Risk Prediction for Early CKD in Type 2 Diabetes (original) (raw)

Abstract

Background and objectives

Quantitative data for prediction of incidence and progression of early CKD are scarce in individuals with type 2 diabetes. Therefore, two risk prediction models were developed for incidence and progression of CKD after 5.5 years and the relative effect of predictors were ascertained.

Design, setting, participants, & measurements

Baseline and prospective follow-up data of two randomized clinical trials, ONgoing Telmisartan Alone and in combination with Ramipril Global Endpoint Trial (ONTARGET) and Outcome Reduction with Initial Glargine Intervention (ORIGIN), were used as development and independent validation cohorts, respectively. Individuals aged ≥55 years with type 2 diabetes and normo- or microalbuminuria at baseline were included. Incidence or progression of CKD after 5.5 years was defined as new micro- or macroalbuminuria, doubling of creatinine, or ESRD. The competing risk of death was considered as an additional outcome state in the multinomial logistic models.

Results

Of the 6766 ONTARGET participants with diabetes, 1079 (15.9%) experienced incidence or progression of CKD, and 1032 (15.3%) died. The well calibrated, parsimonious laboratory prediction model incorporating only baseline albuminuria, eGFR, sex, and age exhibited an externally validated c-statistic of 0.68 and an _R_2 value of 10.6%. Albuminuria, modeled to depict the difference between baseline urinary albumin/creatinine ratio and the threshold for micro- or macroalbuminuria, was mostly responsible for the predictive performance. Inclusion of clinical predictors, such as glucose control, diabetes duration, number of prescribed antihypertensive drugs, previous vascular events, or vascular comorbidities, increased the externally validated c-statistic and _R_2 value only to 0.69 and 12.1%, respectively. Explained variation was largely driven by renal and not clinical predictors.

Conclusions

Albuminuria and eGFR were the most important factors to predict onset and progression of early CKD in individuals with type 2 diabetes. However, their predictive ability is modest. Inclusion of demographic, clinical, and other laboratory predictors barely improved predictive performance.

Keywords: proteinuria, diabetes mellitus, glomerular filtration rate

Introduction

CKD is a significant health care problem: 13% of the adult population of the United States have reduced kidney function or albuminuria (1). Especially, in people with diabetes, predicting the course of CKD is important because decline in kidney function and response to therapy vary considerably between individuals (2). Because unselective screening for kidney disease is not cost-effective (3,4), stratification of people at risk for CKD or progression of CKD is urgently needed in analogy to the highly successful risk calculators in the area of cardiovascular medicine (5). In people with diabetes, for example, screening for established CKD with measurements of albuminuria and eGFR has been shown to be cost-effective, with estimated costs of $18,650 (2009) per quality-adjusted life-year gained (6). Because primary prevention strategies are likely to be most cost-effective (6), it is also desirable to predict incident CKD in people with diabetes before microalbuminuria develops and CKD progression in those with microalbuminuria.

However, no externally validated risk calculator exists for prediction of incidence and progression of early CKD in people with diabetes. Therefore, we used data of the ONgoing Telmisartan Alone and in combination with Ramipril Global Endpoint Trial (ONTARGET) to assess the risk of incidence or progression of CKD after 5.5 years for people aged ≥55 years with diabetes and normo- or microalbuminuria (7–9). The competing risk of death was considered as an additional outcome state by using multinomial logistic regression. The Outcome Reduction with Initial Glargine Intervention (ORIGIN) cohort was used for external validation, and the relative importance of predictors was quantified (10–13).

Materials and Methods

Study Population

Development Cohort.

ONTARGET included individuals aged ≥55 years with vascular disease or diabetes with end-organ damage. Exclusion criteria included serum creatinine >3 mg/dl, clinically significant renal artery stenosis, uncontrolled hypertension, and heart failure. Participants were randomly assigned to receive telmisartan, ramipril, or both. This analysis included participants with type 2 diabetes, normo- or microalbuminuria at baseline, and complete information on incidence and progression of CKD (_n_=6766) (Figure 1).

Figure 1.

