Advertisement

Improved clinical trial enrollment criterion to identify patients with diabetes at risk of end-stage renal disease

  • Author Footnotes
    10 Both authors contributed equally.
    Masayuki Yamanouchi
    Footnotes
    10 Both authors contributed equally.
    Affiliations
    Section on Genetics and Epidemiology, Research Divisions, Joslin Diabetes Center, Boston, Massachusetts, USA

    Department of Medicine, Harvard Medical School, Boston, Massachusetts, USA
    Search for articles by this author
  • Author Footnotes
    10 Both authors contributed equally.
    Jan Skupien
    Correspondence
    Correspondence: Jan Skupien, Department of Metabolic Diseases, Jagiellonian University Medical College, 15 Kopernika Street, 31-501 Krakow, Poland.
    Footnotes
    10 Both authors contributed equally.
    Affiliations
    Section on Genetics and Epidemiology, Research Divisions, Joslin Diabetes Center, Boston, Massachusetts, USA

    Department of Medicine, Harvard Medical School, Boston, Massachusetts, USA

    Department of Metabolic Disease, Jagellonian University Medical College, Krakow, Poland
    Search for articles by this author
  • Monika A. Niewczas
    Affiliations
    Section on Genetics and Epidemiology, Research Divisions, Joslin Diabetes Center, Boston, Massachusetts, USA

    Department of Medicine, Harvard Medical School, Boston, Massachusetts, USA
    Search for articles by this author
  • Adam M. Smiles
    Affiliations
    Section on Genetics and Epidemiology, Research Divisions, Joslin Diabetes Center, Boston, Massachusetts, USA
    Search for articles by this author
  • Alessandro Doria
    Affiliations
    Section on Genetics and Epidemiology, Research Divisions, Joslin Diabetes Center, Boston, Massachusetts, USA

    Department of Medicine, Harvard Medical School, Boston, Massachusetts, USA
    Search for articles by this author
  • Robert C. Stanton
    Affiliations
    Department of Medicine, Harvard Medical School, Boston, Massachusetts, USA

    Renal Division, Beth Israel Deaconess Medical Center, Boston, Massachusetts, USA
    Search for articles by this author
  • Andrzej T. Galecki
    Affiliations
    Institute of Gerontology, University of Michigan Medical School, Ann Arbor, Michigan, USA

    Department of Biostatistics, University of Michigan School of Public Health, Ann Arbor, Michigan, USA
    Search for articles by this author
  • Kevin L. Duffin
    Affiliations
    Lilly Research Laboratories, Eli Lilly & Company Inc. Corporate Center, Indianapolis, Indiana, USA
    Search for articles by this author
  • Nick Pullen
    Affiliations
    Pfizer Inc., 610 Main Street, Cambridge, Massachusetts, 02139, USA
    Search for articles by this author
  • Matthew D. Breyer
    Affiliations
    Lilly Research Laboratories, Eli Lilly & Company Inc. Corporate Center, Indianapolis, Indiana, USA
    Search for articles by this author
  • Joseph V. Bonventre
    Affiliations
    Department of Medicine, Harvard Medical School, Boston, Massachusetts, USA

    Renal Division, Brigham & Women Hospital, Boston, Massachusetts, USA
    Search for articles by this author
  • James H. Warram
    Affiliations
    Section on Genetics and Epidemiology, Research Divisions, Joslin Diabetes Center, Boston, Massachusetts, USA
    Search for articles by this author
  • Andrzej S. Krolewski
    Correspondence
    Andrzej S. Krolewski, Section on Genetics & Epidemiology, Joslin Diabetes Center, One Joslin Place, Boston, Massachusetts, 02215, USA.
    Affiliations
    Section on Genetics and Epidemiology, Research Divisions, Joslin Diabetes Center, Boston, Massachusetts, USA

    Department of Medicine, Harvard Medical School, Boston, Massachusetts, USA
    Search for articles by this author
  • Author Footnotes
    10 Both authors contributed equally.
Open ArchivePublished:April 07, 2017DOI:https://doi.org/10.1016/j.kint.2017.02.010
      Design of Phase III trials for diabetic nephropathy currently requires patients at a high risk of progression defined as within three years of a hard end point (end-stage renal disease, 40% loss of estimated glomerular filtration rate, or death). To improve the design of these trials, we used natural history data from the Joslin Kidney Studies of chronic kidney disease in patients with diabetes to develop an improved criterion to identify such patients. This included a training cohort of 279 patients with type 1 diabetes and 134 end points within three years, and a validation cohort of 221 patients with type 2 diabetes and 88 end points. Previous trials selected patients using clinical criteria for baseline urinary albumin-to-creatinine ratio and estimated glomerular filtration rate. Application of these criteria to our cohort data yielded sensitivities (detection of patients at risk) of 70-80% and prognostic values of only 52-63%. We applied classification and regression trees analysis to select from among all clinical characteristics and markers the optimal prognostic criterion that divided patients with type 1 diabetes according to risk. The optimal criterion was a serum tumor necrosis factor receptor 1 level over 4.3 ng/ml alone or 2.9-4.3 ng/ml with an albumin-to-creatinine ratio over 1900 mg/g. Remarkably, this criterion produced similar results in both type 1 and type 2 diabetic patients. Overall, sensitivity and prognostic value were high (72% and 81%, respectively). Thus, application of this criterion to enrollment in future clinical trials could reduce the sample size required to achieve adequate statistical power for detection of treatment benefits.

