Patterns of Estimated GlomerularFiltration Rate Decline Leadingto End-Stage Renal Diseasein Type 1 DiabetesDiabetes Care 2016;39:2262–2269 | DOI: 10.2337/dc16-0950

OBJECTIVE

Thepatterns of estimated glomerularfiltration rate (eGFR) decline to end-stage renaldisease (ESRD) in patients with type 1 diabetes has not been conclusively described.Decline could be linearly progressive to ESRD but with a variable rate. Conversely,decline may be linear but interrupted by periods of plateaus or improvements.

RESEARCH DESIGN AND METHODS

This observational study included 364 patients with type 1 diabetes attending theJoslin Clinic who developed ESRD between 1991 and 2013. We retrieved serumcreatininemeasurements from clinic visits or research examinations up to 24 years(median 6.7 years) preceding the onset of ESRD. Using serial measurements ofserum creatinine to estimate renal function (eGFR), we used regression-basedspline methods and a data smoothing approach to characterize individual trajec-tories of eGFR over time for the 257 patients with five or more data points.

RESULTS

The rate of eGFR decline per year rangedwidely, from272 to22mL/min/1.73m2

(median 28.5). The trajectories, as characterized with linear regression-basedspline models, were linear or nearly so for 87% of patients, accelerating for 6%,and decelerating for 7%. Smoothed trajectories evaluated by a Bayesian approachdid not significantly depart from a linear fit in 76%.

CONCLUSIONS

The decline of eGFR in type 1 diabetes is predominantly linear. Deviations fromlinearity are small, with little effect on the expected time of ESRD. A single diseaseprocess most likely underlies renal decline from its initiation and continues withthe same intensity to ESRD. Linearity of renal decline suggests using slope re-duction as the measure of effectiveness of interventions to postpone ESRD.

End-stage renal disease (ESRD) and associatedmorbidity andmortality aremajor healthproblems in patients with type 1 diabetes. The lifetime risk of ESRD in type 1 diabetes is10–15% (1–3). ESRD developsmany decades after the disease onset as a result of renaldecline (3–5). The decline trajectory and underlying mechanisms are not well under-stood. Theoretically, the decline may be abrupt, as in acute renal failure (6), episodicwith plateausor periods of improvement (7), or inexorably progressive (8). Eachpatternpoints to a different pathophysiology. Knowledge of the patterns and their frequenciescan guide research of the disease mechanisms and help develop new interventions.

1Research Division, Joslin Diabetes Center, Bos-ton, MA2Department of Medicine, Harvard MedicalSchool, Boston, MA3Department ofMetabolic Diseases, JagiellonianUniversity Medical College, Krakow, Poland4Renal Division at Beth Israel Deaconess MedicalCenter, Harvard Medical School, Boston, MA

Corresponding author: Andrzej S. Krolewski,andrzej.krolewski@joslin.harvard.edu, or JanSkupien, jan.skupien@uj.edu.pl.

Received 2 May 2016 and accepted 29 August2016.

© 2016 by the American Diabetes Association.Readers may use this article as long as the workis properly cited, the use is educational and notfor profit, and the work is not altered. More infor-mation is available at http://www.diabetesjournals.org/content/license.

Jan Skupien,1,2,3 James H. Warram,1,2

Adam M. Smiles,1,2 Robert C. Stanton,1,2,4

and Andrzej S. Krolewski1,2

2262 Diabetes Care Volume 39, December 2016

PATH

OPHYS

IOLO

GY/COMPLICATIONS

Linearity of renal decline in variouskidney conditions was already postu-lated four decades ago (9). Plots of se-rum creatinine reciprocal were used as apredictive tool for ESRD onset (8,10).Several studies have since examined tra-jectories of renal function changes, withfew patients reaching ESRD. Some stud-ies concluded that trajectories were lin-ear (8,11–13), whereas others foundsignificant variation and nonlinearity(7,14,15). These studies were small(7,8,11–14), included many patientswith advanced renal failure, often hadshort follow-up time, and did not useanalytical tools that are now availablefor evaluating this question.In our recent large study of patients

with type 1 diabetes with normal renalfunction and proteinuria, estimated glo-merular filtration rate (eGFR) declinewas present in only one-half. Declinewas predominantly linear within an in-dividual, but slopes differed widelyamong individuals (4). Most of the pa-tients in that study did not reach ESRD,making their trajectories incomplete.Patients who reached ESRD were toofew to draw unambiguous conclusionsand to rule out these ESRD events beingresults of acute kidney injury complicat-ing chronic kidney disease.Different findings were recently re-

ported from the African American Studyof Kidney Disease and Hypertension(AASK) (16). That study included 846 pa-tients with at least 3 years of follow-up(median 9 years) and$8 GFR estimates.Baseline eGFR was impaired (median 50mL/min/1.73 m2). The authors report-ed that more than 41% had nonlineareGFR trajectories. This finding contra-sts sharply with our results in patientswith type 1 diabetes and proteinuria(4). Differences in several factors of abiologic nature may have contributedto this discrepancy: 1) mechanisms under-lying hypertensive and diabetic kidneydiseases, 2) race (17), and 3) character-istics of study subjects and inclusioncriteria. For example, volunteers recruitedinto the AASK study had predomi-nantly slow rates of eGFR decline (18).However, substantially different an-alytical approaches and definitions oflinearity may account for the discrep-ant conclusions (4,19). We therefore ap-plied both methods in the current study:the method used in our previous publi-cation and the one used for analysis of

AASK study data, so the effect of the an-alytic approach can be assessed. Moreimportantly, in the current study we eval-uated, for the first time, complete trajec-tories of eGFR decline specifically inpatients with type 1 diabetes who devel-oped ESRD.

