Early detection of feline chronic kidney disease (CKD) can be challenging, as even the most reliable markers of kidney dysfunction can be influenced by extrarenal factors. Patient history, physical examination, and laboratory diagnostics (eg, BUN, creatinine, urine specific gravity [USG], packed cell volume, electrolytes) are typically used to determine whether kidneys are functioning properly, but evaluation of laboratory results only provides (at best) an indication of kidney dysfunction at 75% nephron loss.1 Newer diagnostics to determine risk factors or biological markers for renal dysfunction are therefore of utmost importance.
This study* used machine learning to attempt to predict whether enrolled cats would develop CKD within 12 months. The model recognized subtle combinations of laboratory tests (eg, BUN, creatinine, USG) that serve as early markers of CKD risk in cats ≥7 years of age. Two strategies to determine a cutoff between cats with high and low risk for developing CKD were considered. The first strategy maximized both sensitivity (87%) and specificity (70%) and appeared most appropriate for scenarios in which correctly identifying cats that will not develop CKD within 12 months is considered more important than correctly identifying cats that will develop CKD (ie, high negative predictive value). The second strategy maximized specificity (98%) but had lower sensitivity (42%). Because this strategy has a higher positive predictive value (87%), it is more appropriate for attempting to limit false-positive results. Clinicians should be aware of the sensitivity and specificity of the exact strategy being used when applying it to clinical practice.