      Keywords

      Although progress has been made over the last 20 years for the prevention and treatment of diabetic kidney disease, the risk of end-stage renal disease (ESRD) remains high.
      • Rosolowsky E.T.
      • Skupien J.
      • Smiles A.M.
      • et al.
      Risk for ESRD in type 1 diabetes remains high despite renoprotection.
      • De Boer I.H.
      • Rue T.C.
      • Hall Y.N.
      • et al.
      Temporal trends in the prevalence of diabetic kidney disease in the United States.
      To mitigate this risk, new interventions must be developed and tested in clinical trials. The enormous expense of such trials effectively limits the trial duration to 3 years, and this time constraint is challenging for diabetic nephropathy, which evolves over many years. To be successful, the trial must accumulate enough end points to give adequate power for detecting a risk reduction between the placebo and the treated group. Thus, the trial size (and cost) is inversely related to how effectively the enrollment criteria enrich the study group with patients who develop ESRD within the timeframe of the trial—typically 3 years. This characteristic is the prognostic value of the enrollment criterion.
      A review of clinical trials in advanced diabetic nephropathy conducted over the last 20 years illustrates the challenge.
      • Lewis E.J.
      • Hunsicker L.G.
      • Bain R.P.
      • Rohde R.D.
      The effect of angiotensin-converting-enzyme inhibition on diabetic nephropathy. The Collaborative Study Group.
      • Lewis E.J.
      • Hunsicker L.G.
      • Clarke W.R.
      • et al.
      Renoprotective effect of the angiotensin-receptor antagonist irbesartan in patients with nephropathy due to type 2 diabetes.
      • Brenner B.M.
      • Cooper M.E.
      • de Zeeuw D.
      • et al.
      Effects of losartan on renal and cardiovascular outcomes in patients with type 2 diabetes and nephropathy.
      • Facchini F.S.
      • Saylor K.L.
      A low-iron-available, polyphenol-enriched, carbohydrate-restricted diet to slow progression of diabetic nephropathy.
      • House A.A.
      • Eliasziw M.
      • Cattran D.C.
      • et al.
      Effect of B-vitamin therapy on progression of diabetic nephropathy: a randomized controlled trial.
      • Heerspink H.J.
      • Ninomiya T.
      • Perkovic V.
      • et al.
      Effects of a fixed combination of perindopril and indapamide in patients with type 2 diabetes and chronic kidney disease.
      • Sharma K.
      • Ix J.H.
      • Mathew A.V.
      • et al.
      Pirfenidone for diabetic nephropathy.
      • Imai E.
      • Chan J.C.
      • Ito S.
      • et al.
      Effects of olmesartan on renal and cardiovascular outcomes in type 2 diabetes with overt nephropathy: a multicentre, randomised, placebo-controlled study.
      • Parving H.H.
      • Brenner B.M.
      • McMurray J.J.
      • et al.
      Cardiorenal end points in a trial of aliskiren for type 2 diabetes.
      • De Zeeuw D.
      • Akizawa T.
      • Audhya P.
      • et al.
      Bardoxolone methyl in type 2 diabetes and stage 4 chronic kidney disease.
      • Navarro-González J.F.
      • Mora-Fernández C.
      • Muros de Fuentes M.
      • et al.
      Effect of pentoxifylline on renal function and urinary albumin excretion in patients with diabetic kidney disease: the PREDIAN trial.
      Altogether, only 941 of 8528 patients (11%) who enrolled in the placebo groups in all the trials
      • Lewis E.J.
      • Hunsicker L.G.
      • Bain R.P.
      • Rohde R.D.
      The effect of angiotensin-converting-enzyme inhibition on diabetic nephropathy. The Collaborative Study Group.
      • Lewis E.J.
      • Hunsicker L.G.
      • Clarke W.R.
      • et al.
      Renoprotective effect of the angiotensin-receptor antagonist irbesartan in patients with nephropathy due to type 2 diabetes.
      • Brenner B.M.
      • Cooper M.E.
      • de Zeeuw D.
      • et al.
      Effects of losartan on renal and cardiovascular outcomes in patients with type 2 diabetes and nephropathy.
      • Facchini F.S.
      • Saylor K.L.
      A low-iron-available, polyphenol-enriched, carbohydrate-restricted diet to slow progression of diabetic nephropathy.
      • House A.A.
      • Eliasziw M.
      • Cattran D.C.
      • et al.
      Effect of B-vitamin therapy on progression of diabetic nephropathy: a randomized controlled trial.
      • Heerspink H.J.
      • Ninomiya T.
      • Perkovic V.
      • et al.
      Effects of a fixed combination of perindopril and indapamide in patients with type 2 diabetes and chronic kidney disease.
      • Sharma K.
      • Ix J.H.
      • Mathew A.V.
      • et al.
      Pirfenidone for diabetic nephropathy.
      • Imai E.
      • Chan J.C.
      • Ito S.
      • et al.
      Effects of olmesartan on renal and cardiovascular outcomes in type 2 diabetes with overt nephropathy: a multicentre, randomised, placebo-controlled study.
      • Parving H.H.
      • Brenner B.M.
      • McMurray J.J.
      • et al.
      Cardiorenal end points in a trial of aliskiren for type 2 diabetes.
      • De Zeeuw D.
      • Akizawa T.
      • Audhya P.
      • et al.
      Bardoxolone methyl in type 2 diabetes and stage 4 chronic kidney disease.
      • Navarro-González J.F.
      • Mora-Fernández C.
      • Muros de Fuentes M.
      • et al.
      Effect of pentoxifylline on renal function and urinary albumin excretion in patients with diabetic kidney disease: the PREDIAN trial.
      reached a primary end point (Supplementary Table S1, Supplementary Material, part A), so the prognostic value of the enrollment criteria was only 11%. The remaining 89% of enrolled patients did not show any benefit, regardless of the intervention’s effectiveness because their outcome (ESRD) was outside the timeframe of the study, and therefore, was uninformative. Surprisingly, there are no published studies regarding any effort made to develop optimal enrollment criteria for Phase III trials for diabetic nephropathy.
      We report here our attempt to address this gap with a comprehensive effort to develop an enrollment criterion with high prognostic value and sensitivity to support successful Phase III trials with a 3-year time constraint. Moreover, we sought to keep the criterion simple by using the fewest markers that take minimal effort to measure and have the greatest reproducibility over time so registries of patients eligible for clinical trials can be easily maintained. Toward this end, we examined baseline and follow-up data of patients with diabetes and chronic kidney disease (CKD) who were enrolled in the studies conducted at the Joslin Diabetes Center. Consideration was restricted to CKD stages 3 and 4 because patients with CKD stage 1 or 2 rarely progress to ESRD within the 3-year period.
      • Rosolowsky E.T.
      • Skupien J.
      • Smiles A.M.
      • et al.
      Risk for ESRD in type 1 diabetes remains high despite renoprotection.
      • Forsblom C.
      • Harjutsalo V.
      • Thorn L.M.
      • et al.
      Competing-risk analysis of ESRD and deaths among patients with Type 1 diabetes and macroalbuminuria.
      • Adler A.I.
      • Stevens R.J.
      • Manley S.E.
      • et al.
      Development and progression of nephropathy in type 2 diabetes: the United Kingdom Prospective Diabetes Study (UKPDS 64).
      Follow-up data of these patients included occurrence of ESRD, 40% loss of baseline estimated glomerular filtration rate (eGFR), or death unrelated to ESRD within 3 years.
      In the search for an optimal enrollment criterion, we evaluated the usual clinical characteristics plus 2 legacy markers for nephropathy, albumin-to-creatinine ratio (ACR) and eGFR, and 2 novel ones, elevated serum tumor necrosis factor receptor (TNFR)1 and TNFR2. Strong association of TNFRs with the risk of ESRD in type 2 (T2D)
      • Niewczas M.A.
      • Gohda T.
      • Skupien J.
      • et al.
      Circulating TNF receptors 1 and 2 predict ESRD in type 2 diabetes.
      and type 1 diabetes (T1D)
      • Skupien J.
      • Warram J.H.
      • Niewczas M.A.
      • et al.
      Synergism between circulating tumor necrosis factor receptor 2 and HbA(1c) in determining renal decline during 5-18 years of follow-up in patients with type 1 diabetes and proteinuria.
      and the rate of early renal decline
      • Gohda T.
      • Niewczas M.A.
      • Ficociello L.H.
      • et al.
      Circulating TNF receptors 1 and 2 predict stage 3 CKD in type 1 diabetes.
      • Krolewski A.
      • Niewczas M.A.
      • Skupien J.
      • et al.
      Early progressive renal decline precedes the onset of microalbuminuria and its progression to macroalbuminuria.
      was previously shown by us and confirmed in cohort studies elsewhere.
      • Lopes-Virella M.F.
      • Baker N.L.
      • Hunt K.J.
      • et al.
      Response to comment on Lopes-Virella, et al. Baseline markers of inflammation are associated with progression to macroalbuminuria in type 1 diabetic subjects. Diabetes Care 2013;36:2317-2323.
      • Saulnier P.J.
      • Gand E.
      • Ragot S.
      • et al.
      Association of serum concentration of TNFR1 with all-cause mortality in patients with type 2 diabetes and chronic kidney disease: follow-up of the SURDIAGENE Cohort.
      • Forsblom C.
      • Moran J.
      • Harjutsalo V.
      • et al.
      Added value of soluble tumor necrosis factor-alpha receptor 1 as a biomarker of ESRD risk in patients with type 1 diabetes.
      • Pavkov M.E.
      • Nelson R.G.
      • Knowler W.C.
      • et al.
      Elevation of circulating TNF receptors 1 and 2 increases the risk of end-stage renal disease in American Indians with type 2 diabetes.
      We used an example of machine learning methods called the classification and regression trees (CARTs).
      • Breiman L.
      • Friedman J.H.
      • Stone C.J.
      • Olshen R.A.
      Classification and Regression Trees.
      CART identifies the smallest set of the best performing clinical characteristics or markers and searches for values (cut-points) for each that maximize the separation between patients who are or are not at a high risk of an outcome.
      To illustrate, we applied the optimal enrollment criterion to the design of a hypothetical 3-year clinical trial to show the impact on reducing the sample size while increasing the statistical power. Finally, we used data of patients with CKD who were excluded from 3-year trials on the basis of this new criterion to determine the magnitude of the ESRD risk problem that cannot be studied in clinical trials as currently constrained. Patients with slower progression to ESRD could be a significant source of enrollees into clinical trials if the change in the eGFR slope is accepted as a hard end point or the duration of the clinical trials is extended to 10 years.

