In 2002, the National Kidney Foundation’s Kidney Disease Outcomes Quality Initiative (NKF’s KDOQI) has introduced a staging system for chronic kidney disease (CKD) [1]. Though widely used for both epidemiological and clinical purposes, this 5-stage system has been criticized for being based primarily on glomerular filtration rate (GFR), as calculated by the use of equations which may not provide accurate estimates, especially in elderly and female individuals [2]. This may lead to erroneous categorization of otherwise healthy individuals with consequent overestimation of CKD prevalence and inappropriate referral to specialist [3]. In addition, the NKF’s KDOQI classification does not entirely meet the fundamental requirements for staging systems, which should reflect the natural history and particularly the prognosis of the disease.

That this system is often inadequate to describe the sequence of events occurring in CKD has become increasingly evident in view of the high prevalence of nonalbuminuric renal impairment, i.e. estimated GFR (eGFR) <60 ml/min/1.73 m² without albuminuria, detected in samples from the general population [4, 5] and also in patients with type 1 diabetes from the Diabetes Control and Complications Trial (DCCT)/Epidemiology of Diabetes Interventions and Complications study [6] and, to a higher extent, in subjects with type 2 diabetes from the United Kingdom Prospective Diabetes Study [7]. This latter finding has been confirmed in large cross-sectional analyses [5, 8] and in the baseline evaluation of two intervention trials [9, 10], thus implying that the majority of individuals with type 2 diabetes develop CKD from stage 3 of the NFK’s KDOQI classification. Therefore, the natural history of CKD in diabetes is now recognized to follow two different pathways, albuminuric and nonalbuminuric [11].

More importantly, the NKF’s KDOQI classification was originally designed to assign people with more severe prognoses to more advanced stages in order to select for referral only individuals at high risk for adverse outcomes [1]. However, for definition of stages 3-5, this classification does not incorporate information about the presence and severity of albuminuria, which is a powerful predictor of renal and cardiovascular disease (CVD) outcomes, independently of eGFR, both in non-diabetic and diabetic individuals [12–16]. The Kidney Disease: Improving Global Outcomes (KDIGO) in a Controversies Conference held in London in 2009 [17] recommended to retain the NKF’s KDOQI definition of CKD and to revise classification by including cause and albuminuria category in addition to GFR category, with subdivision of GFR category 3 at 45 ml/min/1.73 m² (CGA classification). Recently, an alternate classification system based on both eGFR and albuminuria has been proposed and validated by the Alberta Kidney Disease Network (AKDN) in terms of adverse renal outcomes and, to a lesser extent, all-cause mortality. In this classification, eGFR and albuminuria categories with similar relative risks are grouped into risk categories 1-4 (plus risk category 0 corresponding to no CKD of the NKF’s KDOQI system) [18]. In the accompanying editorial, Levey et al. did not recommend substituting risk categories for eGFR and albuminuria categories or relying on risk categories alone to predict clinical outcomes [19]. However, in 2012, the KDIGO has also grouped eGFR and albuminuria categories into 3 risk categories (moderate, high and very high risk, plus low risk category also corresponding to no CKD of the NKF’s KDOQI system) to be used for guiding decisions and predicting outcomes [20]. So far, it has not been investigated which CKD classification system stratifies more accurately diabetic individuals by prevalence of CVD and other complications.

This study was aimed at assessing whether reclassification of subjects with type 2 diabetes from the Renal Insufficiency and Cardiovascular Events (RIACE) Italian Multicenter Study using the AKDN alternate system or the new KDIGO classification provides a better definition of burden from CVD and diabetic retinopathy (DR) than the NKF’s KDOQI classification.

The clinical characteristics of subjects according to NKF’s KDOQI stages and AKDN or KDIGO risk categories are presented in Additional file 1: Tables S1-S3. Across all classification systems, age, age at diabetes diagnosis, diabetes duration, HbA_1c, percentage of insulin users, triglycerides, non-HDL cholesterol, systolic BP, prevalence of dyslipidemia, hypertension, lipid-lowering therapy, and anti-hypertensive treatment (including usage of blockers of the renin-angiotensin system), BMI, and waist circumference tended to increase, whereas HDL cholesterol and, except in the AKDN system, percentage of current smokers decreased. Men predominated in NKF’s KDOQI stages 1-2, whereas women prevailed in stages 3-5, due to the higher percentage of females with nonalbuminuric renal impairment, the most frequent phenotype in subjects with eGFR < 60 ml/min/1.73 m²[5, 8]. Conversely, male gender predominated in all risk categories of the AKDN and KDIGO classifications.

