Background: Registry-based studies have identified risk factors for chronic kidney disease (CKD) and for progression to end-stage renal disease. However, usually, these studies do not incorporate sequential measurements of kidney function and provide little information on the prognosis of individual patients. The aim of this study is to identify which combinations of demographic and clinical characteristics are useful to discriminate patients with a differential annual decline in glomerular filtration rate (GFR).
Methods: This observational retrospective study includes patients enlisted in the registry of the Prevention of Progressive Renal Insufficiency Project of Emilia-Romagna region (Italy) from July 2004 to June 2010, with at least four serum creatinine measurements. Classification tree analysis (CTA) was used to identify subgroups of patients with a different annual GFR decline using demographic and laboratory data collected at study entry.
Results: The CTA procedure generated seven mutually exclusive groups. Among patients with proteinuria, those with a baseline estimated GFR (eGFR) of >33 mL/min/1.73 m(2) exhibited the fastest illness progression in the study population (-3.655 mL/min/1.73 m(2)), followed by patients with a baseline eGFR of <33 mL/min/1.73 m(2) and a baseline serum phosphorus of >4.3 mg/dL (-2.833 mL/min/1.73 m(2)). Among patients without proteinuria, those aged <67 years exhibited a significantly faster progression, which was even faster for the subgroup with diabetes. Among patients aged >67 years, females had on average a stable eGFR over time, with a large variability.
Conclusions: It is possible to rely on a few variables typically accessible in routine clinical practice to stratify patients with a different CKD progression rate. Stratification can be used to guide decisions about the follow-up schedule, treatments to slow progression of kidney disease, prevent its complications and to begin planning for dialysis and transplantation.
Keywords: CKD progression; decision tree; prediction models.