      Results

       Characteristics of the study cohorts

      The study group comprised 2 independent cohorts of patients with diabetes and impaired renal function (CKD stage 3 or 4) enrolled in the follow-up studies conducted between 1991 and 2009 at the Joslin Diabetes Center. The majority were enrolled in the 2000s and followed until 2012-2013. Before enrollment, these patients were under the care of the Joslin clinic for a long time, and their inclusion in prospective studies was unrelated to their unknown future outcomes during the subsequent 4 to 15 years. Thus, the findings of this study reflect the unbiased contemporary natural history of CKD and the development of ESRD in patients with diabetes.
      The T1D cohort with proteinuria and its larger number of renal outcomes served as the training panel, while the T2D cohort with microalbuminuria or proteinuria served as the validation panel. Characteristics of these cohorts are summarized in Table 1. In the T1D cohort, the mean age at onset was 13 years, and all patients were treated with insulin. At enrollment, all patients had long-duration diabetes and poor glycemic control, and 85% received the renin-angiotensin system (RAS) blockade. Eligibility required that proteinuria was documented by multiple ACR measurement during the 2-year interval preceding enrollment; however, the single ACR measurement at baseline examination fell in the proteinuric range for 75% of patients rather than 100%. The median ACR for the whole group was 827 mg/g. Also by design, renal function was impaired (median, 40 ml/min). Median serum concentrations of TNFR1 and TNFR2 at baseline were almost twice as high as those in T1D patients with normoalbuminuria and normal renal function.
      • Krolewski A.
      • Niewczas M.A.
      • Skupien J.
      • et al.
      Early progressive renal decline precedes the onset of microalbuminuria and its progression to macroalbuminuria.
      Table 1Clinical characteristics of the study groups
      CharacteristicJoslin T1D

      N = 279
      Joslin T2D

      N = 221
      At baseline:
      Men48%61%
      Age at DM onset (yr)13 (8, 20)44 (38, 50)
      Age at entry (yr)44 (37, 51)61 (57, 64)
      Duration of DM (yr)28 (22, 36)16 (12, 21)
      Duration of care at Joslin (yr)22 (14, 32)6 (2, 13)
      Insulin Rx100%78%
      HbA1c (%)8.6 (7.6, 9.5)7.6 (6.8, 8.8)
      Systolic BP (mm Hg)133 (124, 149)139 (125, 152)
      Diastolic BP (mm Hg)77 (70, 84)74 (70, 83)
      Renoprotective Rx85%92%
      ACR (mg/g)827 (312, 1855)601 (117, 1734)
      Microalbuminuria25%34%
      eGFR (ml/min)40 (28, 49)42 (31, 50)
      TNFR1 (ng/ml)3.9 (2.8, 5.0)3.5 (2.5, 4.4)
      TNFR2 (ng/ml)7.6 (5.7, 9.5)6.8 (5.0, 8.7)
      During the first 3 years of follow-up:
      Cumulative risk of
      ESRD only
      Calculated under the assumption of competing risks among ESRD, eGFR loss, and deaths unrelated to ESRD.
      32.2% (26.6, 38.0) [85]18.9% (13.8, 24.5) [39]
      >40% eGFR loss
      Calculated under the assumption of competing risks among ESRD, eGFR loss, and deaths unrelated to ESRD.
      14.0% (10.2, 18.3) [39]14.5% (10.2, 19.5) [32]
      Deaths unrelated to ESRD
      Calculated under the assumption of competing risks among ESRD, eGFR loss, and deaths unrelated to ESRD.
      3.9% (2.0, 6.8) [10]8.3% (5.0, 12.6) [17]
      Composite end point48.0% (42.2, 53.9) [134]39.8% (33.5, 46.4) [88]
      ACR, albumin-to-creatinine ratio; BP, blood pressure; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; ESRD, end-stage renal disease; HbA1c, hemoglobin A1c; TNFR, tumor necrosis factor receptor; T1D, type 1 diabetes; T2D, type 2 diabetes.
      Data are percent or median (25th, 75th percentiles) [number of events].
      a Calculated under the assumption of competing risks among ESRD, eGFR loss, and deaths unrelated to ESRD.
      To examine the applicability of the prognostic test developed in the T1D cohort to other patients with diabetes, we applied them to the Joslin T2D cohort. Diabetes was diagnosed at a much older age in this cohort (Table 1), and at enrollment, the duration of diabetes and care at the Joslin clinic was much shorter and glycemic control was better than those for the T1D cohort. Systolic and diastolic blood pressures were similar in both the cohorts, and 92% received RAS blockade. Microalbuminuria or proteinuria was documented during the preceding 2 years. The single ACR measurement at baseline examination was in the proteinuric range for 66% of patients. The median ACR for the whole group was 601 mg/g. The median serum concentrations of TNFR1 and TNFR2 at baseline in the T2D cohort were similar to those in the T1D cohort.
      Because ESRD typically develops over an interval longer than the 3-year duration of a Phase III trial, the US Food and Drug Administration has considered accepting a 40% loss of baseline eGFR as a surrogate end point for ESRD.
      • Levey A.S.
      • Inker L.A.
      • Matsushita K.
      • et al.
      GFR decline as an end point for clinical trials in CKD: a scientific workshop sponsored by the National Kidney Foundation and the US Food and Drug Administration.
      In many trials, death regardless of cause has also been included as a composite trial end point. Therefore, we defined the end point as an onset of ESRD, a ≥40% loss of baseline eGFR, or death occurring within the 3-year follow-up. In the T1D cohort, 134 patients reached the end point, giving a 3-year cumulative risk of 48% (85 with ESRD, 39 with ≥40% eGFR loss, and 10 deaths unrelated to ESRD). Notably, among the 39 patients with ≥40% eGFR loss, ESRD developed in all during the subsequent 4 to 10 years of follow-up. In the T2D cohort, 88 patients reached the end point, yielding a 3-year cumulative risk of 40% (39 with ESRD, 32 with ≥40% eGFR loss, and 17 deaths unrelated to ESRD). Among the 32 patients with ≥40% eGFR loss, ESRD developed in all during the subsequent 4 to 6 years of follow-up. To simplify the text, we refer to the risk of the composite end point as risk of ESRD.

       Prognostic performance of enrollment criteria used in previous studies

      Reported trials of measures to prevent ESRD selected patients with low eGFR or high ACR as high-risk patients using cut-points adopted from clinical practice, that is, <45 ml/min for eGFR and ≥500 mg/g for ACR. We evaluated the performance of these enrollment criteria in the follow-up data of both cohorts (Table 2). Predictive value is the 3-year cumulative risk for ESRD in the eligible enrollees, and sensitivity is the proportion of outcomes in the whole cohort captured in the eligible subset. The predictive value for the 2 markers individually was moderate, approximately 60%, and increased to approximately 70% when the markers were combined. While sensitivity was high for individual markers, approximately 80%, it decreased to 65% when they were combined. For completeness, we also provide a negative predictive value, that is, the 3-year cumulative risk of ESRD in patients excluded from a trial, and specificity, that is, the proportion of patients without end points among those not eligible for a trial.
      Table 2Prognostic values, sensitivity, and specificity of alternative enrollment criteria for detecting end-stage renal disease (composite end point) during 3 years of follow-up in the Joslin cohorts with diabetes and chronic kidney disease
      A composite end point developed in 134 of 279 T1D patients and in 88 of 221 T2D patients.
      Marker and criterionPatients with positive test

      n (%)
      Prognostic value in those with positive test (3-year cumulative risk)

