A hybrid web-based clinical decision support system called NutrIA, trained on harmonized NHANES data spanning 1988–2018, achieved ROC-AUC scores of 0.894, 0.914, and 0.923 for 5-, 10-, and 20-year all-cause mortality prediction respectively. The platform integrates a 151-item adaptive questionnaire, 39 validated clinical instruments, 17 clinical phenotypes, and 31 dietary clustering modules to generate individualized nutritional recommendations and automated clinical reports — translating probabilistic risk into actionable preventive guidance.
These AUC values are competitive with established cardiovascular risk tools like the Framingham Risk Score and PCE calculator, though NutrIA's scope extends further into dietary phenotyping and lifestyle intervention. The integration of machine learning with rule-based clinical reasoning addresses a persistent gap: most existing ML health tools predict risk without translating findings into interpretable, actionable recommendations — a limitation that has slowed clinical adoption broadly.
However, critical caveats apply. This is an internal validation only, meaning the model was tested on data drawn from the same population used for training — a setup prone to optimistic performance estimates. NHANES is observational and US-centric, limiting generalizability globally. No prospective clinical outcomes or real-world patient benefit have been demonstrated. The authors themselves acknowledge external validation is essential before implementation.
As a preprint posted on medRxiv and not yet peer-reviewed, these results should be treated as preliminary. The architecture is technically promising — incremental rather than paradigm-shifting — but meaningful clinical utility awaits rigorous independent validation.