RESEARCH DESIGN AND METHODS

Joslin Clinic Population of PatientsWith Type 1 DiabetesA total of 3,500 adult patients withtype 1 diabetes diagnosed before age40 remain under the care of Joslin Clinic,an institution established in 1898 anddevoted to the treatment of diabetes(5). Most of these patients come to theclinic within 5 years after the diabetesdiagnosis and remain under care for along period, frequently for life. The cur-rent study included the 364 patientswho developed ESRD between 1991and 2013 and were previously enrolledinto one or more studies on the naturalhistory of diabetic nephropathy intype 1 diabetes: Joslin Proteinuria Co-hort (20), 1st Joslin Study on NaturalHistory of Microalbuminuria (21–24),2nd Joslin Study on Natural History ofEarly Renal Decline (25,26), and JoslinStudy on Genetics of Diabetic Nephrop-athy in type 1 diabetes (27,28). Patientswith a diagnosis of nondiabetic kidneydisease present in their medical recordswere excluded from enrollment in thesestudies.

Participants were monitored until2013 with the goal of obtaining bloodandurine specimens at least every 2 yearsand ascertaining onset of ESRD anddeaths. Research specimens were ob-tained during routine clinic visits or spe-cial home visits and stored at 280°C forfuture use. All clinical data and labora-tory determinations for the albumin-to-creatinine ratio (ACR), serum creatinine,and glycatedhemoglobinA1c (HbA1c) for all364 patients included in this study wereretrieved from Joslin medical records.

The Joslin Diabetes Center Institu-tional Review Board approved the studyprotocols and informed consent proce-dures for the above studies.

Assessment of Urinary Albumin andHbA1c

In the Joslin Clinic laboratory, albuminconcentrations in spot urines were mea-sured by immunonephelometry on aBN100 (Behring, Marburg, Germany)

with N Albumin kits. Creatinine measure-ments in urine were assayed by Jaffe’smodified picrate method. Urinary albu-min inmilligramswas expressed per gramof urinary creatinine (ACR) (29). HbA1cwasmeasured during routine clinic visits andwas standardized to Diabetes Controland Complications Trial units (30).

Laboratory Assessment of RenalFunctionRoutine measurements of serum creati-nine are performed for Joslin patients atJoslin Clinic laboratory at least once ayear. In 2011–2014, blood specimens ob-tained as part of research examinationswere sent to the Advanced Researchand Diagnostic Laboratory at Universityof Minnesota to measure serum concen-trations of creatinine using the Roche en-zymatic assay (prod. no. 11775685) ona Roche/Hitachi Mod P analyzer. Thismethod was calibrated to be traceableto an isotope dilutionmass spectrometryreference assay andwas verified bymea-suring National Institutes of Standardsand Technology Standard ReferenceMaterial No. 967. Whenever a routine visit(n = 1,133) coincided with a research ex-amination, we used duplicate measure-ments in samples from the same blooddraw to calibrate Joslin clinical measure-ments to the assay method used at theLaboratory at University of Minnesota(4,30). This calibration diminished thepossible effect of changes in laboratoryassay and equipment over time.

A total of 4,721 calibrated serum cre-atinine determinations from clinic visitsin 364 study patients were used to esti-mate eGFR with the Chronic Kidney Dis-ease Epidemiology Collaboration formula(31). For our goal of tracing the trajec-tory of eGFR decline at the beginning offollow-up (baseline) was dated from ear-liest evaluation of renal function (mea-surement of serum creatinine) while itwas$60 mL/min/1.73 m2 and with per-sistent proteinuria. In patients witheGFR ,60 mL/min/1.73 m2 or enteringthe follow-up without proteinuria, wetraced the decline retrospectively fromESRD up to the last observation in chronickidney disease (CKD) stage 1 to excludethe period of normal and stable renalfunction before the onset of decline.

Ascertainment of ESRD and DeathsAll patients enrolled into the Joslin Stud-ies described above were queried againstrosters of United States Renal Data System

care.diabetesjournals.org Skupien and Associates 2263

(USRDS) and National Death Index (NDI)covering all events up to the end of 2013.ESRD was defined by a match with theUSRDS roster (n = 301) or a listing of re-nal failure among the causes of death onanNDI death certificate (n = 63). Onset ofESRD was given the date of first dialysisor kidney transplantation or date ofdeath for those captured by death cer-tificate. USRDS and NDI data werescreened for disease codes indicatingnondiabetic causes of ESRD.