      % [95% CI]
      Prognostic value in those with negative test (3-year cumulative risk)

      % [95% CI]
      Sensitivity

      %
      Specificity

      %
      eGFR ≤45 ml/min
       T1D172 (62%)63 [56–70]24 [17–34]8156
       T2D132 (60%)52 [43–60]22 [15–33]7752
      ACR >500 mg/g
       T1D178 (64%)61 [54–68]26 [18–35]8152
       T2D116 (52%)59 [51–68]18 [12–27]7865
      eGFR ≤45 ml/min and ACR >500 mg/g
       T1D122 (44%)73 [65–80]29 [22–36]6677
       T2D85 (38%)67 [57–77]23 [17–31]6579
      Multimarker criterion developed by CART
      The following markers measured at baseline were included: sex, age, HbA1c, eGFR, ACR, and serum TNFR1, and TNFR2.
      :
      Serum TNFR1 >4.3 ng/ml only or serum TNFR1 2.9–4.3 ng/ml and ACR >1900 mg/g
       T1D119 (43%)83 [76–89]22 [16–29]7486
       T2D79 (36%)76 [66–85]20 [14–27]6886
      Serum TNFR1 >4.3 ng/ml only or serum TNFR1 2.9–4.3 ng/ml and ACR >1900 mg/g (excluding deaths unrelated to ESRD)
       T1D119 (43%)83 [75–89]17 [12–24]7986
       T2D79 (36%)71 [59–81]15 [10–22]7281
      ACR, albumin-to-creatinine ratio; CART, classification and regression tree; CI, confidence interval eGFR, estimated glomerular filtration rate; ESRD, end-stage renal disease; TNFR, tumor necrosis factor receptor; T1D, type 1 diabetes; T2D, type 2 diabetes.
      a A composite end point developed in 134 of 279 T1D patients and in 88 of 221 T2D patients.
      b The following markers measured at baseline were included: sex, age, HbA1c, eGFR, ACR, and serum TNFR1, and TNFR2.

       CART analysis to develop optimal multimarker criterion to identify patients at a high risk of ESRD

      To determine whether higher prognostic values and sensitivity could be achieved by combining the baseline clinical characteristics, legacy markers, and TNFR1 and TNFR2 into a multimarker criterion, we included all of them in the CART analysis of the data of the T1D cohort. The analysis found only 2 markers that were needed to develop the optimal criterion that identified patients at a high risk of ESRD: TNFR1 with a cut-point of >4.3 ng/ml or if TNFR1 was between 2.4 and 4.3 ng/ml, ACR with a cut-point of >3100 mg/g were at a high risk. This criterion’s predictive value was 85% and sensitivity was 72%. When applied to the T2D cohort, the validation set, the values were 71% and 57%, respectively.
      The CART analysis provided a simple enrollment criterion on the basis of cut-points for just 2 markers, and because results in both the T1D and T2D cohorts were very similar, we re-fitted the diagnostic tree to the combined cohorts (Figure 1) to determine a final clinical trial enrollment criterion. Among all baseline clinical characteristics and marker values, the highest risk of an end point was observed in patients with TNFR1 of >4.3 ng/ml. This cut-point identified 104 cases. A similar high risk was observed in patients with TNFR1 between 2.9 and 4.3 ng/ml and ACR of >1900 mg/g. However, these cut-points identified only 15 cases. These 2 groups were combined as the high-risk group with a risk of end points (prognostic value) of 83% and 76% in T1D and T2D, respectively, and with sensitivity of 74% and 68%, respectively (Table 2).
      Figure thumbnail gr1
      Figure 1Prognostic cut-points derived from the multimarker analysis of the combined type 1 diabetes (T1D) and type 2 diabetes (T2D) cohorts. Plots of the cumulative risk of a study outcome within 3 years in T1D patients (solid line) and T2D patients (interrupted line). Sex, age, systolic blood pressure, hemoglobin A1c, estimated glomerular filtration rate, and tumor necrosis factor receptor (TNFR) 2 were included in the classification and regression tree analysis, but none surpassed the outcome separation achieved by TNFR1 and albumin-to-creatinine ratio (ACR). Letters (a), (b), and (c) in each plot refer to the corresponding areas in .
      Patients in the combined T1D and T2D cohorts are represented by symbols in Figure 2 according to their serum concentrations of TNFR1 and ACR at baseline. The dots are color coded according to the outcome at the 3-year follow-up. Note the overlap of the distribution of red and blue dots representing ESRD and 40% eGFR loss. By moving the cut-points for TNFR1 and ACR around in Figure 2, an investigator can visualize the trade-off between the sensitivity and predictive value of alternative choices. For example, the highest density of renal outcomes (red and blue dots) lies above the serum TNFR1 concentration of >4.3 ng/ml (incidence rate, 40.5/100 patient-years). An intermediate density lies between serum TNFR1 concentrations of 2.9 and 4.3 ng/ml (incidence rate, 13.4/100 patient-years). However, this subgroup can be divided further into patients with a high risk (incidence rate, 35.1/100 patient-years) if ACR was above 1900 mg/g and those with a low risk (incidence rate, 9.5/100 patient-years) if ACR was below this cut-point. Patients with serum TNFR1 concentration of ≤2.9 ng/ml have a very low 3-year risk of ESRD (incidence rate, 5.5/100 patient-years) regardless of baseline concentrations of ACR. In Figure 2, deaths unrelated to ESRD appear to be unrelated to baseline TNFR1 or ACR and when removed from the composite end point, the enrollment criterion’s predictive value is unchanged, but it gains sensitivity (see Table 2).
      Figure thumbnail gr2
      Figure 2Distribution of patient outcomes in the combined study group (type 1 diabetes [T1D] and type 2 diabetes [T2D]) according to baseline values of tumor necrosis factor receptor (TNFR) 1 and albumin-to-creatinine ratio (ACR). Outcomes during the 3-year follow-up are color coded according to type: end-stage renal disease (ESRD) (red), 40% baseline estimated glomerular filtration rate (eGFR) loss (blue), deaths unrelated to ESRD (black ×), or no event (none) (open circle). The cut-points for TNFR1 (2.9 ng/ml and 4.3 ng/ml) and ACR (1900 mg/g) shown in are represented by interrupted horizontal and vertical lines, respectively. The dots within the areas labeled (a), (b), and (c) represent the patients in plots with the corresponding labels. The gray area indicates high risk, whereas the white area indicates a low risk of the outcome. Note that the distributions of blue and red dots closely overlap.

       Effect of optimal multimarker enrollment criterion on statistical power of a clinical trial

      To illustrate how this new optimal enrollment criterion could influence the design of future clinical trials and improve the ability to detect important treatment effects, we considered a study population of patients with diabetes and CKD stage 3 or 4.
      Suppose 1410 patients with diabetes and CKD stage 3 or 4 were identified in medical records. Based on trials using the previous enrollment criteria, 282 patients (20%) are expected to reach an end point and 1128 are not. In such a trial cohort, one would have 50% power to detect a 20% risk reduction of ESRD in the treatment arm (randomized 1:1). If the 1410 patients were invited for screening on the basis of the new enrollment criterion described here, only a subset would be selected for enrollment, and yet the statistical power would increase. For example, the new enrollment criterion with 70% sensitivity and 80% positive prognostic value would select only 246 of the 1410 patients for enrollment in the trial (123 patients for each arm). In this smaller group, 80% are expected to reach an end-point, so a 20% reduction from 80% to 64% (difference of 16%) is more easily detected than that from 20% to 16% (difference of 4%). The power of the trial to detect the 16% difference (between 80% and 64%) would be 86% and not 50%, despite the considerably reduced study size. The influence of an enrollment criterion’s sensitivity and prognostic value on the study group size and statistical power within this hypothetical trial are summarized in Supplementary Figure S1 and Table S2 (Supplementary Material, part B).