Derivation of development (ONTARGET) and validation (ORIGIN) cohorts. The left and right panels show derivation of development and validation cohorts, respectively. ONTARGET, ONgoing Telmisartan Alone and in Combination with Ramipril Global Endpoint Trial; ORIGIN, Outcome Reduction with Initial Glargine Intervention; UACR, urinary albumin/creatinine ratio.

Validation Cohort.

The ORIGIN study was chosen because participants exhibited impaired fasting glucose or early diabetes and were comparable with ONTARGET participants (11). Information starting from the 2-year visit to study end (7 years) was used, resulting in follow-up comparable with ONTARGET (Supplemental Figure 1). Participants with type 2 diabetes, normo- or microalbuminuria at the 2-year visit, and complete information on CKD (_n_=8300) were analyzed.

Both trials were conducted in accordance with the Declaration of Helsinki, were approved by central medical ethics committees, and are registered at www.ClinicalTrials.gov (NCT00153101, NCT00069784). Written informed consent was obtained from all patients.

Study Outcomes

The outcome states after 5.5 years of follow-up were defined as alive without CKD, alive with CKD, or dead. Incidence or progression of CKD was determined as experiencing at least one of the following end points: new micro- or macroalbuminuria, doubling of creatinine, or ESRD. Micro- and macroalbuminuria were defined as urinary albumin/creatinine ratio (UACR) >30 mg/g (3.4 mg/mmol) and 300 mg/g (33.9 mg/mmol) (14). ESRD was defined as eGFR (CKD Epidemiology Collaboration [CKD-EPI] formula) <15 ml/min per 1.73 m2 or dialysis (15).

Prediction Models

Two prediction models were developed: a laboratory model, containing laboratory markers of kidney function, sex and age, and a clinical model, containing the same markers and some clinical variables. The markers and clinical variables were chosen on the basis of medical knowledge and availability in clinical practice (16–18). Predictors were measured at baseline in ONTARGET and at 2 years in the ORIGIN study.

Laboratory Model.

Predictors included UACR, eGFR, albuminuria stage (normo- or microalbuminuria), sex, and age. UACR was modeled as d-UACRtp, defined as the difference between the stage‐specific cut point for development of new micro‐ or macroalbuminuria and UACR at baseline on the log scale (19). We modeled d-UACRtp in this way to account for the higher probability of a stage change for individuals whose UACR values were close to the stage-specific cut point (either 30 or 300mg/g) and used the log scale to account for the right-skewed distribution (Figure 2). In ONTARGET, urinary albumin and creatinine were measured centrally at baseline and after 5 years (20).

Figure 2.

Explanation of the predictor d-UACRtp. Micro- and macroalbuminuria were defined as UACRs>30 mg/g (3.4 mg/mmol) and 300 mg/g (33.9 mg/mmol), respectively. To predict incidence or progression of CKD after 5.5 years the current (at baseline) UACR measurement is clearly an important predictor. Therefore, for each participant, we calculated d-UACRtp as the difference between each participant’s UACR at baseline and the cutoff that was relevant for this participant. For individuals with normoalbuminuria at baseline (UACR<30 mg/g) the relevant cutoff for incidence of albuminuria is 30 mg/g, and for individuals with microalbuminuria at baseline (30<UACR<300) the relevant cutoff for progression of albuminuria is 300 mg/g. To avoid disproportional influence of large values, UACR at baseline was log transformed before analysis. d-UACRtp, difference between the stage‐specific cut point for development of new micro‐ or macroalbuminuria and UACR at baseline on the log scale; UACR, urinary albumin/creatinine ratio.

Clinical Model.

Predictors included d-UACRtp, eGFR, albuminuria stage, sex, age, race (white, Asian, other), diabetes duration (years, log transformed), fasting LDL (mg/dl), glucose (mg/dl), waist circumference (cm), the comorbidities major atherosclerotic cardiac events (myocardial infarction, stable or unstable angina, coronary artery bypass grafting, or percutaneous interventions, including angioplasty, stenting, atherectomy), laser therapy for diabetic retinopathy, peripheral artery disease (peripheral arterial angioplasty, limb or foot amputation), stroke or transient ischemic attack, and number of antihypertensive drugs prescribed. For antihypertensive drugs we created a scale assigning one point for each group (renin-angiotensin system blocker, calcium channel blocker, _α_-blocker, _β_-blocker, diuretics) from which drugs were prescribed.

