Risk Scores for Predicting Outcomes in Patients with Type 2 Diabetes and Nephropathy: The RENAAL Study

Materials and Methods

Results

Of the 1513 patients who were randomly assigned in RENAAL, 341 patients developed ESRD. The relationship between the independent baseline risk factors and the development of ESRD was as follows: Risk score = (1.96 × log [urinary albumin:creatinine ratio (UACR; mg/g)]) − (0.78 × serum albumin [g/dl]) + (1.28 × serum creatinine [mg/dl]) − (0.11 × hemoglobin [g/dl]), with a higher value representing a higher risk (Table 1).

In the RENAAL study, investigators aimed to treat BP to the goal of <140/90 mmHg. We tested for correlation between systolic and diastolic BP and the four elements of the risk score using Pearson product-moment correlation coefficient. Patients with higher BP had larger risk scores and, in particular, more albuminuria. BP did not correlate as well with other elements of the risk score. Although systolic BP is important in determining ESRD risk particularly in patients with type 2 diabetes, in RENAAL, it only explained 12 to 18% of the change in risk score from baseline (data not shown). This was due, in large part, to the fact that the vast majority of patients were on therapy with major reductions in BP from baseline (1).

For comparison of the predictive power of the risk score and its components for ESRD, patients were categorized by the severity of the renal disease into quartiles. As shown in Table 2, the event rate per 1000 patient-years of follow-up increased with increasing albuminuria and serum creatinine and decreasing serum albumin and hemoglobin. Among the four components, baseline albuminuria was the strongest predictor of ESRD, with an event rate per 1000 patient-years of follow-up of 18.7 in the first quartile (albuminuria < 558.0 mg/g [63.1 mg/mmol]) and 227.8 at the last quartile (≥2544.5 [287.5]). Even so, the risk score had a better predictive power than albuminuria alone. For the same patients, the event rate per 1000 patient-years of follow-up was only 6.7 in the first quartile (score < 3.0) and 257.2 at the last quartile (score ≥ 5.1). Therefore, using the risk score improved predictive power for ESRD versus using albuminuria alone, especially for low-risk patients. The relative predictive power can be assessed using the HR for each covariate: The HR between the fourth and first quartiles of the risk score was 49.0, as compared with the corresponding HR for each component: 14.7 for albuminuria, 9.2 for serum creatinine, 5.5 for hemoglobin, and 10.2 for serum albumin. Therefore, the predictive power of the risk score on ESRD increases at least three-fold as compared with albuminuria alone (Table 2).

The applicability of the risk score was evaluated further by examination of the crude ESRD rates by mean risk score deciles (Figure 1). The ESRD rates rose with increased mean risk decile from the fifth to 10th deciles. The risk score was tested for optimistic bias by application of a jackknife procedure, which resulted in similar distribution of the events by mean risk score decile (data not shown). Likewise, the distribution of events by mean risk score decile showed that the risk score was predictive for ESRD in both genders and both white and nonwhite races (data not shown).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Crude ESRD rates stratified by decile of mean risk score.

Figure 2 shows the Kaplan-Meier curves for ESRD stratified by quartile of risk score. Consistent treatment effects were seen in patients across all quartiles (Figure 3). The test for treatment-by-quartile interaction was NS (P = 0.59).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Kaplan-Meier curve for ESRD end point stratified by quartile of risk score.

Quartile 1: Hazard ratio (HR) 0.02 (95% confidence interval [CI] 0.01 to 0.04), P < 0.0001.

Quartile 2: HR 0.07 (95% Cl 0.05 to 0.10), P < 0.0001.

Quartile 3: HR 0.27 (95% Cl 0.21 to 0.35), P < 0.0001.

Quartile 4: Reference.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Kaplan-Meier curve for ESRD end point stratified by quartile of risk score and displaying treatment effects.

Quartile 1: HR 0.95 (95% CI 0.24 to 3.79), P = not significant (NS).

