an ENT outpatient pathway overseen by otolaryngologists. While the use of AI as a clinical decision support tool is a recognised and growing field, there remains a notable evidence gap in its application to this specific clinical context.

The study design ensures patient safety and ethical integrity by maintaining full clinician oversight for all cases. During this evaluation phase, the AI system will function solely as a research tool and will not influence clinical decision-making, i.e. a shadow implementation. This phased, comparative approach enables a rigorous assessment of feasibility, safety, and acceptability before any consideration of broader clinical implementation.

Ultimately, the findings will help determine whether AI-assisted clinical triage could offer a sustainable and scalable response to rising demand in hearing loss and tinnitus services, enhancing patient access, efficiency, and satisfaction while maintaining the highest standards of clinical care.

THEORETICAL FRAMEWORK

This study is grounded in the intersection of clinical decision-making theory, artificial intelligence (AI) in healthcare, and health systems efficiency models. Its theoretical foundation integrates concepts from evidence-based medicine, human-AI collaboration, and service redesign in healthcare delivery.

1. Clinical Decision-Making and Diagnostic Reasoning

   Clinical decision-making in otolaryngology involves interpreting complex data from hearing tests, patient histories, and symptom questionnaires to generate diagnostic and management recommendations. According to dual-process theory, clinicians use both analytical (systematic reasoning based on evidence) and non-analytical (pattern recognition and intuition) approaches.7 These cognitive processes underpin expert clinical judgement but are also limited by human variability, cognitive load, and time constraints-particularly in high-demand services within the NHS.

   By modelling clinician decision-making algorithmically, AI systems can emulate these cognitive processes, providing structured, data-driven, explainable recommendations that complement human expertise. This study therefore builds on clinical reasoning theory, assessing whether AI can achieve concordance with clinician judgement within a controlled, supervised framework.

2. Artificial Intelligence and Explainability in Medicine

   The study applies explainable AI (XAI) principles, which emphasise transparency, interpretability, and accountability in algorithmic decision-making. Traditional "black box" AI models have raised ethical concerns about opacity and trust in clinical contexts. Explainable AI provides interpretable outputs that allow researchers and clinicians to understand why an AI system reached a particular decision, supporting both clinical safety and regulatory compliance.

   This theoretical stance draws on socio-technical systems theory, which views AI not as a replacement for clinicians but as part of an "intervention ensemble approach".8 Within this framework, AI serves as a clinical decision support tool designed to augment human expertise, reduce administrative burden, and enhance diagnostic efficiency, while clinicians retain final authority over all patient care decisions.

3. Health Systems Efficiency, Virtual Care, and National AI Strategy

This study is grounded in the framework of health service efficiency and equitable access to care. Virtual clinics have already proven effective in reducing waiting times and improving access for patients with hearing loss and tinnitus.6 However, efficiency remains constrained by the requirement for clinicians to personally review every case in detail. Integrating AI-assisted triage models offers a mechanism to streamline this process, supporting clinicians by automating routine assessments while preserving human oversight. This approach aligns with lean healthcare principles, which emphasise reducing non-value-adding steps and directing clinical expertise toward complex or high-risk cases.9

The study also aligns with national healthcare policy promoting digital transformation. The UK Government's "Fit for the Future: 10-Year Health Plan for England" identifies AI and data-driven technologies as central to improving NHS capacity, efficiency, and patient experience through innovation and smarter service delivery.10 In parallel, NHS England's Artificial Intelligence Information Governance Guidance outlines the ethical and governance standards for deploying AI in clinical settings-framing AI as a decision-support tool that complements rather than replaces clinician judgement, ensuring accountability, transparency, and patient trust.11

The integration of AI in this study therefore reflects not only a local initiative to enhance virtual care, but also a broader policy-aligned commitment to digital innovation within the NHS. This framework is informed by the Technology Acceptance Model (TAM), which suggests that successful adoption of new technologies depends on perceived usefulness, ease of integration, and professional trust.12 By evaluating the concordance between AI-generated and clinician-generated outcomes-and later, patient satisfaction-the study contributes empirical evidence on the acceptability, feasibility, and safety of AI-assisted clinical workflows within the NHS.

This study is a prospective and retrospective observational mixed-methods service evaluation study, with elements of a diagnostic accuracy and implementation evaluation framework.