Statistical Analyses

Median and interquartile range were used to summarize continuous variables; absolute frequencies and percentages were used for categorical variables. Data availability was very good in both cohorts; the laboratory and clinical models have 0.2% (_n_=15) and 5% (_n_=359) of participants with missing values in ONTARGET. Therefore, a complete-case analysis was applied.

Multinomial logistic regression was applied to develop prediction models for the three outcome states (21). We used the multivariable fractional polynomial algorithm to select predictors at a P value threshold of 0.157 (22), corresponding roughly to a selection according to the Akaike information criterion (23), and to incorporate possible nonlinear associations of UACR and eGFR with the log odds of renal progression. The drugs score was modeled linearly. Binary indicators for comorbidities were either all included or all excluded from the model. All two-way interactions were assessed. The modeling algorithm was applied to ONTARGET to construct both prediction models and to 500 bootstrap resamples for the purpose of internal validation of model stability (24). The bootstrap resamples were also used to compute optimism-corrected performance measures and to estimate global shrinkage. External validity was examined by applying the optimism-corrected models to the ORIGIN study.

Overall, performance and validity of prediction models were evaluated by assessing (1) explained variation, the proportion of variability in the outcome that is explained by the model, using Nagelkerke’s _R_2; (2) discrimination, the ability of prediction models to distinguish individuals with different outcomes, using c-statistics; (3) calibration, the agreement between observed and predicted probabilities, using the calibration slope (which reflects whether predictions are too extreme; i.e., are estimates too low for low predictions and too high for high predictions), calibration-in-the-large (which reflects if predictions are biased, i.e., are they systematically too high or low), and visual inspection of calibration plots; and (4) clinical usefulness using decision curve analysis (Supplemental Table 1).

The importance of individual predictors was quantified by the proportion of variation explained (i.e., the drop in explained variation when each predictor is individually removed from the model) (25).

If applicable, performance measures and plots are given in two versions: one summarizing all three outcome states (overall), and one comparing each outcome state to the other two (state-specific).

Sensitivity Analyses.

We used eGFR using Modification of Diet in Renal Disease (MDRD) instead of eGFR CKD-EPI (Supplemental Figure 2) (15); changed the significance level to 0.1 and 0.05; changed the study population, including only participants with normoalbuminuria at baseline; and treated each comorbidity as a separate predictor for inclusion or exclusion.

For a more detailed description of the statistical analysis see Supplemental Table 2. R software was used for analysis (26).

Results

In ONTARGET, 1079 participants (15.9%) experienced incidence or progression of CKD, and 1032 (15.3%) died; 694 (10.3%) developed microalbuminuria, 312 (4.6%) developed macroalbuminuria, 105 (1.6%) doubled their serum creatinine, and 62 (0.9%) progressed to ESRD or required dialysis after 5.5 years. The study outcomes were similarly distributed in the ORIGIN study. Baseline characteristics of participants of both cohorts are given in Table 1.

Table 1.

Baseline characteristics of participants with diabetes in the development and validation cohorts

Laboratory Model

This model contained d-UACRtp, eGFR, albuminuria stage, sex, and age and exhibits an externally validated explained variation (Nagelkerke’s _R_2) of 10.6% (Table 2). For the outcome states alive with CKD and death, _R_2 values were 7.3% and 10.2%, respectively. The externally validated c-statistic was 0.68, and state-specific c-statistics were 0.66 and 0.70.

Table 2.

Performance of prediction models in the development and validation cohorts

For CKD, by far the most important predictor was d-UACRtp, with a partial explained variation of 7.23% (Table 3). This predictor depicts the difference between the stage‐specific cut point for development of new micro‐ or macroalbuminuria and UACR at baseline on the log scale. For death, the partial explained variation of age was highest. The optimism-corrected calibration-in-the-large reached the ideal value of 0, indicating that predictions were not systematically biased. The lowest optimism-corrected calibration slope with 0.98 is very close to the calibration slope of a perfectly calibrated model of 1. Externally validated calibration-in-the-large and calibration slopes were very similar (Supplemental Figure 4).

Table 3.

