Precision Nutrition in Diabetic Foot Ulcers: AI-Enabled Personalized Metabolic Management Remains a Future Vision, Not Current Reality

The application of artificial intelligence to nutritional management in diabetic foot ulcer (搜索)搜索该疾病 (DFU) care represents a compelling but entirely unrealized frontier, according to a sweeping review published in Frontiers in Nutrition. Despite robust evidence linking nutritional status to DFU outcomes and mature AI architectures developed in adjacent disease areas, no AI-driven personalized nutrition intervention has been directly validated in DFU populations. The gap between technical feasibility and clinical reality remains wide, with authors cautioning that "substantial methodological and validation work is required before clinical translation can be realistically anticipated."

The Nutritional Foundation: Strong Associations, Weak Causal Evidence

The review synthesizes a growing body of evidence demonstrating that nutritional biomarkers are independently associated with DFU prognosis. The Prognostic Nutritional Index (PNI) (搜索)搜索该靶点, calculated from serum albumin and total lymphocyte count, has emerged as a particularly powerful predictor. In a retrospective study of 386 patients, Coşkun et al. found that PNI predicted amputation with excellent accuracy (AUC = 0.937, 95% CI: 0.911–0.963); at an optimal cut-off of 39.005, sensitivity was 82.7% and specificity reached 93.1%. The odds ratio for amputation was 81.8 (95% CI: 38.5–173.7) for patients with PNI values below this threshold.

The Controlling Nutritional Status (CONUT) score, which integrates serum albumin, total cholesterol, and total lymphocyte count, has similarly been validated. Shi et al., in a retrospective study of 357 DFU patients, identified moderate-to-severe malnutrition (CONUT score (搜索)搜索该靶点 5–12) as an independent risk factor for amputation (OR = 2.685, 95% CI: 1.141–6.314).

Among micronutrients, zinc查看药品详情 and vitamin D查看药品详情 have received the most attention. Nakamura et al. found that serum zinc levels were significantly lower in DFU patients compared to controls, with multivariable analysis identifying lower zinc as an independent factor associated with wound healing failure. A comprehensive meta-analysis by Tang et al., incorporating 36 studies and 11,298 individuals, demonstrated a dose–response association between vitamin D deficiency and elevated DFU risk, with odds ratios of 3.28 for levels below 25 nmol/L and 2.25 for levels below 50 nmol/L.

However, the review emphasizes a critical interpretive limitation: "after adjusting for age, serum albumin, serum creatinine, known ischemic heart disease, and initial ulcer area, baseline zinc查看药品详情 tertiles were no longer significantly associated with final healing rates, suggesting that zinc levels may be confounded by comorbidity rather than exerting a direct independent effect on wound healing."

The Evidence Gap: Very Low Certainty for Supplementation

A pivotal systematic review and meta-analysis by Donnelly et al., which synthesized 23 studies on nutritional interventions in DFU populations and employed GRADE methodology, concluded that the evidence supporting nutrient supplementation for wound healing remains of "low to very low certainty." Although meta-analyses showed that supplements significantly reduced wound depth (WMD −0.200 mm, 95% CI −0.364 to −0.035), width (WMD −0.466 mm, −0.724 to −0.208), and length (WMD −0.443 mm, −0.841 to −0.045), substantial between-study heterogeneity (I² = 56–68%) and low methodological quality undermine confidence in these findings.

The review notes that "no randomized controlled trial to date has definitively separated zinc查看药品详情 deficiency as an independent causal factor from the residual confounding inherent in observational studies," and that normalizing zinc levels through supplementation "does not necessarily reverse the wound healing trajectory if zinc deficiency is a surrogate for broader nutritional deterioration."

AI in DFU: Three Categories of Evidence, Zero Validated Interventions

The authors organize the available literature into three distinct categories. Category A encompasses direct DFU-specific prognostic studies using nutritional biomarkers—but these rely on conventional regression modeling without AI-driven dietary recommendations or adaptive interventions. Category B includes AI tools developed for DFU wound assessment, such as the ScoreDFUNet convolutional neural network, which classifies DFU images with 95.34% accuracy. While technically impressive, these tools are entirely non-nutritional.

Category C comprises transferable AI-nutrition methods from adjacent diseases, including GPT-based dietary recommendation systems for hemodialysis patients, XGBoost models integrated with continuous glucose monitoring (CGM) for type 2 diabetes, and digital twin-enabled personalized nutrition platforms. None of these methods has been validated in DFU-specific cohorts.

"Patients with DFU have metabolic demands—including high protein and micronutrient requirements for wound healing, frequent renal impairment, polypharmacy, and a high prevalence of sarcopenia—that differ substantially from the populations in which these AI tools were developed," the authors write.

A Conceptual Framework for Future Development

The review proposes a four-module architecture for a hypothetical DFU-specific AI nutrition tool: a dietary intake sensor using smartphone-based image recognition, a metabolic monitor processing CGM data, a clinical data integrator extracting electronic health record data, and a personalized recommendation engine driven by reinforcement learning.

A clinical vignette illustrates the concept: a 62-year-old male with a chronic neuropathic DFU, HbA1c of 8.9%, and moderate malnutrition (CONUT score (搜索)搜索该靶点 of 5) would be monitored via CGM and food photography. The AI model would identify that his time-in-range is limited to 45%, postprandial glucose excursions exceed 250 mg/dL, and midday protein intake averages only 15 g. The system could then generate personalized interventions—redistributing carbohydrate loads, prompting protein supplementation, and scheduling micronutrient dosing based on documented deficiencies.

The authors stress that this framework "serves as a conceptual model to guide future research and development rather than a declaration of clinical readiness."

Validation Roadmap and Methodological Concerns

Before clinical adoption, the review outlines a stepwise validation process: technical feasibility testing of food image classifiers in DFU patients, a single-arm pilot study in 30–50 patients, and ultimately a multi-center randomized controlled trial comparing AI-guided personalized nutrition to standard care with healing rates and amputation incidence as primary endpoints. "To date, none of these translational steps has been executed."

Broader concerns about AI in nutrition research are also highlighted. A 2025 systematic review by Cofre et al. found that 61.5% of AI-based dietary intake assessment studies were conducted in preclinical settings, with 58.3% exhibiting moderate risk of bias. A 2026 validation study of the SNAQ app against doubly labeled water—the gold standard for energy expenditure—yielded an intraclass correlation coefficient of 0.00, indicating "an absence of individual-level agreement."

The review concludes that while the foundational elements for AI-enabled precision nutrition in DFU management exist, "their application to patients with DFU remains conceptual." The path forward requires dedicated DFU-specific datasets, prospective nutritional phenotyping, and rigorous clinical validation before these technologies can contribute to reducing the global burden of diabetes-related amputations.