Abstract
Research Type:
Level 3 - Retrospective cohort study, Case-control study, Meta-analysis of Level 3 studies
Introduction/Purpose:
Diabetic foot ulcers (DFUs) are a major cause of morbidity, leading to 40–60% of non-traumatic lower limb amputations worldwide. Identifying patients at risk for major amputations remains a critical challenge, as traditional clinical models often fail to capture the complex interplay of risk factors. Artificial intelligence (AI) offers an advanced approach to amputation risk prediction by leveraging machine learning and network-based analysis. This study aimed to develop AI-driven models capable of accurately predicting amputation risk and distinguishing between major and minor amputations. By integrating patient demographics, comorbidities, and laboratory data, the study sought to improve clinical decision-making, optimize treatment strategies, and prevent unnecessary major amputations, particularly in resource-limited settings where efficient allocation of healthcare resources is crucial.
Methods:
This retrospective cohort study analyzed 459 diabetic patients with PEDIS 3 or 4 DFUs who underwent surgical treatment at a private tertiary hospital in Brazil between 2008 and 2023. Patient data included demographics, comorbidities, laboratory biomarkers, and surgical outcomes. Two AI-based predictive models were developed: (1) amputation risk prediction (surgery vs. no surgery) and (2) major vs. minor amputation classification. The models utilized the Network Node Dispersion (NND) metric, a novel AI approach that quantifies the structural complexity of clinical risk factors. Performance was validated using Leave-One-Out Cross-Validation (LOOCV). Evaluation metrics included accuracy, F1-score, and Area Under the Curve (AUC) for model performance assessment. Key predictive variables were identified through a variable importance analysis, allowing AI to uncover the most significant risk factors associated with amputation severity in DFU patients.
Results:
The AI models demonstrated exceptional predictive performance. The amputation risk model achieved an accuracy of 78.5% with an AUC of 79%, effectively distinguishing patients requiring surgery from those who did not. The major vs. minor amputation model performed even better, reaching 88.8% accuracy and an AUC of 94%, significantly outperforming traditional statistical methods. Key predictive variables identified by the AI included C-reactive protein (CRP), segmented neutrophils, and lymphocyte counts. The model’s ability to capture complex interactions between these biomarkers allowed for superior risk stratification. Compared to previous studies where amputation prediction models achieved AUCs ranging from 0.6 to 0.84, our AI-driven approach demonstrated substantial improvements in predictive power, underscoring its potential for clinical integration in decision support systems.
Conclusion:
AI-driven models provide highly accurate and reliable predictions for amputation risk and severity in diabetic foot ulcer patients. The 94% AUC for major vs. minor amputation classification represents a significant advancement over conventional prediction methods. By identifying key biomarkers such as CRP, segmented neutrophils, and lymphocytes, AI enables earlier intervention, reducing unnecessary major amputations and optimizing patient outcomes. Future studies should focus on external validation and real-world implementation of AI-based decision support tools for precision medicine in diabetic foot care.
(A) Surgery required vs. No surgery required: The radar plot compares biomarker levels between patients who required surgical intervention (red) and those who did not (blue). Higher levels of C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), segmented neutrophils, and glycated hemoglobin (HbA1c) were associated with a greater need for surgery, while higher lymphocyte and albumin levels were more prevalent in patients who did not require surgery. (B) Major vs. Minor Amputation: This radar plot visualizes biomarker differences between patients undergoing major amputations (red) and minor amputations (blue). Elevated CRP, white blood cell count (WBC), and segmented neutrophils were strongly associated with major amputations, while higher levels of hematocrit, hemoglobin, lymphocytes, and albumin correlated with minor amputations. These findings highlight key predictive biomarkers for amputation severity in diabetic foot ulcers.