Partial explained variation of predictors in the development cohort

Applying the laboratory model to a 65-year-old man with UACR of 5 mg/g and eGFR of 75 ml/min per 1.73 m2 gives a 12.4% probability of CKD in 5.5 years, whereas for another individual with 15 mg/g UACR and the same sex, age, and eGFR, a probability of 25.9% is obtained (Figure 3). A different sex or age barely changes the probability of CKD in 5.5 years (Supplemental Figure 5).

Figure 3.

Predicted probabilities for incidence or progression of CKD after 5.5 years and death within 5.5 years computed by the laboratory model. For a hypothetical 65-year-old man with normoalbuminuria (top panels) or with microalbuminuria (bottom panels), predicted probabilities were computed by inserting eGFR CKD-EPI values and selected UACR values into the laboratory model. The plots reveal that the predicted probability for death within 5.5 years is higher in an individual with microalbuminuria at baseline than in an individual with normoalbuminuria. As a consequence, an individual with microalbuminuria at baseline has a lower predicted probability for CKD after 5.5 years than an individual with normoalbuminuria. CKD-EPI, CKD Epidemiology Collaboration; UACR, urinary albumin/creatinine ratio.

Clinical Model

This model has an externally validated _R_2 of 12.1%, and for the outcomes alive with CKD and death _R_2 values were 8.4% and 12.1%. The externally validated c-statistic was 0.69, and the outcomes alive with CKD and death have c-statistics of 0.68 and 0.71.

For CKD the partial explained variation of the renal predictors (d-UACRtp, eGFR, albuminuria stage) was 5.58%, whereas for the clinical (antihypertensive drugs, glucose, fasting LDL, waist circumference, diabetes duration, organ-specific comorbidities) and demographic (age, sex, race) predictors it was only 1.06% and 0.49%, respectively. For the outcome death, the highest partial explained variation was observed for age, eGFR, and peripheral artery disease. Optimism-corrected calibration-in-the-large reached the ideal value of 0, and externally validated calibration-in-the-large was between 0.2 and 0.14. Optimism-corrected and externally validated calibration slopes were between 0.88 and 0.90, and 1.04 and 1.16, respectively. This indicates a slightly higher overfit than the laboratory model (Supplemental Figure 4).

Sensitivity Analysis

Using eGFR MDRD instead of eGFR CKD-EPI for prediction mainly affects individuals with eGFR>60 ml/min per 1.73 m2; consequently, the performance of the prediction models barely changed, which has been also shown by O'Seaghdha et al. (27). Generally, prediction models in the sensitivity analyses resulted in very similar optimism-corrected performance measures. In the clinical model restricted to individuals with normoalbuminuria at baseline, diabetes duration was not selected. When comorbidities were not simultaneously included or excluded, a more parsimonious clinical model without laser therapy for diabetic retinopathy and diabetes duration, but with a similar predictive performance, was selected.

Online Risk Calculator

Both prediction models were implemented as online risk calculators (available from: http://www.meduniwien.ac.at/nephrogene/index.php/data). The prediction equations are given in Supplemental Table 4.

Discussion

Our analysis shows that risk stratification of people with diabetes for incidence and progression of early CKD and mortality can be accomplished using prediction models. UACR and eGFR were the most important factors predicting CKD. UACR was transformed to reflect how close baseline UACR was to the next threshold for micro- or macroalbuminuria. Inclusion of demographic, clinical, and other laboratory predictors barely improved predictive performance. For mortality, predictive performance was largely driven by age. Further addition of clinical predictors, such as a previous vascular event or vascular comorbidity, enhanced the predictive performance only for the outcome mortality. External validation with the ORIGIN study confirmed the transportability of the prediction models to other cohorts of individuals with diabetes and who were older than 55 years of age. In the ONTARGET cohort, after 5.5 years of follow-up, most (84%) patients developed CKD because of an increase of UACR above the thresholds for micro- or macroalbuminuria. The parsimonious laboratory model explained approximately 11% of the variability of CKD. The externally validated c-statistic of CKD of 0.68 seems low and indicates that other factors associated with incidence or progression of CKD are still not identified. However, considering that we studied a population at very early CKD stages and therefore observed low event rates (within 5.5 years CKD or death occurred in 16% and 15% of participants, respectively), the prediction is reasonable. Moreover, our study represents the only externally validated decision tool available for people aged ≥55 years with diabetes. Given the easy application of the parsimonious prediction model in people with diabetes and cardiovascular risk and the huge expenses of lifelong treatment, a benefit of a few percent is likely cost-effective (4). For example, the incremental cost-effectiveness of primary care screening for albuminuria in people with diabetes between 50–69 years, combined with prescription of renin-angiotensin system blockers for all positive individuals, has been estimated at $4370 (2010) per quality-adjusted life-year gained (28). Utilization of online risk calculators may assist in risk communication with patients motivating them to lifestyle changes and/or adherence to prescribed therapies (29).

