Back to Home

Validation & Science

How we measure accuracy & ensure patient safety

Validation Methodology

Nefra's nutrient estimation pipeline is validated against reference laboratory analyses using the following protocol:

  1. Sample collection — representative food items from each supported market are photographed under controlled and real-world conditions.
  2. Reference analysis — identical portions are sent to an accredited laboratory for wet-chemistry nutrient determination.
  3. Comparison — Nefra estimates are compared to lab values. We report Mean Absolute Error (MAE), sensitivity (true positive rate for exceeding daily limits), and specificity.
  4. Continuous monitoring — production predictions are sampled monthly and re-validated against updated reference databases.

Clinical validation study is in planning. The targets below represent our acceptance criteria; measured results will be published here once the pilot study is complete.

Accuracy Targets

NutrientMAE TargetSensitivitySpecificityStatus
Potassium (K)≤ 30 mg≥ 90%≥ 85%Pilot planned
Phosphorus (P)≤ 25 mg≥ 88%≥ 85%Pilot planned
Sodium (Na)≤ 50 mg≥ 90%≥ 87%Pilot planned
Protein≤ 3 g≥ 85%≥ 82%Pilot planned
Fluid≤ 50 ml≥ 85%≥ 80%Pilot planned
Calories≤ 40 kcal≥ 88%≥ 85%Pilot planned

Food Database Sources

SourceRegionEntries
USDA FoodData CentralUnited States~380,000
McCance & Widdowson'sUnited Kingdom~3,500
IFCT (NIN)India~1,500
Standard Tables of Food (MEXT)Japan~2,500
NutriSurvey DOST-FNRIPhilippines~1,600

Uncertainty & Human-in-the-Loop

When Nefra's AI confidence falls below 75%, the estimated uncertainty exceeds 15%, or the food is classified as high-risk for CKD patients (e.g., banana, coconut water, dal, miso), the app:

  • Displays a prominent warning with the confidence interval for K, P, and Na
  • Presents three options: select from database, enter manually, or accept with caution
  • Logs the uncertain prediction to ai_uncertainty_log for post-market surveillance

This human-in-the-loop system ensures that no critical dietary decision is made on unreliable AI output alone.

References

  • KDOQI Clinical Practice Guidelines — National Kidney Foundation. Nutrition in CKD: 2020 Update. Am J Kidney Dis. 2020;76(3)(Suppl 1):S1-S107.
  • KDIGO 2024 Guidelines — Kidney Disease: Improving Global Outcomes (KDIGO) CKD Work Group. Kidney Int Suppl. 2024.
  • IEC 62304:2006+A1:2015 — Medical device software — Software life cycle processes.
  • EU MDR 2017/745 — Regulation on medical devices, Annex I General Safety and Performance Requirements.

Nefra is a decision-support tool providing estimated nutritional data.

It is not a substitute for professional medical advice. Always consult your nephrologist or renal dietitian.