       Patients at risk of ESRD not recognized by our multimarker criterion

      Patients who met our multimarker criterion of high risk of ESRD within 3 years comprised 43% of the T1D cohort and captured 74% (sensitivity) of those at a risk for this outcome. Similarly, they comprised 36% of the T2D cohort and included 68% (sensitivity) of those at a risk for the outcome (Table 2). All of these patients had a fast renal decline to ESRD.
      • Krolewski A.S.
      • Skupien J.
      • Rossing P.
      • et al.
      Fast renal decline to end-stage renal disease: an unrecognized feature of nephropathy in diabetes.
      Moderate decliners are not captured by this criterion, although they experienced a risk of ESRD of approximately 50% within 10 years (Figure 3a and b ). Our study had 80 moderate decliners in the T1D cohort and 71 in the T2D cohort. These patients comprised 29% and 32% of the T1D and T2D cohorts, respectively, and while they would have been uninformative participants in a 3-year trial, their clinical course may be more susceptible to beneficial intervention than the very fast and fast decliners eligible for current US Food and Drug Administration-approved trials.
      Figure thumbnail gr3
      Figure 3Cumulative incidence of end-stage renal disease (ESRD) (interrupted line) and ESRD and deaths (solid line) in patients who are not eligible for a clinical trial according to the multimarker prognostic criterion. (a) Type 1 diabetes (T1D) patients (n = 160) and (b) type 2 diabetes (T2D) patients (n = 142). While the incidence of ESRD and mortality was low within the first 3 years, by the 10th year of follow-up, it reached 45% and 63% in T1D and T2D patients, respectively. Overall, during the 10 years of follow-up, 48 ESRD cases occurred in the T1D cohort and 38 in the T2D cohort that were not eligible by the multimarker criterion.

      Discussion

      For patients with diabetes, the risk of ESRD remains high, and any new intervention to reduce it must be tested by a clinical trial. The enrollment criterion, determined by the CART analysis of 4 clinical characteristics (sex, age, systolic blood pressure [BP], and hemoglobin A1c [HbA1c]) and 4 markers (eGFR, ACR, TNFR1, and TNFR2), enriches the trial group with patients (fast decliners) having a high risk of progression to ESRD. Two markers determine the optimal criterion: serum TNFR1 concentration of >4.3 ng/ml, which identified most cases, and if serum TNFR1 concentration was between >2.9 and 4.3 ng/ml, only patients with ACR of >1900 mg/g were at a high risk of the outcome. This composite criterion performs similarly well in the T1D and T2D cohorts. Because the performance of serum TNFR2 was similar to that of serum TNFR1, it is not presented.
      We hope that designers of future clinical trials will give more serious consideration to the prognostic values of their enrollment criteria than that given in the past. For ESRD trials, an adaptation of our optimal criterion will permit smaller and more efficient trials, as elaborated in our results section. Recruitment of patients for diabetic nephropathy trials would be facilitated if serum TNFR1 concentrations were measured in all patients with CKD stages 3 and 4 to create registries of patients eligible for clinical trials. Patients with a serum TNFR1 concentration of >4.3 ng/ml would be good candidates. Note that this TNFR1 concentration indicates a high risk of ESRD even in patients with normoalbuminuria (see Figure 2). Additional high-risk patients could be found among those with serum TNFR1 concentrations between >2.9 and <4.3 ng/ml if ACR is also high.
      Our enrollment criterion predicts a much higher ESRD risk than recent predictive models of kidney failure.
      • Tangri N.
      • Stevens L.A.
      • Griffith J.
      • et al.
      A predictive model for progression of chronic kidney disease to kidney failure.
      • Elley C.R.
      • Robinson T.
      • Moyes S.A.
      • et al.
      Derivation and validation of a renal risk score for people with type 2 diabetes.
      • Jardine M.J.
      • Hata J.
      • Woodward M.
      • et al.
      Prediction of kidney-related outcomes in patients with type 2 diabetes.
      Those models were based on traditional renal markers, such as eGFR and ACR, and several clinical characteristics, such as sex, age, BP, ethnicity, diabetes duration, or HbA1c. Excellent performance was reported for the models, with the C-statistic reaching 0.9. Unfortunately, high C-statistics are misleading. The high values are because of the correct classification of low-risk patients, who were overwhelmingly predominant in the cohorts used for derivation and validation. The models do not efficiently capture the patients at the highest risk. Unlike our criterion, which selects a group in which the 3-year ESRD risk may exceed 70% to 80%, the models (assuming a reasonable range of inclusion characteristics) predict a 5-year ESRD risk of 15% to 30% that reaches 40% to 50% if patients have serum ACR and eGFR concentrations of 2000 mg/g and 30 ml/min per 1.73 m2, respectively. Once again, the utility of an inclusion criterion should be judged on the basis of its positive predictive value (also see Supplementary Material, part D).
      While our enrollment criterion identifies the very fast and fast renal decliners and would serve the needs of 3-year trials using the approved composite end point, it excludes patients who are moderate decliners and need 10 years to reach ESRD. These patients comprised 30% of all patients who were followed up. If efforts are directed toward developing an enrollment criterion with a high positive predictive value and sensitivity for the subset of moderate decliners, it could be used in a screened population in tandem with enrollment into a 3-year trial. Those who did not qualify for a 3-year trial could be evaluated for enrollment in a trial designed for moderate decliners.
      • Krolewski A.S.
      • Skupien J.
      • Rossing P.
      • et al.
      Fast renal decline to end-stage renal disease: an unrecognized feature of nephropathy in diabetes.
      Before a trial can be initiated for these patients, a novel end point based on changes in the eGFR slope must be approved. One example of the use of such a design is the ongoing Preventing Early Renal Loss trial (http://www.perl-study.org/) that tests the effectiveness of allopurinol in reducing renal function decline as determined by the eGFR slope.

      Doria A, et al. PERL (Preventing Early Renal Loss in Diabetes): A Multicenter Clinical Trial of Allopurinol to Prevent Kidney Function Loss in Type 1 Diabetes. ClinicalTrials.gov Identifier: NCT02017171.