Quartile 2: HR 0.43 (95% Cl 0.19 to 0.98), P = 0.044.

Quartile 3: HR 0.51 (95% Cl 0.33 to 0.79), P = 0.003.

Quartile 4: HR 0.81 (95% Cl 0.62 to 1.06), P = NS.

Time-varying covariate analyses showed that at 36 mo, approximately 54% of patients in fourth quartile had developed ESRD (i.e., for a hypothetical cohort of patients who remain in the fourth quartile throughout the trial, 54% would be expected to develop ESRD within 36 mo). For patients in the first through third quartiles, 1 to 10% developed ESRD at 36 mo. At month 36, approximately 58% of the patients in the placebo group versus approximately 50% of the patients in the losartan group who were in the fourth quartile of the risk score had developed ESRD, reflecting the influence of losartan treatment on slowing the progression of albuminuria and development of ESRD.

Among the four risk factors, proteinuria and serum creatinine were positively correlated with each other but negatively correlated with serum albumin and hemoglobin. As shown in Table 3, patients who had baseline proteinuria <2000 mg/g (226 mg/mmol) had overlapping quartile ranges of serum creatinine, serum albumin, and hemoglobin with patients who had baseline proteinuria ≥2000 mg/g (226 mg/mmol). Similar results were seen when patients were stratified by baseline serum creatinine <2 or ≥2 mg/dl (176.8 μmol/L). None of the risk factors could be reproduced by the other ones. For different subpopulations, similar risk score equations were derived using the four risk factors. As shown in Table 4, when patients were stratified by renal disease severity, either baseline proteinuria <2000 or ≥2000 mg/g (226 mg/mmol) or baseline serum creatinine <2 or ≥2 mg/dl (176.8 μmol/L), by gender, white or nonwhite race, and treatment group, the four parameters were almost independent risk factors for each subgroup even though the number of outcomes was reduced dramatically. The coefficients for each risk factor vary somewhat with subgroups, but the 95% confidence intervals overlap each other. In particular, the risk scores are very similar between losartan and placebo treatment groups.

Of the 1513 patients who were randomly assigned in RENAAL, 555 patients developed ESRD or died. The risk score for the composite end point of ESRD or death was as follows: Risk score (ESRD or death) = (1.14 × log [UACR]) − (0.61 × serum albumin [g/dl]) + (0.97 × serum creatinine [mg/dl]) − (0.07 × hemoglobin [g/dl]) + (0.08 × glycosylated hemoglobin [HbA_1c; %]) (Table 1). In this case, HbA_1c entered the model. The frequency table shows that the two risk scores are highly comparable in their abilities to predict outcomes (Table 5).

Discussion

In the RENAAL study of 1513 patients who had type 2 diabetes and nephropathy and in whom BP was treated aggressively, 341 patients developed ESRD. Independent risk factors for the development of ESRD were albuminuria, increased serum creatinine, hypoalbuminemia, and decreased hemoglobin level, which demonstrated that these nontraditional risk factors are important in patients with diabetic kidney disease. We present the relationship between the independent risk factors and developed a risk score to predict ESRD in patients with type 2 diabetes and kidney disease. This relationship was shown to be applicable to both genders, and it also was independent of ethnicity, initial level of BP, and treatment intervention. Patients with renal disease progression often die before reaching ESRD (5); therefore, we also examined the risk score for the composite end point of ESRD or death. It was comparably effective in predicting outcome. HbA_1c entered the model when the risk score for the composite end point of ESRD or death was calculated.

Albuminuria is a critical baseline risk predictor for ESRD in this and other studies of patients with diabetic nephropathy (6,7). Level of albuminuria is the most important clinical marker for future renal events, and UACR should be monitored in all patients with diabetic nephropathy. In RENAAL, there was a nearly linear relationship between progression to ESRD and albuminuria at baseline and on treatment (6). However, the effect of treatment with losartan on albuminuria was responsible for only 50% of the effect of losartan on ESRD (6).

