Revolutionizing disease management using AI

Veterinary use of artificial intelligence 

In recent years, the veterinary field has witnessed a technological revolution driven by the rapid advancement of artificial intelligence (AI). This transformation is not merely a trend, it is a fundamental shift in how veterinary professionals can and will approach diagnosis, treatment, and preventive care (1). Among the most promising applications of AI is its capacity to predict disease risk, enabling proactive interventions that can significantly improve pets’ health and quality of life. One such critical area is feline diabetes mellitus (DM), a condition with serious health implications for cats worldwide, and this short paper will show how AI may assist the clinician when dealing with this common endocrinopathy. As a supplementary illustration of AI’s abilities, much of this paper was co-written by a proprietary AI platform, as explained in Box 1. 

 

Box 1. An introduction to MAX.

 

The growing challenge of feline diabetes mellitus

Diabetes mellitus is a complex, multifactorial disease characterized by impaired insulin production or action, leading to elevated blood glucose levels (2). It shares similarities with human type 2 diabetes and has become increasingly prevalent, especially among overweight and obese cats (3,4). According to recent studies, cats with obesity are up to four times more likely to develop DM than their lean counterparts (5), and recent results have shown that the longer a cat has been overweight or obese, the greater the risk of developing DM (Figure 1) (6). The disease's insidious progression often results in severe complications – such as ketoacidosis, neuropathy, and even death – if not diagnosed and managed early (7). The challenge for veterinarians lies in the difficulty of predicting which cats are at risk before clinical signs manifest. Traditional diagnostic methods rely on observing signs or conducting blood tests after the disease has already developed (Figure 2), but this reactive approach often results in delayed diagnosis, increased treatment costs, and compromised animal welfare.

Figure 2. Regular sampling to monitor a cat’s blood glucose levels is necessary once diabetes mellitus has been diagnosed and is under insulin treatment; prevention is better than cure, and the possibility of identifying at-risk cats before the disease develops is obviously a major advantage in terms of animal health and welfare. © Shutterstock

 

The promise of AI in disease prediction

Artificial intelligence offers a transformative solution to this challenge. By analyzing vast amounts of data – ranging from medical records, lifestyle factors, genetic predispositions, and biometric measurements – AI algorithms can identify patterns and risk factors that may elude human observation. This capability enables the development of predictive models that assess an individual cat’s likelihood of developing certain diseases, using data such as:

• Medical records: Historical health data, previous diagnoses, medication history.
• Lifestyle factors: Diet, activity levels, indoor vs. outdoor living.
• Physical measurements: Body weight, body condition score (BCS), age.
• Laboratory results: Blood glucose levels, insulin sensitivity markers.
• Genetic data: Breed predispositions, hereditary factors.

Using this data, researchers and veterinary practitioners can train machine learning models to recognize the combination of factors that increase the risk of disease development, prior to the appearance of classical clinical symptoms.

Introducing DiabetesPredict

DiabetesPredict is an algorithm that can detect if a cat is at risk of becoming diabetic in the following 3 or 12 months. The algorithm can be applied at any time, and only needs data from a single visit to the clinic; this includes both clinical data and standard blood parameters (Figure 3). The algorithm was developed and validated using Banfield medical records between 2013 and 2023, employing data from 2,557,817 cats that made 10,184,040 clinic visits over this period. The algorithm was further validated using data from a secondary source, namely medical records from VCA clinics (69,422 cats with 116,306 visits between 2017 and 2023). Validation of an algorithm in an external dataset is an important step in the development of any model, as it involves testing the model on separate datasets to ensure it performs reliably across different populations.

The effectiveness of the model is measured through metrics such as accuracy, sensitivity and specificity. DiabetesPredict can achieve a 96% accuracy rate in predicting diabetes risk over a three-month period, with a sensitivity of 80%, meaning it can correctly identify 80% of cats that go on to develop DM. The models also show high specificity, which means that false positives are very low, so this should lead to few unnecessary interventions.