      Unfortunately, while a study design based on the eGFR slope is acceptable for allopurinol (an already approved drug), it is not approved by the US Food and Drug Administration for new drugs or interventions.
      Our study has several strengths that should be emphasized. The cohorts used in the study comprised patients with diabetes who had CKD stages 3 and 4 while under long-term care at the Joslin clinic. Reliable measurements of the examined markers were performed at enrollment, and the cohorts were treated according to contemporary standards of diabetes and CKD care. They were followed up for 4 to 15 years with complete ascertainment of relevant end points. These features of the study allowed us to generalize our findings to patients with diabetes in the general population. The fact that the frequency of patients at a high risk of ESRD was much lower in past trials (Supplementary Table S1, Supplementary Material, part A) than that in the Joslin cohorts suggests that past recruitment efforts were biased toward nonprogressing patients despite their impaired renal function. Applying the enrollment criterion that we developed may help to ameliorate this tendency. Finally, the CART analysis appears to be a very robust analytical tool for developing prognostic tests, as illustrated by the excellent replication in the T2D validation cohort. Furthermore, the enrollment criterion developed based on the CART analysis is much simpler and performs slightly better than that based on an alternative approach, a logistic regression score (see Supplementary Material, part C). Advantages of a CART approach over that based on logistic modeling have been reviewed.
      • Henrard S.
      • Speybroeck N.
      • Hermans C.
      Classification and regression tree analysis vs. multivariable linear and logistic regression methods as statistical tools for studying hemophilia.
      Several important qualifications to the findings of our study need to be recognized. As demonstrated in our recent publication
      • Skupien J.
      • Warram J.H.
      • Smiles A.M.
      • Stanton R.C.
      • Krolewski A.S.
      Patterns of estimated glomerular filtration rate decline leading to end-stage renal disease in type 1 diabetes.
      and review article,
      • Krolewski A.S.
      • Skupien J.
      • Rossing P.
      • et al.
      Fast renal decline to end-stage renal disease: an unrecognized feature of nephropathy in diabetes.
      the majority of patients with diabetes who developed ESRD had fast renal decline that began when their renal function was normal. That decline was linear, and such patients needed 3 to 10 years to progress from normal renal function to ESRD. Interventions to prevent or delay ESRD need to be implemented early, that is, when renal function is normal and when sufficient time remains for the effect to be realized while patients have significant renal function remaining.
      • Krolewski A.S.
      • Skupien J.
      • Rossing P.
      • et al.
      Fast renal decline to end-stage renal disease: an unrecognized feature of nephropathy in diabetes.
      Unfortunately, no such intervention has been developed, and prognostic criteria to identify patients while renal function is normal are nonexistent.
      • Krolewski A.S.
      • Skupien J.
      • Rossing P.
      • et al.
      Fast renal decline to end-stage renal disease: an unrecognized feature of nephropathy in diabetes.
      Findings from our study are restricted to patients with diabetes who have CKD and impaired renal function. Our prognostic criterion identifies patients with very fast and fast renal decline, with manifestation of a hard end point within 3 years. Our study does not shed light on the design of the trials for moderate decliners—neither enrollment criteria nor end points. Therapies whose effects manifest after a lag interval will never be detected in short-term trials. The analytical plan for longer trials must anticipate such an outcome. This is more than a hypothetical consideration because we demonstrated in a study of T1D patients with proteinuria that a sustained period of improved glycemic control (3–5 years) significantly reduced the risk of ESRD but only during a 5- to 10-year follow-up interval.
      • Skupien J.
      • Warram J.H.
      • Smiles A.
      • et al.
      Improved glycemic control and risk of ESRD in patients with type 1 diabetes and proteinuria.
      Finally, two additional uncertainties of our findings should be considered. First, our criterion was developed in the Joslin cohorts that were relatively young, and their risk of death unrelated to ESRD was low. Future clinical trials may enroll much older T2D patients in whom mortality unrelated to ESRD is much higher. Application of our criterion for recruitment into such trials will prove insufficient in its ability to identify such patients. Fortunately, this will result in an increased sensitivity and prognostic value regarding enrichment of the trial population for patients at a risk of ESRD (see Table 2). Second, calibration and validation of our prognostic criterion may need to be separately evaluated in patients of other ethnic ancestry. For example, in a prospective cohort of Pima Indians with a comparable renal phenotype, we also observed robust contributions of TNFR concentrations with a similar threshold pattern for the progression to ESRD; nevertheless, the concentrations in that population were twice as high compared with those in Caucasian Joslin cohorts.
      • Pavkov M.E.
      • Nelson R.G.
      • Knowler W.C.
      • et al.
      Elevation of circulating TNF receptors 1 and 2 increases the risk of end-stage renal disease in American Indians with type 2 diabetes.
      These differences may be related to the ethnicity itself, a higher degree of obesity, or other factors that will need to be determined.

      Materials and Methods

      The study group comprised patients with diabetes who had CKD stages 3 or 4 when they enrolled in the studies conducted at the Joslin Diabetes Center. The Joslin T1D cohort with proteinuria was used as the exploratory or training panel, and the Joslin T2D cohort was used as the validation panel. Study protocols and informed consent procedures for the 2 studies were approved by the Institutional Review Board of the Joslin Diabetes Center.
      Description of the study groups, measurements of clinical characteristics, and determination of examined markers are described in the Supplementary Material, part C.
      • Krolewski A.S.
      • Laffel L.M.
      • Krolewski M.
      • et al.
      Glycosylated hemoglobin and the risk of microalbuminuria in patients with insulin-dependent diabetes mellitus.
      • Mueller P.M.
      • Rogus J.J.
      • Cleary P.A.
      • et al.
      The Genetics of Kidneys in Diabetes (GoKinD) Study: A genetics collection available for identifying the genetic susceptibility factors for diabetic nephropathy in type 1 diabetes mellitus.
      • Warram J.H.
      • Gearin G.
      • Laffel L.
      • Krolewski A.S.
      Effect of duration of type I diabetes on the prevalence of stages of diabetic nephropathy defined by urinary albumin/creatinine ratio.
      • Skupien J.
      • Warram J.H.
      • Smiles A.M.
      • et al.
      Early renal function decline predicts risk of ESRD: 5-18 year follow-up of patients with type 1 diabetes and proteinuria.
      • Levey A.S.
      • Stevens L.A.
      • Schmid C.H.
      • et al.
      A new equation to estimate glomerular filtration rate.

       Ascertainment of onset of ESRD, mortality, and time of 40% eGFR loss

      All patients enrolled in the studies conducted at the Joslin Diabetes Center who were included in this study were queried against rosters of the United States Renal Data System and the National Death Index that covered all events up to the end of 2012. United States Renal Data System maintains a roster of US patients who undergo renal replacement therapy, which includes the dates of dialysis and transplantation.
      • Agodoa L.Y.
      • Eggers P.W.
      Renal replacement therapy in the United States: data from the United States Renal Data System.
      The National Death Index is a comprehensive roster of deaths in the US, which includes the date and cause of death.

      Centers for Disease Control and Prevention, National Center for Health Statistics: Data Access—National Death Index. Available at: http://www.cdc.gov/nchs/ndi.htm. Accessed October 10, 2010.

      • Cowper D.C.
      • Kubal J.D.
      • Maynard C.
      • Hynes D.M.
      A primer and comparative review of major US mortality databases.
      ESRD was defined by a match with the United States Renal Data System roster or a listing of renal failure among the causes of death on an National Death Index death certificate. The date given for the onset of ESRD was that of the first dialysis or transplantation or that of death for those captured with respect to the death certificate. If a date of death was obtained from the National Death Index and ESRD had not developed, the outcome was defined as “death unrelated to ESRD.”
      If ESRD had not developed during the first 3 years of follow-up or death unrelated to ESRD occurred in that interval, the patient was evaluated for 40% eGFR loss. We used all available eGFR determinations of patients, which were performed during the first 5 years of follow-up, to estimate eGFR slopes using the general linear model. The slopes were projected against the follow-up time to determine if and when eGFR declined by 40% from the baseline eGFR.
      Because the study aimed to evaluate risk of ESRD during the first 3 years, all patients who developed ESRD or died or lost ≥40% of baseline eGFR after the 3-year period were considered alive as of the end of the third year of follow-up.