The estimation of risk for the progression of CKD in the general population and high-risk cohorts has been investigated by several authors. Gansevoort et al. showed in a combined analysis of nine studies in the general population comprising roughly 840,000 individuals and seven studies of about 170,000 individuals with elevated risk of CKD that UACR and eGFR were also the most important predictors for progression of CKD and ESRD in both cohorts (30). However, the authors did not explicitly study patients with diabetes. Prediction of ESRD in patients with diabetes and advanced renal disease has been performed in the RENAAL (Reduction of Endpoints in NIDDM with the Angiotensin II Antagonist Losartan) study, in cohorts of patients in the Kaiser Permanente database, and in a population in England and Wales (31–33). Although not directly comparable with our study which focuses on very early kidney disease, the landmark articles of later CKD stages showed c-statistics in the range of 0.84–0.90 using the predictors of UACR, eGFR, age, sex, and systolic BP, but the competing risk of death was not addressed (31–36).

Studies of the prediction of early CKD in people with diabetes are scarce, (16,37). Jardine et al. published a prediction model for CKD using the ADVANCE (Action in Diabetes and Vascular Disease: Preterax and Diamicron MR Controlled Evaluation) study, a clinical trial that followed 11,140 participants with type 2 diabetes (16). c-Statistics of a prediction model for major kidney-related events were 0.85, and for a second prediction model for new-onset albuminuria it was 0.65. Asian race and greater waist circumference were predictive of new-onset albuminuria. (In the ONTARGET population with diabetes the incremental explained variation of race and waist circumference was <0.3%.) Of all of the risk prediction models, only the ADVANCE model predicting new-onset microalbuminuria is directly comparable with our prediction models. But, in the ADVANCE study, no external validation was performed, and the competing risk of death was not considered. However, Adler et al. showed the risk of death is higher than the risk of ESRD in patients with diabetes (38). By including the competing risk of death in our models we were also able to observe that eGFR was more important for the prediction of death, whereas UACR had a much greater effect on CKD. With the biomarkers currently available in clinical practice, it appears that prediction of early renal disease is more challenging, even in those with diabetes (c-statistic, 0.66), than the prediction of late events in people with early or intermediate disease. However, it is of great importance because earlier intervention or prevention has the potential to have much greater effect over a patient’s life span.

The prediction models and performance measures were corrected for optimism by internal validation using bootstrap resamples. Transportability to other cohorts of people with diabetes was assessed by external validation with data from the ORIGIN study, a large, well conducted clinical trial. However, our findings apply only to people with type 2 diabetes who were aged ≥55 years and may not be generalizable. ONTARGET included mostly participants of white (68%) and Asian (17%) origin and therefore the validity of our findings in populations of other ethnicities are still to be proven. We included nonlinear effects in our prediction models. In contrast with most other studies the overriding competing risk of death in people with diabetes was considered. We quantified relative importance of individual predictors of CKD and death.

Conclusions

We present the first externally validated risk prediction models for incidence or progression of early CKD, which also consider the competing risk of death, in individuals aged ≥55 years with type 2 diabetes. Given the high prevalence of diabetes worldwide, the increase of precision in the prediction of nephropathy by using easily applicable risk models can facilitate an early identification of individuals with elevated risk for CKD.

Disclosures

None.

Supplementary Material

Supplemental Data

Acknowledgments

ONTARGET and ORIGIN were sponsored by Boehringer Ingelheim and Sanofi. Funding for this study was provided by the European Union’s seven-framework program (SysKid HEALTH–F2–2009–241544).

The funding agency and the sponsors had no role in the design and conduct of the study, the collection, analysis, and interpretation of the data in the preparation, review, or approval of the manuscript.