Statistical AnalysisContinuous variables are summarized asmedians and quartiles, and categoricalvariables are presented as numbersand percentages. Comparisons betweenmedians were done with the Wilcoxontest, and categorical variableswere com-pared with the x2 test.Linear trajectories were estimated

with least-squares regression. To char-acterize eGFR trajectories, as previouslyreported (4), we used an approach de-scribed by Jones and Molitoris (32) in257 patients with$5 eGFR observationsplaced at least 1 month apart. A splinefunction was fitted by least squares withposition of the knot determined by min-imizing the residual sum of the squares.Spline function was tested against the

linear model with a partial F test at asignificance level of 0.05. Nonlinear tra-jectories were classified as clinically con-sequential only if predicted time toESRD (eGFR = 10 mL/min/1.73 m2) bythe simple linear fit missed the observedtime to ESRD by$1 year; otherwise theywere classified as clinically inconse-quential. Analysis was performed inSAS 9.3 software (SAS Institute, Inc.,Cary, NC).

The Bayesian approach to modelingtrajectories in the 257 patients used pe-nalized thin-plate splines, as previouslydescribed (16,19). The analysis wasperformed with R 3.1.0 software (TheR Foundation for Statistical Comput-ing, Vienna, Austria) and OpenBUGS3.2.3 software (University of Helsinki,Helsinki, Finland). Knots were placed1 year apart in patients with $3 yearsof follow-up and every 0.2 year in pa-tients with ,3 years of follow-up. Weused noninformative and conjugate pri-ors. Starting values of fixed-effect param-eters were estimated with a generaladditive model using the SemiPar pack-age for R. We performed 700,000 itera-tions in 4 chains to assess convergence.We discarded the first 200,000 run-in it-erations and saved 1 of every 50 from theremaining 500,000 to obtain 10,000

posterior samples. A linear trajectorywas then fitted to the smoothed data,and the smoothed trajectory was clas-sified as clinically consequential non-linearity if any point-wise differencebetween the smooth and linear trajec-tories was$15mL/min/1.73 m2. Becausethe threshold of 15 mL/min/1.73 m2 wasan arbitrary choice, we performed a sen-sitivity analysis using the values of 5 and10 mL/min/1.73 m2.

RESULTS

Cohort CharacteristicsThe study group comprised 364 EasternMassachusetts residents, 186 men and178 women (48.9%) with type 1 diabe-tes. All were long-term (.11 years) pa-tients of Joslin Clinic (Boston, MA), and95% were Caucasian (self-reported).Characteristics are summarized in Table 1.Median age at baseline was 33 years, andmedian diabetes duration was 19 years.Diabetes was diagnosed in most patientsduring childhood or adolescence. Glyce-mic control was poor: median HbA1c forthe group was 9.5% (80.3 mmol/mol).Median ACR was 1,262 mg/g. Medianbaseline eGFR was 73 mL/min/1.73 m2.At time of enrollment into Joslin studies,median blood pressure was 134/80 mmHg.There were 66.2% patients treated with

Table 1—Characteristics of the study group according to availability of five observations

eGFR observations

All $5 ,5 P value

Characteristic N = 364 N = 257 N = 107 d

Women 178 (48.9) 128 (49.8) 50 (46.7) 0.59

At the time of earliest eGFR observationAge (years) 33 (27, 40) 34 (28, 40) 32 (26, 40) 0.42Diabetes duration (years) 19 (15, 26) 19 (16, 26) 18 (14, 24) 0.047Age of diabetes onset (years) 12 (8, 17) 12 (7, 17) 12 (8, 17) 0.50ACR (mg/g) 1,262 (526, 2,500) 1,299 (559, 2,500) 966 (471, 2,500) 0.18HbA1c (%) 9.5 (8.4, 10.8) 9.5 (8.4, 10.8) 9.6 (8.3, 11.0) 0.98HbA1c (mmol/mol) 80.3 (68.3, 94.5) 80.3 (68.3, 94.5) 81.4 (67.2, 96.7)eGFR (mL/min/1.73 m2) 73 (57, 97) 71 (56, 87) 84 (63, 109) 0.001

Recorded at the research enrollment examinationSystolic blood pressure (mmHg) 134 (125, 148) 135 (127, 150) 134 (120, 144) 0.21Diastolic blood pressure (mmHg) 80 (74, 88) 80 (74, 86) 82 (78, 88) 0.16Treatment with ACE-I or ARB 241 (66.2) 169 (65.8) 72 (67.3) 0.78SmokingCurrent 75 (20.6) 54 (21.0) 21 (19.6) 0.77Past 111 (30.5) 74 (28.8) 37 (34.6) 0.21

Follow-upFollow-up duration (years) 6.6 (4.2, 9.9) 6.7 (4.4, 9.8) 6.1 (3.7, 10.5) 0.22eGFR measurements 10 (4, 18) 15 (9, 24) 3 (2, 4)eGFR measurements per year 1.7 (0.6, 3.4) 2.7 (1.3, 3.9) 0.5 (0.3, 0.8)Death at ESRD onset 63 (17.3) 48 (18.7) 15 (14.0) 0.28Preemptive kidney transplant 17 (4.7) 13 (5.1) 4 (3.7) 0.59

Data are presented as n (%) or median (1st, 3rd quartile). P value for comparison between patients with ,5 and $5 eGFR observations.

