Machine-learning algorithm as a prognostic tool in non-obstructive acute-on-chronic kidney disease in the cat

Area of Science:

• Veterinary Medicine
• Nephrology
• Data Science in Animal Health

Background:

• Acute-on-chronic kidney disease (ACKD) presents a significant challenge in feline medicine.
• Accurate prediction of survival is crucial for managing ACKD cases and informing treatment decisions.

Purpose of the Study:

• To develop a predictive algorithm for short- and medium-term survival in cats diagnosed with ACKD.
• To identify key clinical and clinicopathological variables that influence ACKD prognosis.

Main Methods:

• A machine learning approach using decision tree algorithms was employed.
• Data from hospitalized cats with ACKD were analyzed, including clinical severity scores, serum creatinine at 48 hours (Crea48), age, body weight, spontaneous feeding at 48 hours (SpF48), and ultrasonographic findings.
• The algorithms were trained to predict outcomes at 7, 30, 90, and 180 days.

Main Results:

• Serum creatinine at 48 hours (Crea48) demonstrated strong predictive power for survival at 7, 30, and 90 days.
• A decision tree incorporating age and Crea48 was optimal for predicting 180-day outcomes.
• Excluding Crea48, other variables like clinical severity score, age, body weight, SpF48, and kidney appearance also contributed to prediction with similar diagnostic performance.

Conclusions:

• Crea48 is a valuable predictor for short- and medium-term survival in feline ACKD.
• Age, clinical severity score, body weight, spontaneous feeding, and kidney morphology are important secondary prognostic indicators.
• The developed algorithms offer a tool for improved prognostication in cats with ACKD.