Figure 3. An overview of the values to input, and prediction model output, for DiabetesPredict. © John Flanagan/redrawn by Sandrine Fontègne

 

Dietary recommendations following DiabetesPredict

While no data currently exists regarding the optimal nutritional management of cats which are identified to be at-risk of developing DM, common sense would dictate that nutritional strategies similar to cats with clinical DM could be applied. In this context, a cat identified to be at risk of developing DM in the following 3 or 12 months should be recommended a therapeutic diet low in digestible carbohydrates (to minimize post-prandial glucose and insulin responses). For cats in healthy body condition, weight gain should be carefully avoided, and owners may benefit from clear feeding and weight-monitoring guidance to support healthy weight maintenance. For a cat that is overweight, the preferred diet should not only be low in digestible carbohydrates, it should also be suitable for therapeutic weight reduction (to achieve a weight reduction rate of 0.5-2% per week until a healthy target weight is achieved). A 12-week weight reduction program in overweight diabetic cats has been shown to more than double the likelihood of diabetic remission (8), and, while yet to be proven, weight reduction in cats in a “pre-diabetic” stage is also likely to be highly beneficial in preventing transition to clinical DM. Finally, for underweight cats, a diet low in digestible carbohydrate and with an elevated energy content should be favored, with the aim being to achieve weight gain.

Owners should also be made aware of potential clinical signs to monitor in their pet, and to conduct a recheck in 3-6 months to generate data to rerun the DiabetesPredict algorithm.

Practical implementation in veterinary practice

The integration of AI-based risk prediction tools into veterinary clinics can be seamless and highly beneficial (1). Tools such as DiabetesPredict can be embedded within existing electronic health record systems or accessed via dedicated platforms, providing real-time risk assessments during routine check-ups. This offers benefits for both veterinarians and pet owners:

• Early intervention: Identifying high-risk cats allows for targeted preventive measures, such as weight management and/or dietary modifications.
• Enhanced animal welfare: Early detection and management can improve a cat’s quality of life, reducing suffering and prolonging health span.
• Reputation: Employing such tools within a clinic will serve to showcase the expertise of the veterinarian, leading to increased footfall, and reinforcing client loyalty.
• Cost savings: Preventing or delaying the onset of diabetes reduces long-term treatment expenses and improves economic efficiency.

Conclusion

Currently, veterinarians cannot predict the risk of cats becoming diabetic, and there is no possibility to detect pre-diabetes in cats. Algorithms such as DiabetesPredict offers a unique opportunity for early intervention to potentially avoid transition of cats at risk of developing DM to clinical diabetes. In addition, detection of increased disease risk at an early stage can inform whether recommendations for more frequent blood testing should be made or whether to initiate early lifestyle modifications, notably via dietary solutions. More generally, AI models will in the future offer data-driven assistance to veterinarians that will help them make evidence-based decisions for many disease conditions, reducing reliance on subjective judgment.

 

 

References

1. Albergante L, O’Flynn C, De Meyer G. Artificial intelligence is beginning to create value for selected small animal veterinary applications while remaining immature for others. J. Am. Vet. Med. Assoc. 2025;263:388-394.
2. Osto M, Zini E, Reusch CE, et al. Diabetes from humans to cats. Gen. Comp. Endocrinol. 2013;182:48-53.
3. Scarlett JM, Donoghue S. Associations between body condition and disease in cats. J. Am. Vet. Med. Assoc. 1998;212:1725-1731.
4. O'Neill DG, Gostelow R, Orme C, et al. Epidemiology of diabetes mellitus among 193,435 cats attending primary-care veterinary practices in England. J. Vet. Intern. Med. 2016;30:964-972.
5. Verbrugghe A, Hesta M. Cats and carbohydrates: the carnivore fantasy? Vet. Sci. 2017;4:55.
6. Flanagan J, Kocevar G, Reinert B, et al. Association between prior obesity and development of feline diabetes mellitus: a pair-matched 1:3 case-control study. In: Proceedings, 35^th European Veterinary Internal Medicine-Companion Animals Congress 2025; Maastricht, the Netherlands.
7. Gilor C, Fleeman L. Diabetes mellitus in cats. In: Ettinger SJ, Feldman EC and Cote E (eds). Ettinger’s Textbook of Veterinary Internal Medicine. eBook: Elsevier, 2024;292.
8. Jørgensen FK, Kieler IN, Xia D, et al. Effect of 12-week, intentional caloric restriction, using a novel therapeutic diabetic diet suitable for weight reduction, on glycemic control in overweight cats with diabetes mellitus: an international prospective, randomized clinical trial. In: Proceedings, 35^th European Congress of Veterinary Internal Medicine for Companion Animals 2025; Maastricht, the Netherlands.