2264 eGFR Decline in Type 1 Diabetes Diabetes Care Volume 39, December 2016

ACE inhibitors (ACE-I) or angiotensin re-ceptor blockers (ARB). This proportionreflects gradual entry of renoprotectivetreatment into clinical practice in 1990s.Observed time from baseline to the

initiation of renal replacement therapyvaried widely. Median (1st, 3rd quar-tiles) interval was 6.6 years (4.2, 9.9).Median number of eGFR observationsper patient was 10 with a median den-sity of 1.7 per year. Median (1st, 3rdquartile) individual-specific rate ofeGFR decline (estimated with linear re-gression) was29.1mL/min/1.73m2/year(25.8, 214.9), whereas the mean peryear was212.2 mL/min/1.73 m2, reflect-ing that all patients developed ESRD. Formost patients, renal replacement ther-apy began with dialysis, and mortality atESRD diagnosis was high, at 17.3%. Fewpatients (4.7%) received preemptivetransplants.The slope of eGFR decline was not

associated with age at study entry ordiabetes duration (adjusting for base-line eGFR). There was no associationwith the age at diabetes onset or anysignificant protective effect of prepu-bertal diabetes diagnosis. The distribu-tions of slopes were similar in men andwomen. A steep eGFR slope was asso-ciated with higher urinary ACR (P =0.017) and worse glycemic control(1.2 mL/min/1.73 m2/year steeper per1% HbA1c increase, P , 0.001,). Higherbaseline eGFR was associated withsteeper decline (P , 0.001), which re-flects the study design, because the pa-tients with slow eGFR decline werelikely to be recruited with an alreadylow eGFR. Systolic blood pressure wasnot associated with eGFR slope, butthere was a statistically significantdifference between slopes in patientstreated and not treated with an ACE-Ior ARB (median 27.2 vs. 211.9 mL/min/1.73 m2/year, P, 0.001). No associationof eGFR slope was found with to-bacco use.Because 5 observations are required

for tests of nonlinearity, patients withfewer eGFR observations could not beincluded in the analysis of trajectories.Clinical characteristics of the 257 patientswith$5 observations and the 107with,5were generally similar (Table 1). Exceptionswere steeper slopes of eGFR decline inthosewith,5observations (median211.7vs. 28.5 mL/min/1.73 m2/year, P ,0.001 (Fig. 1A and B), slightly shorter

diabetes duration (P = 0.047 (Table 1),and higher baseline eGFR (84 vs.71 mL/min/1.73 m2, P = 0.001) (Table1). Paucity of observations in these pa-tients was perhaps caused by rapidrates of eGFR decline.

For the 257 patients with $5 obser-vations, the median interval to ESRDwas 6.7 years, with a median of 15 eGFRobservations for a median density of2.7 observations per year. At the firstmeasurement of serum creatinine, dis-tribution according to CKD stage was55 at stage 1, 127 at stage 2, and 70 atstage 3.

Patterns of eGFR DeclineCriteria for classifying trajectories asnonlinear are summarized in Table 2.We used an approach described byJones and Molitoris (32) for fitting a

nonlinear splinemodel to an individual’seGFR data. Nonlinearity was subse-quently classified as clinically conse-quential if predicted time to ESRD(eGFR = 10 mL/min/1.73 m2) by simplelinear fit missed the observed time toESRD by 1 year or more. Clinically conse-quential nonlinear trajectories were fur-ther characterized as decelerating oraccelerating (Table 2).

The test of nonlinearity failed to re-ject the null hypothesis of linearity for124 trajectories. Furthermore, 101 non-linear trajectories failed to meet the cri-terion of clinical consequence. Thus,225 trajectories (87%) were satisfacto-rily characterized by a constant slope,and representation of eGFR declineas a single slope, as shown in Fig. 1A, isvalid for the vast majority of patients.The remaining 32 trajectories were

Figure 1—Distribution of eGFR decline slopes with $5 (A) and ,5 (B) eGFR observations.

care.diabetesjournals.org Skupien and Associates 2265

equally divided between 15 significantaccelerations (6%) and 17 significantdecelerations (7%). Examples are pre-sented in Fig. 2.

Apart from the previously appliedmethod of testing linearity with thespline model, we analyzed data with aBayesian technique previously used in

the analysis of data from the AASK(16,19). The same criteria of linearityapplied to our ESRD cohort (Table 2) re-sulted in rejection of the linear model inmost patients. A careful inspection ofeGFR decline trajectories of patientsdeclared nonlinear identified a largenumber of trajectories, for which thecriterion suggested in the AASK study(16) was clearly too stringent, as shownin the examples presented in Fig. 3. Inthe examples in Fig. 3C and D, the tra-jectory is apparently linear, but only aminority of the Bayesian trajectoriesconform to AASK study criteria. As analternative, we based the test of nonlin-earity on the magnitude of the largestdeparture of the smoothed trajectoryfrom a simple regression line fitted toit (Table 2). Provisionally, we chose adeparture $15 mL/min/1.73 m2 as thecriterion for nonlinearity and thenperformed a sensitivity analysis usingvalues of 5 and 10 mL/min/1.73 m2. Ac-cording to the15mL/min/1.73m2criterion,overall prevalence of linear trajectories inposterior Bayesian samples was 76.0%,and 5 and 10 mL/min/1.73 m2 departuresyielded 42.2% and 62.7% of linear tra-jectories, respectively.