The value of the RENAAL ESRD risk score is that it demonstrates that factors in addition to albuminuria predict ESRD: hypoalbuminemia, increased serum creatinine level, and decreased hemoglobin level. Although the four factors were correlated with each other, one still worked beyond the correlation and could not be fully predicted by the others. The inclusion of these factors improved the risk prediction for progression of nephropathy to ESRD from 50% with consideration of albuminuria alone to >80% when the risk score was used. Hypoalbuminemia worsens as renal function deteriorates (8) and may be related to albuminuria, inflammation, and nutrition. Low serum albumin at initiation of dialysis has been associated with increased morbidity and mortality in dialysis patients (9,10). As expected, the incidence of ESRD increased with increasing serum creatinine (e.g., 40.5% in the serum creatinine 2.1 to 3.6 mg/dl [185.6 to 318.2 μmol/L] tertile versus 7.3% in the 0.9 to 1.6 mg/dl [79.6 to 141.4 μmol/L] tertile) (11). Anemia is common in diabetic kidney disease and contributes to cardiovascular morbidity and mortality in patients with nephropathy (12,13). Anemia also was identified as an independent predictor for progression to ESRD in this study (14). Even mild anemia (<13.8 g/dl [138 g/L]) was associated with increased risk. After adjustment for known risk factors for ESRD, the average increase in relative risk was 11% for each 1 g/dl (10 g/L) decrease in hemoglobin concentration. The contribution of each factor on ESRD was robust regardless of whether patients were at low or high risk, as defined by either baseline proteinuria < or ≥2000 mg/g (226 mg/mmol) or baseline serum creatinine < or ≥2 mg/dl (176.8 μmol/L), also regardless of gender, white/nonwhite race, and treatment group. The four risk predictors for progression of kidney disease in the risk score were independent of therapy.

Control of hypertension is of primary importance in patients with diabetic nephropathy. Current guidelines state that the BP goal in patients with diabetes is ≤130/80 mmHg. Because BP was treated aggressively and equivalent levels were achieved in both treatment groups in the RENAAL study, this may have resulted in our inability to identify BP as an independent risk factor in multivariate analyses. However, it is possible—but unlikely—that the lack of correlation may have been because hypertension was not associated with risk for ESRD.

Glycemic control is important in this population, and HbA_1c entered the model only when the risk score for the composite end point of ESRD or death was calculated. This is consistent with observations of other large observational studies that demonstrated a role for glycemic control in cardiovascular disease but not kidney disease (15).

Identifying covariates that are highly associated with outcome plays an important role in clinical trial design. In the International Conference on Harmonization guidelines, investigators are advised to identify the covariates that may influence the primary outcome and to prespecify the method to be used to account for them to compensate for any imbalance between groups (16). The risk score that combines the identified important covariates helps to avoid the use of multiple covariates to adjust for the treatment effect in statistical analyses. The homogeneity of the treatment effect across severity of disease can be explored by using subgroup analysis of the risk score.

An important limitation of the risk score validation via examination of ESRD events by quartiles and deciles was that few events occurred in the lower quartiles and deciles. An additional important limitation was that this analysis used clinical trial data, and the same data from which score was derived to validate score; however, the outcome of the application of the jackknife procedure causes us to be optimistic about the broad applicability of this score. Because the risk score was derived from the Cox model using RENAAL baseline data, the absolute risk score may not be interpretable. The difference in risk score between patients suggests relative risks on the development of ESRD.

The RENAAL risk score for ESRD emphasizes the importance of identification of level of albuminuria, hypoalbuminemia, increased serum creatinine, and decreased hemoglobin level to predict the development of ESRD in patients with type 2 diabetes and nephropathy. Although albuminuria is a very strong predictor for ESRD, the contribution of serum albumin, serum creatinine, and hemoglobin level further enhances the prediction of ESRD. Future trials with a similar patient population and outcome measures as those of the RENAAL study should consider adjusting analyses for baseline risk factors.