AI-Enabled Mobile App for Safe Eating in Older Adults With Dysphagia

Dysphagia (swallowing impairment) is a highly prevalent condition in older adults and is associated with increased risks of aspiration, malnutrition, dehydration, and reduced quality of life. While the International Dysphagia Diet Standardization Initiative (IDDSI) framework provides standardized terminology and practical methods for modifying food and liquid textures, ensuring accurate and consistent adherence to prescribed texture levels remains challenging in community settings. In particular, adherence is often limited by a lack of real-time feedback, variable caregiver skills, and difficulty verifying food texture during routine meal preparation.

This study evaluates a novel, implementation-oriented digital health intervention designed to address these barriers. The intervention is an AI-enabled mobile application that supports real-time classification of food and liquid textures using smartphone-based image capture. The application incorporates a safety-first decision logic, providing conservative classification outputs and prompts that encourage users to verify textures using established IDDSI field tests (e.g., flow test, fork pressure test) when classification uncertainty is detected. The system is designed to operate within everyday meal-preparation workflows, thereby targeting behavioral adherence at the point of consumption rather than relying solely on retrospective education. The application is intended as a decision-support tool and does not replace clinician judgment or prescribed dietary recommendations.

The trial adopts a pragmatic, community-based randomized controlled design to assess real-world effectiveness under typical home-use conditions. The intervention is delivered over a defined 12-week period of active use, during which participants and/or caregivers may engage with the application during meal preparation or consumption. The application records usage metrics (e.g., frequency of image captures, classification outputs, and uncertainty prompts), enabling evaluation of engagement and fidelity. The intervention is supported by structured onboarding and ongoing technical support to ensure usability among older adults and their caregivers. The 12-week intervention period is followed by an 8-week follow-up period, resulting in a total study period of 20 weeks per participant.

The comparator reflects current standard practice in community dysphagia management and allows evaluation of the incremental benefit of digital decision support beyond education alone. The study is designed within a superiority framework to determine whether access to the AI-enabled tool leads to meaningful improvements in behavioral adherence, which is considered the proximal mechanism linking dietary modification to downstream clinical outcomes. By focusing on adherence as the primary target, the study aligns with an implementation science perspective, recognizing that the efficacy of dietary recommendations depends on their consistent and correct application in daily life.

The primary outcome is adherence to the clinician-prescribed IDDSI dietary recommendation at Week 12. Adherence will be assessed using a structured 3-day dietary record and accompanying meal photographs collected over two weekdays and one weekend day within the assessment week. A speech-language pathologist blinded to group allocation will review the dietary records and meal photographs and determine whether each meal is consistent with the participant's prescribed IDDSI level. Participant-level adherence will be calculated as the number of adherent meals divided by the total number of meals recorded. Adherence will also be assessed at Week 20 to evaluate maintenance of any intervention effects.

In addition to assessing behavioral outcomes, the trial incorporates a multidimensional evaluation framework spanning dysphagia symptom severity, swallowing-related quality of life, dysphagia-related adverse events, and dysphagia management self-efficacy. These domains provide a comprehensive understanding of both intended benefits and potential unintended consequences, including risks related to misclassification, over-reliance on automated guidance, or increased user burden.

The study also integrates human-technology interaction considerations, including usability and perceived workload, to evaluate the feasibility and scalability of implementation. App usability will be evaluated using the mHealth App Usability Questionnaire (MAUQ), and perceived workload associated with app use will be assessed using the NASA Task Load Index (NASA-TLX) at Week 12 in the intervention arm. These measures are important for determining whether the intervention can be sustainably adopted in routine practice, particularly among older populations with varying levels of digital literacy.

From an analytical perspective, the trial is designed to estimate real-world effectiveness using an intention-to-treat framework, capturing the impact of offering the intervention under typical conditions rather than ideal adherence scenarios. Exploratory analyses will examine associations between engagement metrics and clinical outcomes.

This research addresses a critical gap in dysphagia care by evaluating a scalable, technology-assisted approach that operationalizes standardized dietary guidelines in real-world settings. It also contributes to the emerging field of AI in healthcare by providing rigorous evidence from a randomized controlled trial conducted outside of highly controlled clinical environments. The findings are expected to inform both clinical practice and public health strategies aimed at improving adherence to recommended food textures and supporting safer eating behaviors among community-dwelling older adults with dysphagia.