The linear spline approach classified13% of the trajectories as sufficientlynonlinear to have a clinically consequen-tial effect on the anticipated time toESRD. The Bayesian approach classified24% of the posterior samples as nonlin-ear because they departed by at least15 mL/min/1.73 m2 from a linear trajec-tory. To determine concordance of clas-sifications by the two approaches, we

Table 2—Criteria for classifying trajectories of eGFR decline according to analytic approach

Classification of trajectoryif criterion met Criterion

Classification of trajectoryif criterion not met

Spline regression method

Nonlinear Linear model is rejected in favor of spline model at P, 0.05 with partial F test. Linear

Clinically consequentialnonlinearity

Linear model rejected and linear projection of time to reach eGFR = 10 mL/min/1.73 m2 misses the observed time of ESRD by 1 year or more.

Clinically inconsequentialnonlinearity

Acceleration Clinically consequential nonlinear trajectory and the slope of the second splinesegment is steeper than the first.

Deceleration Clinically consequential nonlinear trajectory and the slope of the first splinesegment is steeper than the second.

Bayesian method

Nonlinear Any eGFR difference between the smoothed trajectory and the line fitted to itis $15 mL/min/1.73 m2.

Linear

Bayesian method with criteria of linearityproposed in AASK study

Nonlinear Mean slope for the half of the follow-up months with faster decline differs fromthe mean slope for the half with slower yearly decline .3 mL/min/1.73 m2.

Linear

Figure 2—Examples and frequency of linear and nonlinear of trajectories of eGFR declineclassified with the linear spline approach. Trajectories were classified by comparing the linearmodel (solid gray line) with spline regression (solid black line). The dots represent eGFR obser-vations, the horizontal gridlines correspond to boundaries of CKD stages, and the vertical dashedline indicates the onset of ESRD.

2266 eGFR Decline in Type 1 Diabetes Diabetes Care Volume 39, December 2016

calculated the frequency of Bayesiannonlinear posterior samples accordingto class of trajectory determined by thelinear spline approach. The frequencyof Bayesian nonlinearity was 10.8% in124 patients whose trajectories wereclassified as linear by the spline method,31.6% in 104 patients classified as in-consequentially nonlinear, and 51.1%among 32 patients whose decline wasconcluded consequentially nonlinearwith the spline method.We tested patients’ characteristics,

such as sex, age, duration of diabe-tes, age at diabetes onset, ACR, HbA1c,eGFR at baseline, blood pressure, ACE-Ior ARB treatment, and smoking statusfor their association with deviationsfrom linearity and with accelerationsor decelerations. None were statisticallysignificant.

CONCLUSIONS

We characterized trajectories of eGFRdecline that preceded onset of ESRDin a large group of patients with type 1diabetes. For more than 70%, serial mea-surements of serum creatinine beganwhile their eGFR was.60 mL/min/1.73 m2.The eGFR declined at a very rapid ratein most patients. In half of the group,the yearly rate of eGFR loss was steep-er than 9 mL/min/1.73 m2 and up to72 mL/min/1.73 m2. In the other half, theyearly rate of eGFR loss was between2 and 9 mL/min/1.73 m2.

In most of the patients studied, tra-jectories of eGFR decline were linear onthe arithmetic scale and not constant onthe percentage scale, as it is erroneouslyassumed and used in outcome defini-tions recommended for clinical trials(33–35). Furthermore, acceleration or

deceleration was observed in a small mi-nority of trajectories. There was no evi-dence that acute kidney injury or seriesof it, as suggested by others (6,7), leadsto onset of ESRD in a significant propor-tion of patients with type 1 diabetes.

There was a statistically significant as-sociationof glycemic control, urinaryACR,and renoprotective/antihypertensivetreatment with eGFR decline. No sex dif-ferences in the slopeswere present in ourstudy group, and the protective effect ofprepubertal diabetes onset was absent(3,36,37). We were not able to identifypredictors of deviations from linearity inour study group. This may be partly ex-plained by the lack of statistical powerand the inclusion, by design, of only thosepatients who developed ESRD.

Infrequent presence of accelerationsand decelerations, regardless of the ini-tial slope, suggests that eGFR decline isdetermined by constitutive factors,possibly genetic susceptibilities. A dis-ease process that initiates the onset ofeGFR decline occurs when patientshave normal eGFR, as found in this studyand in our previous report (26). In theJoslin Proteinuria Cohort, baseline HbA1cand blood pressure were higher in thosewho reached ESRD (20). An intriguingquestion is, “Is persistence of a triggernecessary to sustain the disease processthat underlies progressive eGFR declineonce it is initiated?” Our recent reportssuggest that persistence of triggers isimportant. Improvement in glycemiccontrol sustained for several years inpatients with previously poor controlhad salutary effects on the incidencerate of ESRD (30) and on the influenceof markers of constitutive susceptibil-ity (38).

