Equivalence of Ina, an AI-based nutrition platform, to human dietitians for counseling patients with cancer.

Abstract

12117 Background: Addressing nutritional issues in patients with cancer can reduce symptoms, shorten hospitalizations, enhance treatment adherence, and improve quality of life. However, most patients never receive nutrition counseling due to dietitian workforce shortages and healthcare access disparities. Advances in artificial intelligence (AI) and high rates of mobile device utilization among all demographics provide opportunities to expand reach. This study assessed if guidance from an expert-designed AI nutrition platform, “Ina”, is equivalent to that of human Oncology-Credentialed Registered Dietitians (RD-CSO). Methods: RD-CSOs were recruited from a professional message board and grouped by quartiles of RD experience (yrs). We randomly selected 1 from each quartile as “Responders” (n = 4) and 3 with >10 yrs experience as expert “Reviewers” (n = 3). To compare Ina to RD Responders, a list of 20 top oncology nutrition queries was developed and assigned to 10 hypothetical patient profiles representing common cancer types, comorbidities, side effects, food allergies and preferences. Both Ina and RD Responders answered the queries. The Reviewers then blindly rated each answer (n = 100) using a modified version of the validated Quality Assessment of Medical AI (mQAMAI) instrument, which individually scores domains of accuracy, clarity, relevance, completeness, and usefulness yielding a total score between 5-25. Within-query differences between Ina and Responders were expressed as mean differences (SD) and tested for significance using a Signed Rank or two-tailed paired T-test. Equivalence was defined a priori as a mean within-query difference < 5. Results: The criteria for equivalence was met and no statistically significant differences were found between the total scores of Ina and each RD Responder. Ina’s average total score was superior to the combined RD Responders (19.3 vs 18.3, 95% CI [0.1,1.8]; p = 0.02). A descriptive analysis of individual mQAMAI domains demonstrated higher scores for Ina compared to averaged RD Responder scores (Table). Ina had a 54% faster response time (8.5 vs 18.5 min per 2 queries; CI [-0.21,-0.13]; p < 0.001) and 27% easier readability (Flesch-Kincaid grade level 7.2 vs 9.9; CI [-3.3,-1.9]; p < 0.001). Conclusions: Ina provides equivalent nutritional guidance to that of human dietitians for patients with cancer, with greater speed and readability. This technology offers a solution to meet the needs of patients with cancer in settings where access to dietitians is limited. Mean mQAMAI score. Total Quality Score Accuracy Clarity Relevance Completeness Usefulness AI-PlatformMean (SD) 19.3 (2.7) 3.9 (0.7) 4.2 (0.5) 4.1 (0.6) 3.3 (0.8) 3.8 (0.7) RD Responders Mean (SD) 18.3 (3.0) 3.8 (0.7) 3.9 (0.6) 3.9 (0.6) 3.2 (0.8) 3.5 (0.7) Mean Difference [95% CI] 0.9 [0.1,1.8] 0.1 [-0.1, 0.3] 0.3 [0.1, 0.5] 0.2 [0.0, 0.4] 0.1 [-0.1, 0.3] 0.2 [0.0, 0.4] P val 0.02 0.06 0.001 0.04 0.36 0.02