Footnotes

Published online ahead of print. Publication date available at www.cjasn.org.

References

• 1.Coresh J, Selvin E, Stevens LA, Manzi J, Kusek JW, Eggers P, Van Lente F, Levey AS: Prevalence of chronic kidney disease in the United States. JAMA 298: 2038–2047, 2007 [DOI] [PubMed] [Google Scholar]
• 2.Hunsicker LG, Adler S, Caggiula A, England BK, Greene T, Kusek JW, Rogers NL, Teschan PE: Predictors of the progression of renal disease in the Modification of Diet in Renal Disease Study. Kidney Int 51: 1908–1919, 1997 [DOI] [PubMed] [Google Scholar]
• 3.Taal MW: Predicting renal risk in the general population: Do we have the right formula? Clin J Am Soc Nephrol 6: 1523–1525, 2011 [DOI] [PubMed] [Google Scholar]
• 4.Boulware LE, Jaar BG, Tarver-Carr ME, Brancati FL, Powe NR: Screening for proteinuria in US adults: A cost-effectiveness analysis. JAMA 290: 3101–3114, 2003 [DOI] [PubMed] [Google Scholar]
• 5.Anderson KM, Odell PM, Wilson PW, Kannel WB: Cardiovascular disease risk profiles. Am Heart J 121: 293–298, 1991 [DOI] [PubMed] [Google Scholar]
• 6.Manns B, Hemmelgarn B, Tonelli M, Au F, Chiasson TC, Dong J, Klarenbach S, Alberta Kidney Disease Network : Population based screening for chronic kidney disease: Cost effectiveness study. BMJ 341: c5869, 2010 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 7.Teo K, Yusuf S, Sleight P, Anderson C, Mookadam F, Ramos B, Hilbrich L, Pogue J, Schumacher H, ONTARGET/TRANSCEND Investigators : Rationale, design, and baseline characteristics of 2 large, simple, randomized trials evaluating telmisartan, ramipril, and their combination in high-risk patients: The Ongoing Telmisartan Alone and in Combination with Ramipril Global Endpoint Trial/Telmisartan Randomized Assessment Study in ACE Intolerant Subjects with Cardiovascular Disease (ONTARGET/TRANSCEND) trials. Am Heart J 148: 52–61, 2004 [DOI] [PubMed] [Google Scholar]
• 8.Yusuf S, Teo KK, Pogue J, Dyal L, Copland I, Schumacher H, Dagenais G, Sleight P, Anderson C, ONTARGET Investigators : Telmisartan, ramipril, or both in patients at high risk for vascular events. N Engl J Med 358: 1547–1559, 2008 [DOI] [PubMed] [Google Scholar]
• 9.Mann JF, Schmieder RE, McQueen M, Dyal L, Schumacher H, Pogue J, Wang X, Maggioni A, Budaj A, Chaithiraphan S, Dickstein K, Keltai M, Metsärinne K, Oto A, Parkhomenko A, Piegas LS, Svendsen TL, Teo KK, Yusuf S, ONTARGET investigators : Renal outcomes with telmisartan, ramipril, or both, in people at high vascular risk (the ONTARGET study): A multicentre, randomised, double-blind, controlled trial. Lancet 372: 547–553, 2008 [DOI] [PubMed] [Google Scholar]
• 10.Heinze G, Schemper M: Comparing the importance of prognostic factors in Cox and logistic regression using SAS. Comput Methods Programs Biomed 71: 155–163, 2003 [DOI] [PubMed] [Google Scholar]
• 11.Gerstein H, Yusuf S, Riddle MC, Ryden L, Bosch J, Origin Trial Investigators : Rationale, design, and baseline characteristics for a large international trial of cardiovascular disease prevention in people with dysglycemia: the ORIGIN Trial (Outcome Reduction with an Initial Glargine Intervention). Am Heart J 155: 26–32, e1–e6, 2008 [DOI] [PubMed] [Google Scholar]
• 12.Bosch J, Gerstein HC, Dagenais GR, Díaz R, Dyal L, Jung H, Maggiono AP, Probstfield J, Ramachandran A, Riddle MC, Rydén LE, Yusuf S, ORIGIN Trial Investigators : n-3 fatty acids and cardiovascular outcomes in patients with dysglycemia. N Engl J Med 367: 309–318, 2012 [DOI] [PubMed] [Google Scholar]