We previously (4) traced eGFR declinetrajectories in 244 patients with type 1diabetes and persistent proteinuria whohad normal baseline renal function. Re-nal function was stable in approximatelyhalf of the patients. Significant nonlin-ear decline was rare, and decelerationswere more frequent than accelerations.Most patients did not reach ESRD, mak-ing their trajectories incomplete. Pa-tients in that cohort who reached ESRDwere too few to draw unambiguous con-clusions and to rule out that these ESRDevents resulted from acute kidney injurycomplicating the chronic condition. Ourcurrent results, for the first time, de-scribe complete trajectories in a large

Figure 3—Examples of Bayesian trajectories of eGFR decline. The solid black line represents themean estimated trajectory, the gray zone indicates the 95% point-wise Bayesian credible in-terval, the black dots represent eGFR observations, the horizontal gridlines represent CKD stagesboundaries, and the vertical gridline indicates the time of ESRD onset. A and B: Patients havevarying slopes of eGFR decline, but all of their trajectories are linear. C and D: Patients havevisibly linear trajectories but present some fluctuations of eGFR. E and F: Patients have clearlynonlinear trajectories. Probability of linearity [P(I)] according to AASK study criteria is provided.

care.diabetesjournals.org Skupien and Associates 2267

number of patients and indicate thateGFR decline is linear during the earlystage of the process (4) and just beforeESRD. The same disease process seemsto be responsible for the early (eGFR$60 mL/min/1.73 m2) and late (eGFR,60 mL/min/1.73 m2) decline in eGFR.Our findings have important implica-

tions for the design and evaluation ofclinical trials of interventions to reducerisk of ESRD in type 1 diabetes. First, thepredominance of linear trajectories isfortunate because the slope is easy tointerpret and makes possible simplepredictions of a patient’s near andmore distant future. Its stability overtime allows it to be used as a dependentvariable in studies of determinants ofeGFR decline and ESRD risk biomarkers(M. Yamanouchi, A.S.K., unpublished ob-servations). Linearity of eGFR decline im-plies that an association with the slopeobserved during 3–6 years of follow-uptranslates strongly into an associationwith the risk of ESRD. An early slope ofrenal decline may serve as a surrogateend point as long as it is estimatedfrom sufficiently dense data (we postu-late measurements for research pur-poses every 3–6 months).From data presented in Fig. 1 we may

estimate that .80% of patients withtype 1 diabetes who will develop ESRD inthe fourth to sixth decade of life will havea yearly eGFR loss $5 mL/min/1.73 m2.Once prognostic algorithms are devel-oped, these high-risk patients shouldbe screened to identify them whilethey have still normal renal function. Ef-fective interventions at this early stagemight decelerate eGFR slopes and post-pone onset of ESRD by years rather thanby several months, as is expected whentreatment is implemented in CKD stage3 or 4. Our recent study suggests thatimprovement in HbA1c can decrease therate of eGFR decline and postpone ESRDonset (30); however, such improvementhad to be sustained for several years. In-terventions aimed at glycemic control, ifinitiated too late, will likely fail.Patients described in this study re-

present a group in which available pre-ventive measures failed. In our previousstudy, where only a minority of patientswith type 1 diabetes with proteinuriadeveloped ESRD, there were moreeGFR trajectories with clinically signifi-cant deceleration (4). Such decelerationsmay represent responses to effective

therapeutic interventions (30). Changein slope of eGFR may be consideredfor future use as the primary end pointin clinical trials, possibly tested usinga quadratic model of decline (30). Un-der such a model, an expected lineardecline in the placebo group will becompared with a change of slope (de-celeration) in the treated group. Useof a quantitative outcome may providemore power than current threshold-based end points.

Emphasizing the strength of our studywe would like to point out the large sizeand unbiased selection of our studygroup, the reliable measurements of se-rum creatinine (most eGFR observa-tions in the range ,90 mL/min/1.73 m2,where less noise is observed), and twoalternative analytical methods to ana-lyze eGFR trajectories that providedsimilar results. The Bayesian approachis appealing because the fit to the datais not constrained to just two or threelinear segments (20,27) and it providesestimates of uncertainty about the char-acter of a trajectory. However, the ap-proach based on fitting linear splinesegments provides results that are easierto summarize numerically in clinical trialsand epidemiologic studies. Once the pre-dominance of the linear pattern is estab-lished, the preferred summarymeasure ofeGFRdecline should bea single slope. Thismetric is also suitable as an outcomemea-sure for evaluating clinical interventions.We would like to emphasize that the con-clusionabout linearity is highly sensitive toachosen definition of what constitutes a sig-nificant departure from it. Current def-initions are arbitrary and data driven;therefore, we postulate adopting uniformconsensus criteria for the future research.