       Statistical analysis

      Characteristics of patients, including serum and urine marker concentrations, were summarized as medians (25th and 75th percentiles) or as counts and percentages. Medians were compared by Wilcoxon tests and proportions by χ2 or Fisher’s exact tests. The cumulative incidence of ESRD and deaths unrelated to ESRD was calculated under the assumption of competing risks between these 2 events using the cumulative incidence competing risk approach.
      • Caplan R.J.
      • Pajak T.F.
      • Cox J.D.
      Analysis of the probability and risk of cause-specific failure.
      In this search for an optimal enrollment criterion for future clinical trials for diabetic nephropathy to prevent ESRD, we considered patients in the study cohorts who within 3 years had any of the following criteria: (i) the onset of ESRD, (ii) a ≥40% eGFR loss, or (iii) death. Our analytical approach was an implementation of the CART method available
      • Breiman L.
      • Friedman J.H.
      • Stone C.J.
      • Olshen R.A.
      Classification and Regression Trees.
      in the “rpart” package of the R software. The method involved hierarchical partitioning of the study group on the basis of optimal cut-points selected from all possible cut-points in the distribution of each marker. An optimal cut-point is the one producing the widest separation in the frequency of cases in the 2 daughter groups. Partitioning was optimized first within the total sample and then separately within the daughters. The process was repeated until stopping criteria were met.
      Each resulting set of cut-points is a candidate for a set of enrollment criteria and was evaluated in a manner similar to a diagnostic test. Patients in the high-risk ESRD group, identified by the set of cut-points, were the “test positives,” and all others were “test negatives.” Our focus was primarily on the predictive value of the criteria, that is, the proportion of test-positive patients who became cases measured as the cumulative incidence of ESRD by the end of the 3rd year of follow-up. The secondary consideration was the set’s sensitivity; that is, the proportion of cases who were test positive. For planning a clinical trial with a specified power, these 2 characteristics of a set of enrollment criteria (sensitivity and positive predictive value) determine how many test-positive patients must be identified by screening the source population (i.e., patients with proteinuria and CKD stage 3 or 4).
      For recursive partitioning, we assumed standard default parameters. The minimum size of terminal nodes was set at 5 patients, and a partition was attempted in a node with at least 20 individuals. For simplicity of the final eligibility rule, we additionally limited the tree size by removing cut-points where both daughter nodes provided the same binary event prediction, that is, the risk of the outcome was either >50% or <50% in both.
      We applied recursive partitioning to the T1D cohort as the training or discovery set, and we used all 4 clinical characteristics (sex, age, systolic BP, and HbA1c) and 4 candidate markers (ACR, eGFR, TNFR1, and TNFR2) to build a multimarker tree. To test the replicability of the findings, we applied the cut-points identified in the T1D cohort to the T2D cohort and calculated the proportion of cases in the terminal nodes to compare the predictions the in T1D and T2D cohorts. Finally, on the basis of good replication of our findings, we reestimated the multimarker decision rule in the T1D and T2D cohorts pooled together.
      As an alternative, we used a logistic regression model to derive a score calculated from markers weighted by regression parameters. To compare the performance of CART and logistic regression analyses, we selected a cutoff in the score, which identified as trial eligible (a high-risk group) the same number of T1D patients as the CART method (see the results presented in Supplementary Material, part C).
      Comparisons of our approach with other methods of characterizing prognostic tests
      • Kerr K.F.
      • Wang Z.
      • Janes H.
      • et al.
      Net reclassification indices for evaluating risk prediction instruments: a critical review.
      are discussed in Supplementary Material, part D.

      Disclosures

      ASK and MAN are co-inventors of the TNFR1 and TNFR2 patent for predicting the risk of ESRD. This patent was licensed by the Joslin Diabetes Center to EKF Diagnostics. All the other authors declared no competing interests.

      Author Contributions

      MY contributed to the study design, performed data analysis, and drafted the manuscript; JS contributed to the study design, performed data analysis, and drafted the manuscript; MAN contributed to research data collection and reviewed and edited the manuscript; AMS contributed to data collection and was responsible for data management and contributed to data analysis; AD, RSC, ATG, KLD, NP, MDB, and JVB all contributed intellectually to the final plans of data analysis, final interpretation of the results, and drafting/editing the manuscript; JHW contributed to the study design, performed data analysis, and drafted and edited the manuscript; and ASK was responsible for the study design, supervised data collection and data analysis, and contributed to drafting and editing of the manuscript.

      Acknowledgments

      This study was supported by the JDRF grant No. 3-SRA-2015-106-Q-R for subproject “Predictors of progressive renal decline in Type 1 diabetes” to ASK and JVB; by the National Institutes of Health grants DK-041526 to ASK, and DK-072381 to JVB; by Novo Nordisk Foundation grant NNF14OC0013659 (PROTON) to ASK; by the NIH DRC grant P30 DK03683 ; by the JDRF grant 5-CDA-2015-89-A-B to MAN; and by grants from Lilly Inc. and Pfizer Inc. to ASK to study biomarkers of renal decline in T2D.

      Supplementary Material

      • Part A, Table S1

        Characteristics of clinical trials conducted with patients having diabetic nephropathy with emphasis on the risk of end-stage renal disease (ESRD) or the composite outcome and mortality in the placebo subgroups of the trials.

      • Part B, Figure S1

        Hypothetical clinical trial with 705 patients per arm and designed to have 50% power to detect a 20% reduction in the 20% 3-year risk of end-stage renal disease (ESRD). Graph showing the effect on statistical power of the application of an enrollment criterion to the study group of 1410 patients according to sensitivity and several positive prognostic values.

      • Part B, Table S2

        Hypothetical clinical trial with 705 patients per arm and designed to have 50% power to detect a 20% reduction in the 20% 3-year risk of end-stage renal disease (ESRD). Table shows the study group sizes that result from the application of an enrollment criterion to the study group of 1410 patients according to several sensitivities and positive prognostic values. Note that all combinations of sensitivity and positive prognostic value shown would have greater than 50% power (see Figure S1 in Supplemental Material part B).