Irrespective of the classification system, 62.5% of patients had no CKD and 37.5% had various degrees of renal impairment, as previously detailed [8]: 18.8% had albuminuria alone, and 18.7% had reduced eGFR, either with (43.4%) or without (56.6%) albuminuria. However, while the percentages of subjects with stages 1-2 and stages 3-5 of the NKF’s KDOQI classification were almost identical, reclassification of nonalbuminuric patients assigned to stages 3a (n = 1,253) and 3b (n = 351) and microalbuminuric patients assigned to stage 2 (n = 1,653) and 3a (n = 562) into risk category 1 or moderate and risk category 2 or high with the AKDN and KDIGO systems, respectively, resulted in a decreasing number of subjects with increasing risk category, in keeping with the goal of reducing referral for specialist care. Moreover, since macroalbuminuric subjects with normal-to-high or mildly reduced eGFR (n = 120 and n = 244, respectively; NKF’s KDOQI stages 1-2) are classified into a higher risk category with the AKDN alternate system (i.e. risk category 3) than with the KDIGO classification (i.e. high), the number of patients decreased less strikingly from AKDN risk category 2 to 3 than from the KDIGO risk category high to very high (Figure 2).

Though prevalence of complications increased with increasing CKD severity with all systems (P for trend <0.0001), it differed significantly between NKF’s KDOQI stages and AKDN or KDIGO risk categories (Figures 2 and 3). In fact, since prevalence of CVD, but not DR, was higher in patients with nonalbuminuric renal impairment (NKF’s KDOQI stages 3-5) than in micro and macroalbuminuric subjects with normal-to-high or mildly reduced eGFR (NKF’s KDOQI stages 1-2) (Table 1) [8, 25], prevalence of any CVD, any coronary events, and myocardial infarction was significantly higher in risk category 1 or moderate than in stage 1 (P at least <0.001) as well as in risk category 2 (P < 0.0001) or high (P at least <0.005) than in stage 2, whereas prevalence of advanced DR was higher in risk category 3 or very high than in stage 3 (P < 0.0001). Moreover, since CVD prevalence was relatively low in macroalbuminuric subjects with non-reduced eGFR (included in AKDN and KDIGO risk category 3 and high, respectively) [8], it tended to plateau between AKDN risk categories 2 and 3, whereas they increased markedly from KDIGO risk category high to very high (Figures 2 and 3). However, a higher number of subjects without CVD (2,788) or DR (2,808) were appropriately classified in the lowest risk categories of the new systems (i.e. 1 and moderate, respectively) than in NKF’s KDOQI stage 1 (820 and 757, respectively). Logistic regression analysis showed that the strength of association of complications, independent of confounders, increased more progressively with AKDN and particularly KDIGO risk categories than with NKF’s KDOQI stages, except for cerebrovascular and peripheral events. In addition, NKF’s KDOQI stage 1 was not significantly associated with any CVD event and CVD events by vascular bed, except ulcer/gangrene, whereas stage 2 did not correlate with any CVD and coronary events. Finally, the odd ratios for CVD, except cerebrovascular events, tended to plateau between AKDN risk categories 3 and 4 or even 2 and 4 (Figure 4).

In this cross-sectional analysis of the RIACE study, we compared three CKD classification systems regarding their ability to define CVD and DR burden. As reported in the general population [18, 20, 28], the percentage of patients assigned to more advanced stages or risk categories decreased markedly when individuals staged with the NKF’ KDOQI classification (18.75% assigned to stage ≥3) were reclassified with the AKDN (7.34% assigned to risk category ≥3) and particularly the KDIGO (5.03% assigned to risk category very high) systems. This implies that, according to the rationale of the new classifications [18, 20], a lower number of subjects with type 2 diabetes would be referred for specialist care, with significant cost savings. It remains to be clarified whether reclassification also improves accuracy of referral in terms of better prediction of morbidity and mortality in diabetic individuals.