Although our study hasmany strengths,we acknowledge its limitations. First, itsgeneralizability would benefit from repli-cation, especially in type 2 diabetes andnon-Caucasian individuals. Second, wewere unable to test trajectories of allESRD patients at Joslin Clinic because toofew eGFR measurements were available,often because of the very rapid declineof renal function. Also, eGFR measure-mentswerenot uniformly spaced, aweak-ness that favors detection of too manyrather than too few nonlinear trajectoriesas they become unstable in areas withsparse observations. Finally, we acknowl-edge that direct GFR measurements werenot available in our patients; however, a

recent study found that measured GFRcompared with eGFR did not show anenhanced association with risk of ESRD,cardiovascular events, or death (39).

Funding. This project was supported throughJDRF fellowship grant 3-2009-397 (J.S.) and re-search grant 17-2013-8 (A.S.K.) and NationalInstitutes of Health grant DK41526 (A.S.K.).Duality of Interest. No potential conflicts ofinterest relevant to this article were reported.Author Contributions. J.S. performed statisti-cal analysis. J.S., J.H.W., R.C.S., andA.S.K. draftedthe manuscript. J.S., J.H.W., and A.S.K. contrib-uted to the study concept anddesign. All authorswere involved in data analysis and interpreta-tion. J.S. and A.S.K. and obtained the funding.J.H.W., R.C.S., and A.S.K. supervised the study.A.M.S. acquired thedataandwas responsible forthe administrative, technical, and material sup-port. R.C.S. and A.S.K. critically revised themanuscript for important intellectual content.J.S. and A.S.K. are the guarantors of this workand, as such, had full access to all the data in thestudy and take responsibility for the integrity ofthe data and the accuracy of the data analysis.

References1. Laing SP, Swerdlow AJ, Slater SD, et al. Mor-tality from heart disease in a cohort of 23,000patients with insulin-treated diabetes. Diabeto-logia 2003;46:760–7652. Nishimura R, Dorman JS, Bosnyak Z, TajimaN, Becker DJ, Orchard TJ; Diabetes EpidemiologyResearch InternationalMortality Study; AlleghenyCounty Registry. Incidence of ESRD and survivalafter renal replacement therapy in patients withtype 1 diabetes: a report from the AlleghenyCounty Registry. Am J Kidney Dis 2003;42:117–1243. Finne P, Reunanen A, Stenman S, Groop PH,Gronhagen-Riska C. Incidence of end-stage re-nal disease in patients with type 1 diabetes.JAMA 2005;294:1782–17874. Skupien J,WarramJH,SmilesAM,etal. Theearlydecline in renal function in patients with type 1 di-abetes and proteinuria predicts the risk of end-stage renal disease. Kidney Int 2012;82:589–5975. Krolewski AS. Progressive renal decline: thenew paradigm of diabetic nephropathy in type 1diabetes. Diabetes Care 2015;38:954–9626. Onuigbo MA, Agbasi N. Diabetic nephropa-thy and CKD-analysis of individual patient serumcreatinine trajectories: a forgotten diagnosticmethodology for diabetic CKD prognosticationand prediction. J Clin Med 2015;4:1348–13687. Kelly KJ, Dominguez JH. Rapid progression ofdiabetic nephropathy is linked to inflammationand episodes of acute renal failure. Am J Neph-rol 2010;32:469–4758. Jones RH, Hayakawa H, Mackay JD, ParsonsV, Watkins PJ. Progression of diabetic nephrop-athy. Lancet 1979;1:1105–11069. Mitch WE, Walser M, Buffington GA, LemannJ Jr. A simplemethod of estimating progression ofchronic renal failure. Lancet 1976;2:1326–132810. Bleyer AJ. A reciprocal graph to plot thereciprocal serum creatinine over time. Am J Kid-ney Dis 1999;34:576–578

2268 eGFR Decline in Type 1 Diabetes Diabetes Care Volume 39, December 2016

11. Mogensen CE. Progression of nephropathyin long-term diabetics with proteinuria andeffect of initial anti-hypertensive treatment.Scand J Clin Lab Invest 1976;36:383–38812. Parving HH, Smidt UM, Friisberg B, Bonnevie-Nielsen V, Andersen AR. A prospective study ofglomerular filtration rate and arterial bloodpressure in insulin-dependent diabetics withdiabetic nephropathy. Diabetologia 1981;20:457–46113. Viberti GC, Bilous RW, Mackintosh D, KeenH. Monitoring glomerular function in diabeticnephropathy. A prospective study. Am J Med1983;74:256–26414. Shah BV, Levey AS. Spontaneous changes inthe rate of decline in reciprocal serum creati-nine: errors in predicting the progression of re-nal disease from extrapolation of the slope.J Am Soc Nephrol 1992;2:1186–119115. Szeto CC, Leung CB, Wong TY, et al. Extrap-olation of reciprocal creatinine plot is not reli-able in predicting the onset of dialysis inpatients with progressive renal insufficiency.J Intern Med 2003;253:335–34216. Li L, Astor BC, Lewis J, et al. Longitudinalprogression trajectory of GFR among patientswith CKD. Am J Kidney Dis 2012;59:504–51217. Genovese G, Friedman DJ, Ross MD, et al.Association of trypanolytic ApoL1 variants withkidney disease in African Americans. Science2010;329:841–84518. Wright JT Jr, Bakris G, Greene T, et al.; Afri-can American Study of Kidney Disease andHypertension Study Group. Effect of bloodpressure lowering and antihypertensive drugclass on progression of hypertensive kidney dis-ease: results from the AASK trial. JAMA 2002;288:2421–243119. Crainiceanu CM, Ruppert D, Wand MP.Bayesian analysis for penalized spline regres-sion using WinBUGS. J Stat Softw 2005;14:1–2420. Rosolowsky ET, Skupien J, Smiles AM, et al.Risk for ESRD in type 1 diabetes remains highdespite renoprotection. J Am Soc Nephrol 2011;22:545–553