• 13.Gerstein HC, Bosch J, Dagenais GR, Díaz R, Jung H, Maggioni AP, Pogue J, Probstfield J, Ramachandran A, Riddle MC, Rydén LE, Yusuf S, ORIGIN Trial Investigators : Basal insulin and cardiovascular and other outcomes in dysglycemia. N Engl J Med 367: 319–328, 2012 [DOI] [PubMed] [Google Scholar]
• 14.Justesen TI, Petersen JL, Ekbom P, Damm P, Mathiesen ER: Albumin-to-creatinine ratio in random urine samples might replace 24-h urine collections in screening for micro- and macroalbuminuria in pregnant woman with type 1 diabetes. Diabetes Care 29: 924–925, 2006 [DOI] [PubMed] [Google Scholar]
• 15.Levey AS, Coresh J, Balk E, Kausz AT, Levin A, Steffes MW, Hogg RJ, Perrone RD, Lau J, Eknoyan G, National Kidney Foundation : National Kidney Foundation practice guidelines for chronic kidney disease: Evaluation, classification, and stratification. Ann Intern Med 139: 137–147, 2003 [DOI] [PubMed] [Google Scholar]
• 16.Jardine MJ, Hata J, Woodward M, Perkovic V, Ninomiya T, Arima H, Zoungas S, Cass A, Patel A, Marre M, Mancia G, Mogensen CE, Poulter N, Chalmers J, ADVANCE Collaborative Group : Prediction of kidney-related outcomes in patients with type 2 diabetes. Am J Kidney Dis 60: 770–778, 2012 [DOI] [PubMed] [Google Scholar]
• 17.Ravid M, Brosh D, Ravid-Safran D, Levy Z, Rachmani R: Main risk factors for nephropathy in type 2 diabetes mellitus are plasma cholesterol levels, mean blood pressure, and hyperglycemia. Arch Intern Med 158: 998–1004, 1998 [DOI] [PubMed] [Google Scholar]
• 18.Retnakaran R, Cull CA, Thorne KI, Adler AI, Holman RR, UKPDS Study Group : Risk factors for renal dysfunction in type 2 diabetes: U.K. Prospective Diabetes Study 74. Diabetes 55: 1832–1839, 2006 [DOI] [PubMed] [Google Scholar]
• 19.Dunkler D, Dehghan M, Teo KK, Heinze G, Gao P, Kohl M, Clase CM, Mann JF, Yusuf S, Oberbauer R, ONTARGET Investigators : Diet and kidney disease in high-risk individuals with type 2 diabetes mellitus. JAMA Intern Med 173: 1682–1692, 2013 [DOI] [PubMed] [Google Scholar]
• 20.Schmieder RE, Mann JF, Schumacher H, Gao P, Mancia G, Weber MA, McQueen M, Koon T, Yusuf S, ONTARGET Investigators : Changes in albuminuria predict mortality and morbidity in patients with vascular disease. J Am Soc Nephrol 22: 1353–1364, 2011 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 21.Biesheuvel CJ, Vergouwe Y, Steyerberg EW, Grobbee DE, Moons KG: Polytomous logistic regression analysis could be applied more often in diagnostic research. J Clin Epidemiol 61: 125–134, 2008 [DOI] [PubMed] [Google Scholar]
• 22.Royston P, Sauerbrei W: Building multivariable regression models with continuous covariates in clinical epidemiology--with an emphasis on fractional polynomials. Methods Inf Med 44: 561–571, 2005 [PubMed] [Google Scholar]
• 23.Royston P, Sauerbrei W: Multivariable model-building. A pragmatic approach to regression analysis based on fractional polynomials for modelling continuous variables, Chichester, England, John Wiley & Sons, 2008, pp 1–322 [Google Scholar]
• 24.Steyerberg EW, Harrell FE, Jr, Borsboom GJ, Eijkemans MJ, Vergouwe Y, Habbema JD: Internal validation of predictive models: Efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 54: 774–781, 2001 [DOI] [PubMed] [Google Scholar]
• 25.Schemper M: The relative importance of prognostic factors in studies of survival. Stat Med 12: 2377–2382, 1993 [DOI] [PubMed] [Google Scholar]
• 26.R Core Team : R: A language and environment for statistical computing, Vienna, Austria, R Foundation for Statistical Computing, 2013 [Google Scholar]