Previous studies in the general population using the AKDN alternate system have shown that it better predicts adverse renal outcomes and all-cause mortality [18], whereas the NKF’ KDOQI classification is superior for identifying CKD complications such as anemia, acidosis, hyperphosphatemia, hyperparathyroidism and hypertension (but not hypoalbuminemia) [28]. However, reclassification with the alternate system was less accurate for mortality than for the renal outcome, since it consistently and incorrectly reclassified more patients who died to a lower category [18], whereas it more likely assigned to lower stages patients without each CKD complication [28], as compared the NKF’ KDOQI classification.

Our study focused on CVD and DR, the main complications of diabetes which affect more frequently diabetic patients with CKD than those without [25, 29], and profoundly impact on life expectancy and quality of these individuals [30, 31].

Results showed that prevalence of complications differed significantly between NKF’s KDOQI stages and AKDN or KDIGO risk categories and also among eGFR and albuminuria categories grouped into the same risk category. As expected, a large number of subjects without CVD and DR were correctly reclassified in the lowest risk category of the new systems. Moreover, the AKDN and particularly the KDIGO classification showed a more graded association with CVD and DR than the NKF’s KDOQI staging system, independently of confounders, thus supporting the concept that the new systems provide a better definition of CVD and DR burden in type 2 diabetes, consistent with previous data on renal outcomes and all-cause mortality from the general population [18]. Under the assumption that subjects assigned to the lowest risk category would not be referred for specialist care, at variance with those classified as AKDN risk categories ≥2 or KDIGO risk categories ≥ high, this would result in more accurate and less expensive management of diabetic individuals with CKD.

However, CVD, but not DR prevalence was higher in risk category 1 or moderate than in stage 1 (except ulcer/gangrene) as well as in risk category 2 or high than in stage 2, consistent with previous data comparing the AKDN alternate system with the NKF’s KDOQI classification in the NHANES cohort [28]. In addition, prevalence differed also among eGFR and albuminuria categories grouped into AKDN and KDIGO risk category 1 and moderate, respectively, and to a lesser extent into higher risk categories of the new systems. Both findings are attributable to the relatively high prevalence of CVD in individuals with nonalbuminuric renal impairment, higher than that in subjects with albuminuria and normal-to-high or mildly reduced eGFR [8], especially for coronary events [26]. This might imply that assignment of individuals with type 2 diabetes and normoalbuminuria who fall into eGFR category 3a and 3b to risk categories 1 (or moderate) and 2 (or high), respectively, should be tested in large prospective studies and eventually reconsidered with classification of these subjects into higher risk categories. Hopefully, the follow-up of the RIACE Study will provide insight into this issue.

Strengths of this study include the large size of the cohort, the completeness of data and the analysis of a contemporary dataset. The main limitation is the cross-sectional design of the study. Potential limitations concerning assessment of CKD and DR, including non-centralized measurements of albuminuria and serum creatinine and the use of funduscopy have been addressed in previous RIACE reports [8, 21, 25]. Briefly, as an external quality control of urinary albumin assays, 50 samples from each center were re-analyzed at the reference laboratory of the Coordinating center, confirming that the coefficient of variations between the peripheral and the central values were <15%, at least in the relevant clinical range of 15-500 mg/L [8, 21]. Moreover, fundus was examined by an expert an ophthalmologist in each center, who was asked to fill in a standardized report format for classifying the RIACE participants [25].

In conclusions, the new AKDN and KDIGO systems seem to perform better than the NKF’s KDOQI classification in grading CVD and DR burden in subjects with type 2 diabetes. However, though these systems reclassify upward a large proportion of patients without complications, thus reducing unnecessary referrals, they appear to underestimate the burden from CVD, but not DR in diabetic subjects assigned to lower risk categories, especially those with reduced eGFR without albuminuria (see Additional file 1: Table S4). This suggests that the new systems should be tested for CVD and DR outcomes in prospective studies, in order to improve risk stratification of diabetic individuals with CKD.