21. Krolewski AS, Laffel LM, KrolewskiM, QuinnM, Warram JH. Glycosylated hemoglobin andthe risk of microalbuminuria in patients withinsulin-dependent diabetes mellitus. N Engl JMed 1995;332:1251–125522. Perkins BA, Ficociello LH, Silva KH,Finkelstein DM, Warram JH, Krolewski AS. Re-gression of microalbuminuria in type 1 diabetes.N Engl J Med 2003;348:2285–229323. Ficociello LH, Perkins BA, Roshan B, et al.Renal hyperfiltration and the development ofmicroalbuminuria in type 1 diabetes. DiabetesCare 2009;32:889–89324. Perkins BA, Ficociello LH, Ostrander BE,et al. Microalbuminuria and the risk for earlyprogressive renal function decline in type 1 di-abetes. J Am Soc Nephrol 2007;18:1353–136125. Nowak N, Skupien J, Niewczas MA, et al.Increased plasma kidney injury molecule-1 sug-gests early progressive renal decline in non-proteinuric patients with type 1 diabetes. KidneyInt 2016;89:459–46726. Krolewski AS, Niewczas MA, Skupien J, et al.Early progressive renal decline precedes the on-set of microalbuminuria and its progression tomacroalbuminuria. Diabetes Care 2014;37:226–23427. Mueller PW, Rogus JJ, Cleary PA, et al. Ge-netics of Kidneys in Diabetes (GoKinD) study: agenetics collection available for identifying ge-netic susceptibility factors for diabetic nephrop-athy in type 1 diabetes. J Am Soc Nephrol 2006;17:1782–179028. Pezzolesi MG, Poznik GD, Mychaleckyj JC,et al.; DCCT/EDIC Research Group. Genome-wide association scan for diabetic nephropathysusceptibility genes in type 1 diabetes. Diabetes2009;58:1403–141029. Warram JH, Gearin G, Laffel L, Krolewski AS.Effect of duration of type I diabetes on the prev-alence of stages of diabetic nephropathy de-fined by urinary albumin/creatinine ratio. J AmSoc Nephrol 1996;7:930–93730. Skupien J, Warram JH, Smiles A, Galecki A,Stanton RC, Krolewski AS. Improved glycemic

control and risk of ESRD in patients withtype 1 diabetes and proteinuria. J Am Soc Neph-rol 2014;25:2916–292531. Levey AS, Stevens LA, Schmid CH, et al.;CKD-EPI (Chronic Kidney Disease EpidemiologyCollaboration). A new equation to estimate glo-merular filtration rate. Ann Intern Med 2009;150:604–61232. Jones RH, Molitoris BA. A statistical methodfor determining the breakpoint of two lines.Anal Biochem 1984;141:287–29033. Levey AS, Inker LA, Matsushita K, et al. GFRdecline as an end point for clinical trials in CKD: ascientific workshop sponsored by the NationalKidney Foundation and the US Food and DrugAdministration. Am J Kidney Dis 2014;64:821–83534. Coresh J, Turin TC, Matsushita K, et al.; CKDPrognosis Consortium. Decline in estimated glo-merular filtration rate and subsequent risk ofend-stage renal disease and mortality. JAMA2014;311:2518–253135. Inker LA, Lambers Heerspink HJ, Mondal H,et al. GFR decline as an alternative endpoint forkidney failure: a meta-analysis of treatment ef-fects from 37 randomized trials. Am J Kidney Dis2014;64:848–85936. Harvey JN. The influence of sex and pu-berty on the progression of diabetic nephrop-athy and retinopathy. Diabetologia 2011;54:1943–194537. Donaghue KC, Fairchild JM, Craig ME, et al.Do all prepubertal years of diabetes durationcontribute equally to diabetes complications?Diabetes Care 2003;26:1224–122938. Skupien J, Warram JH, Niewczas MA, et al.Synergism between circulating tumor necrosisfactor receptor 2 and HbA(1c) in determiningrenal decline during 5-18 years of follow-up inpatients with type 1 diabetes and proteinuria.Diabetes Care 2014;37:2601–260839. Ku E, Xie D, Shlipak M, et al.; CRIC StudyInvestigators. Change in measured GFR versuseGFR and CKD outcomes. J Am Soc Nephrol2016;27:2196–2204

care.diabetesjournals.org Skupien and Associates 2269