Abstract: FR-PO0054
PhosTrack: Artificial Intelligence (AI)-Assisted Premeal Phosphate Binder Dosing in in Patients with CKD on Dialysis: Dietitian-Validated Accuracy
Session Information
- Artificial Intelligence: Clinical Prediction, Decision Support, and Digital Kidney Care
October 23, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Singh, Manisha, University of Arkansas for Medical Sciences, Little Rock, Arkansas, United States
- Hartwick, Audra B., Central Arkansas Veterans Healthcare System John L McClellan Memorial Veterans Hospital, Little Rock, Arkansas, United States
- Sharma, Shree G., Arkana Laboratories, Little Rock, Arkansas, United States
Background
Hyperphosphatemia in CKD/ESRD drives mineral bone disorder, vascular calcification, and mortality. Fixed binder dosing ignores variable dietary phosphorus bioavailability, poor adherence, and the pharmacokinetic requirement for pre-meal timing. We developed PhosTrack, an AI tool providing meal-specific, pre-meal phosphate binder guidance
Methods
Built by a nephrologist and nephropathologist using Claude (Anthropic) without dedicated software engineering. Patients log meals via natural language or photograph. An LLM analyzes each entry using USDA FoodData Central with source-specific bioavailability fractions: animal protein 40–80%, dairy 50–60% with subtype differentiation, plant/phytate-bound 20–40%, food additives 90–100%. Doses calibrated to agent-specific binding capacity; tenapanor included as phosphate absorption inhibitor; calcium binders flagged per KDIGO CKD-MBD. A renal dietitian calculated bioavailable phosphorus for 10 CKD meal types across 14 assessments (4 via photograph). Agreement threshold: ±20%
Results
Phase 1 (14 assessments): 12/14 within ±20%, 9/14 within ±10% of dietitian reference. The critical finding was systematic underestimation — not overestimation — in processed foods. Chain restaurant pizza mozzarella misclassified at natural dairy bioavailability (55%) vs. processed cheese blend (100%); correction raised bioavailable phosphorus from 345 to 438 mg, changing sevelamer from 1 to 2 tablets. Fast food beef patties reclassified from animal protein (65%) to injected-phosphate category (95%). Targeted reclassification incorporated; re-validation planned. Photograph assessments concordant with text-based input.
Conclusion
In this proof-of-concept validation (n=14), PhosTrack achieved accuracy within the pre-specified ±20% threshold for whole-food CKD meals. The principal finding — systematic underestimation in processed foods, the highest phosphate additive burden category for dialysis patients — was driven by bioavailability misclassification rather than random error. This failure mode is correctable: targeted reclassification has been incorporated. Prospective re-validation, patient usability testing, and a pilot correlating meal-specific phosphorus estimates with monthly serum phosphorus are planned to demonstrate that meal-specific binder guidance reduces phosphorus variability in home dialysis patients.