Lower-limb amputations remain one of the most devastating complications of diabetes, yet many are preventable with timely screening and intervention. Through a collaboration with GEMINI, the Vector Institute, and Unity Health Toronto, Diabetes Action Canada researchers helped develop and validate a new artificial intelligence (AI) model designed to identify people at high risk for diabetic foot complications.
The goal is simple but critical: help clinicians intervene earlier and prevent avoidable amputations.
People living with diabetes face more than a 30% lifetime risk of developing foot complications, and even after a wound heals there is roughly a 40% chance of recurrence within one year. Evidence shows that routine foot screening significantly reduces amputation rates, but clinicians need better tools to identify which patients require the most urgent follow-up.
Most modern AI prediction models operate as “black boxes,” generating risk scores without explaining how they reached their conclusions. This lack of transparency can make it difficult for clinicians to trust the predictions or use them to guide conversations with patients.
To address this challenge, the research team developed an interpretable AI model designed specifically for clinical use. Rather than simply generating a risk score, the model shows how individual factors, such as age, kidney function, or blood glucose levels, contribute to a patient’s overall risk. This transparency allows clinicians to understand the reasoning behind each prediction and integrate it more confidently into care decisions.
The model was trained using health data from 107,836 adults with diabetes discharged from 29 hospitals across Ontario between 2016 and 2023. Unlike many traditional prediction tools, it also accounts for competing health events, such as in-hospital death, that may occur before a foot complication develops.
When evaluated, the model demonstrated competitive or superior accuracy compared with existing black-box AI approaches, while remaining fully transparent and auditable.
Most importantly, the tool can be used at the point of hospital discharge to flag patients who may need urgent follow-up foot screening in the community. This creates an opportunity to prioritize care for those at highest risk and strengthens provincial efforts focused on lower-limb amputation prevention.
The model’s analytical engine — Competing Risks Survival Prediction using Neural Additive Models: Fine Gray (CRISPNAM-FG) — was recently presented in a spotlight session at International Association for Safe and Ethical AI (IASEAI) 2026, an international conference on trustworthy AI held at UNESCO in Paris.
This work represents another step toward using responsible artificial intelligence to support clinicians, improve care pathways, and prevent avoidable complications for people living with diabetes.
Read the full study:
Explore how an interpretable deep survival model was developed to predict post-discharge diabetic foot complications using data from hospitals across Ontario here: https://arxiv.org/pdf/2511.12409