      References

        • Rosolowsky E.T.
        • Skupien J.
        • Smiles A.M.
        • et al.
        Risk for ESRD in type 1 diabetes remains high despite renoprotection.
        J Am Soc Nephrol. 2011; 22: 545-553
        • De Boer I.H.
        • Rue T.C.
        • Hall Y.N.
        • et al.
        Temporal trends in the prevalence of diabetic kidney disease in the United States.
        JAMA. 2011; 305: 2532-2539
        • Lewis E.J.
        • Hunsicker L.G.
        • Bain R.P.
        • Rohde R.D.
        The effect of angiotensin-converting-enzyme inhibition on diabetic nephropathy. The Collaborative Study Group.
        N Engl J Med. 1993; 329: 1456-1462
        • Lewis E.J.
        • Hunsicker L.G.
        • Clarke W.R.
        • et al.
        Renoprotective effect of the angiotensin-receptor antagonist irbesartan in patients with nephropathy due to type 2 diabetes.
        N Engl J Med. 2001; 345: 851-860
        • Brenner B.M.
        • Cooper M.E.
        • de Zeeuw D.
        • et al.
        Effects of losartan on renal and cardiovascular outcomes in patients with type 2 diabetes and nephropathy.
        N Engl J Med. 2001; 345: 861-869
        • Facchini F.S.
        • Saylor K.L.
        A low-iron-available, polyphenol-enriched, carbohydrate-restricted diet to slow progression of diabetic nephropathy.
        Diabetes. 2003; 52: 1204-1209
        • House A.A.
        • Eliasziw M.
        • Cattran D.C.
        • et al.
        Effect of B-vitamin therapy on progression of diabetic nephropathy: a randomized controlled trial.
        JAMA. 2010; 303: 1603-1609
        • Heerspink H.J.
        • Ninomiya T.
        • Perkovic V.
        • et al.
        Effects of a fixed combination of perindopril and indapamide in patients with type 2 diabetes and chronic kidney disease.
        Eur Heart J. 2010; 31: 2888-2896
        • Sharma K.
        • Ix J.H.
        • Mathew A.V.
        • et al.
        Pirfenidone for diabetic nephropathy.
        J Am Soc Nephrol. 2011; 22: 1144-1151
        • Imai E.
        • Chan J.C.
        • Ito S.
        • et al.
        Effects of olmesartan on renal and cardiovascular outcomes in type 2 diabetes with overt nephropathy: a multicentre, randomised, placebo-controlled study.
        Diabetologia. 2011; 54: 2978-2986
        • Parving H.H.
        • Brenner B.M.
        • McMurray J.J.
        • et al.
        Cardiorenal end points in a trial of aliskiren for type 2 diabetes.
        N Engl J Med. 2012; 367: 2204-2213
        • De Zeeuw D.
        • Akizawa T.
        • Audhya P.
        • et al.
        Bardoxolone methyl in type 2 diabetes and stage 4 chronic kidney disease.
        N Engl J Med. 2013; 369: 2492-2503
        • Navarro-González J.F.
        • Mora-Fernández C.
        • Muros de Fuentes M.
        • et al.
        Effect of pentoxifylline on renal function and urinary albumin excretion in patients with diabetic kidney disease: the PREDIAN trial.
        J Am Soc Nephrol. 2015; 26: 220-229
        • Forsblom C.
        • Harjutsalo V.
        • Thorn L.M.
        • et al.
        Competing-risk analysis of ESRD and deaths among patients with Type 1 diabetes and macroalbuminuria.
        J Am Soc Nephrol. 2011; 22: 537-544
        • Adler A.I.
        • Stevens R.J.
        • Manley S.E.
        • et al.
        Development and progression of nephropathy in type 2 diabetes: the United Kingdom Prospective Diabetes Study (UKPDS 64).
        Kidney Int. 2003; 63: 225-232
        • Niewczas M.A.
        • Gohda T.
        • Skupien J.
        • et al.
        Circulating TNF receptors 1 and 2 predict ESRD in type 2 diabetes.
        J Am Soc Nephrol. 2012; 23: 507-515
        • Skupien J.
        • Warram J.H.
        • Niewczas M.A.
        • et al.
        Synergism between circulating tumor necrosis factor receptor 2 and HbA(1c) in determining renal decline during 5-18 years of follow-up in patients with type 1 diabetes and proteinuria.
        Diabetes Care. 2014; 37: 2601-2608
        • Gohda T.
        • Niewczas M.A.
        • Ficociello L.H.
        • et al.
        Circulating TNF receptors 1 and 2 predict stage 3 CKD in type 1 diabetes.
        J Am Soc Nephrol. 2012; 23: 516-524
        • Krolewski A.
        • Niewczas M.A.
        • Skupien J.
        • et al.
        Early progressive renal decline precedes the onset of microalbuminuria and its progression to macroalbuminuria.
        Diabetes Care. 2014; 37: 226-234
        • Lopes-Virella M.F.
        • Baker N.L.
        • Hunt K.J.
        • et al.
        Response to comment on Lopes-Virella, et al. Baseline markers of inflammation are associated with progression to macroalbuminuria in type 1 diabetic subjects. Diabetes Care 2013;36:2317-2323.
        Diabetes Care. 2014; 37: e108-e109
        • Saulnier P.J.
        • Gand E.
        • Ragot S.
        • et al.
        Association of serum concentration of TNFR1 with all-cause mortality in patients with type 2 diabetes and chronic kidney disease: follow-up of the SURDIAGENE Cohort.
        Diabetes Care. 2014; 37: 1425-1431
        • Forsblom C.
        • Moran J.
        • Harjutsalo V.
        • et al.
        Added value of soluble tumor necrosis factor-alpha receptor 1 as a biomarker of ESRD risk in patients with type 1 diabetes.
        Diabetes Care. 2014; 37: 2334-2342
        • Pavkov M.E.
        • Nelson R.G.
        • Knowler W.C.
        • et al.
        Elevation of circulating TNF receptors 1 and 2 increases the risk of end-stage renal disease in American Indians with type 2 diabetes.
        Kidney Int. 2015; 87: 812-819
        • Breiman L.
        • Friedman J.H.
        • Stone C.J.
        • Olshen R.A.
        Classification and Regression Trees.
        Wadsworth, Belmont, CA1984
        • Levey A.S.
        • Inker L.A.
        • Matsushita K.
        • et al.
        GFR decline as an end point for clinical trials in CKD: a scientific workshop sponsored by the National Kidney Foundation and the US Food and Drug Administration.
        Am J Kidney Dis. 2014; 64: 821-835
        • Krolewski A.S.
        • Skupien J.
        • Rossing P.
        • et al.
        Fast renal decline to end-stage renal disease: an unrecognized feature of nephropathy in diabetes.
        Kidney Int. 2017; 91: 1300-1311
        • Tangri N.
        • Stevens L.A.
        • Griffith J.
        • et al.
        A predictive model for progression of chronic kidney disease to kidney failure.
        JAMA. 2011; 305: 1553-1559
        • Elley C.R.
        • Robinson T.
        • Moyes S.A.
        • et al.
        Derivation and validation of a renal risk score for people with type 2 diabetes.
        Diabetes Care. 2013; 36: 3113-3120
        • Jardine M.J.
        • Hata J.
        • Woodward M.
        • et al.
        Prediction of kidney-related outcomes in patients with type 2 diabetes.
        Am J Kidney Dis. 2012; 60: 770-778
      1. Doria A, et al. PERL (Preventing Early Renal Loss in Diabetes): A Multicenter Clinical Trial of Allopurinol to Prevent Kidney Function Loss in Type 1 Diabetes. ClinicalTrials.gov Identifier: NCT02017171.

        • Henrard S.
        • Speybroeck N.
        • Hermans C.
        Classification and regression tree analysis vs. multivariable linear and logistic regression methods as statistical tools for studying hemophilia.
        Haemophilia. 2015; 21: 713-722
        • Skupien J.
        • Warram J.H.
        • Smiles A.M.
        • Stanton R.C.
        • Krolewski A.S.
        Patterns of estimated glomerular filtration rate decline leading to end-stage renal disease in type 1 diabetes.
        Diabetes Care. 2016; 39: 2262-2269
        • Skupien J.
        • Warram J.H.
        • Smiles A.
        • et al.
        Improved glycemic control and risk of ESRD in patients with type 1 diabetes and proteinuria.
        J Am Soc Nephrol. 2014; 25: 2916-2925
        • Krolewski A.S.
        • Laffel L.M.
        • Krolewski M.
        • et al.
        Glycosylated hemoglobin and the risk of microalbuminuria in patients with insulin-dependent diabetes mellitus.
        N Engl J Med. 1995; 332: 1251-1255
        • Mueller P.M.
        • Rogus J.J.
        • Cleary P.A.
        • et al.
        The Genetics of Kidneys in Diabetes (GoKinD) Study: A genetics collection available for identifying the genetic susceptibility factors for diabetic nephropathy in type 1 diabetes mellitus.
        J Am Soc Neph. 2006; 17: 1782-1790
        • Warram J.H.
        • Gearin G.
        • Laffel L.
        • Krolewski A.S.
        Effect of duration of type I diabetes on the prevalence of stages of diabetic nephropathy defined by urinary albumin/creatinine ratio.
        J Am Soc Nephrol. 1996; 7: 930-937
        • Skupien J.
        • Warram J.H.
        • Smiles A.M.
        • et al.
        Early renal function decline predicts risk of ESRD: 5-18 year follow-up of patients with type 1 diabetes and proteinuria.
        Kidney Int. 2012; 82: 589-597
        • Levey A.S.
        • Stevens L.A.
        • Schmid C.H.
        • et al.
        A new equation to estimate glomerular filtration rate.
        Ann Intern Med. 2009; 150: 604-612
        • Agodoa L.Y.
        • Eggers P.W.
        Renal replacement therapy in the United States: data from the United States Renal Data System.
        Am J Kidney Dis. 1995; 25: 119-133
      2. Centers for Disease Control and Prevention, National Center for Health Statistics: Data Access—National Death Index. Available at: http://www.cdc.gov/nchs/ndi.htm. Accessed October 10, 2010.

        • Cowper D.C.
        • Kubal J.D.
        • Maynard C.
        • Hynes D.M.
        A primer and comparative review of major US mortality databases.
        Ann Epidemiol. 2002; 12: 462-468
        • Caplan R.J.
        • Pajak T.F.
        • Cox J.D.
        Analysis of the probability and risk of cause-specific failure.
        Int J Radiat Oncol Biol Phys. 1994; 29: 1183-1186
        • Kerr K.F.
        • Wang Z.
        • Janes H.
        • et al.
        Net reclassification indices for evaluating risk prediction instruments: a critical review.
        Epidemiology. 2014; 25: 114-121