• 27.O’Seaghdha CM, Lyass A, Massaro JM, Meigs JB, Coresh J, D’Agostino RB, Sr, Astor BC, Fox CS: A risk score for chronic kidney disease in the general population. Am J Med 125: 270–277, 2012 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 28.Howard K, White S, Salkeld G, McDonald S, Craig JC, Chadban S, Cass A: Cost-effectiveness of screening and optimal management for diabetes, hypertension, and chronic kidney disease: A modeled analysis. Value Health 13: 196–208, 2010 [DOI] [PubMed] [Google Scholar]
• 29.Echouffo-Tcheugui JB, Kengne AP: Risk models to predict chronic kidney disease and its progression: A systematic review. PLoS Med 9: e1001344, 2012 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 30.Gansevoort RT, Matsushita K, van der Velde M, Astor BC, Woodward M, Levey AS, de Jong PE, Coresh J, Chronic Kidney Disease Prognosis Consortium : Lower estimated GFR and higher albuminuria are associated with adverse kidney outcomes. A collaborative meta-analysis of general and high-risk population cohorts. Kidney Int 80: 93–104, 2011 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 31.Keane WF, Zhang Z, Lyle PA, Cooper ME, de Zeeuw D, Grunfeld JP, Lash JP, McGill JB, Mitch WE, Remuzzi G, Shahinfar S, Snapinn SM, Toto R, Brenner BM, RENAAL Study Investigators : Risk scores for predicting outcomes in patients with type 2 diabetes and nephropathy: The RENAAL study. Clin J Am Soc Nephrol 1: 761–767, 2006 [DOI] [PubMed] [Google Scholar]
• 32.Johnson ES, Thorp ML, Platt RW, Smith DH: Predicting the risk of dialysis and transplant among patients with CKD: A retrospective cohort study. Am J Kidney Dis 52: 653–660, 2008 [DOI] [PubMed] [Google Scholar]
• 33.Hippisley-Cox J, Coupland C: Predicting the risk of chronic kidney disease in men and women in England and Wales: Prospective derivation and external validation of the QKidney Scores. BMC Fam Pract 11: 49, 2010 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 34.Tangri N, Stevens LA, Griffith J, Tighiouart H, Djurdjev O, Naimark D, Levin A, Levey AS: A predictive model for progression of chronic kidney disease to kidney failure. JAMA 305: 1553–1559, 2011 [DOI] [PubMed] [Google Scholar]
• 35.Halbesma N, Jansen DF, Heymans MW, Stolk RP, de Jong PE, Gansevoort RT, PREVEND Study Group : Development and validation of a general population renal risk score. Clin J Am Soc Nephrol 6: 1731–1738, 2011 [DOI] [PubMed] [Google Scholar]
• 36.Dimitrov BD, Perna A, Ruggenenti P, Remuzzi G: Predicting end-stage renal disease: Bayesian perspective of information transfer in the clinical decision-making process at the individual level. Kidney Int 63: 1924–1933, 2003 [DOI] [PubMed] [Google Scholar]
• 37.Clase CM, Gao P, Tobe SW, McQueen MJ, Grosshennig A, Teo KK, Yusuf S, Mann JF, ONTARGET (ONgoing Telmisartan Alone and in combination with Ramipril Global Endpoint Trial) and TRANSCEND (Telmisartan Randomized Assessment Study in Angiotensin-Converting-Enzyme-Inhibitor Intolerant Subjects with Cardiovascular Disease) : Estimated glomerular filtration rate and albuminuria as predictors of outcomes in patients with high cardiovascular risk: A cohort study. Ann Intern Med 154: 310–318, 2011 [DOI] [PubMed] [Google Scholar]
• 38.Adler AI, Stevens RJ, Manley SE, Bilous RW, Cull CA, Holman RR, UKPDS GROUP : Development and progression of nephropathy in type 2 diabetes: The United Kingdom Prospective Diabetes Study (UKPDS 64). Kidney Int 63: 225–232, 2003 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplemental Data