Journal of Diabetes and Its Complications xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Journal of Diabetes and Its Complications journal homepage: WWW.JDCJOURNAL.COM

Prevalence of chronic kidney disease among individuals with diabetes in the SUPREME-DM Project, 2005–2011 Emily B. Schroeder a, b,⁎, J. David Powers a, Patrick J. O’Connor c, Gregory A. Nichols d, Stanley Xu a, Jay R. Desai c, Andrew J. Karter e, Leo S. Morales f, 1, Katherine M. Newton f, Ram D. Pathak g, Gabriela Vazquez-Benitez c, Marsha A. Raebel a, h, Melissa G. Butler i, 2, Jennifer Elston Lafata j, k, Kristi Reynolds l, Abraham Thomas k, 3, Beth E. Waitzfelder m, John F. Steiner a, b, On behalf of the SUPREME-DM Study Group a

Institute for Health Research, Kaiser Permanente Colorado, Denver, Colorado University of Colorado School of Medicine, Aurora, Colorado HealthPartners Institute for Education and Research, Minneapolis, Minnesota d Kaiser Permanente Center for Health Research, Portland, Oregon e Division of Research, Kaiser Permanente Northern California, Oakland, California f Group Health Research Institute, Seattle, Washington g Marshfield Clinic, Marshfield, Wisconsin h University of Colorado Skaggs School of Pharmacy and Pharmaceutical Sciences, Aurora, Colorado i Center for Clinical and Outcomes Research, Kaiser Permanente Georgia, Atlanta, Georgia j Department of Social and Behavioral Health, Virginia Commonwealth University, Richmond, Virginia k Henry Ford Health System, Detroit, Michigan l Department of Research & Evaluation, Kaiser Permanente Southern California, Pasadena, California m Kaiser Permanente Center for Health Research, Honolulu, Hawaii b c

a r t i c l e

i n f o

Article history: Received 4 March 2015 Accepted 11 April 2015 Available online xxxx Keywords: Chronic renal insufficiency Diabetes mellitus Electronic health records Epidemiology Prevalence

a b s t r a c t Aims: Diabetes is a leading cause of chronic kidney disease (CKD). Different methods of CKD ascertainment may impact prevalence estimates. We used data from 11 integrated health systems in the United States to estimate CKD prevalence in adults with diabetes (2005–2011), and compare the effect of different ascertainment methods on prevalence estimates. Methods: We used the SUPREME-DM DataLink (n = 879,312) to estimate annual CKD prevalence. Methods of CKD ascertainment included: diagnosis codes alone, impaired estimated glomerular filtration rate (eGFR) alone (eGFR b 60 mL/min/1.73 m 2), albuminuria alone (spot urine albumin creatinine ratio N 30 mg/g or equivalent), and combinations of these approaches. Results: CKD prevalence was 20.0% using diagnosis codes, 17.7% using impaired eGFR, 11.9% using albuminuria, and 32.7% when one or more method suggested CKD. The criteria had poor concordance. After age- and sex-standardization to the 2010 U.S. Census population, prevalence using diagnosis codes increased from 10.7% in 2005 to 14.3% in 2011 (P b 0.001). The prevalence using eGFR decreased from 9.7% in 2005 to 8.6% in 2011 (P b 0.001). Conclusions: Our data indicate that CKD prevalence and prevalence trends differ according to the CKD ascertainment method, highlighting the necessity for multiple sources of data to accurately estimate and track CKD prevalence. © 2015 Elsevier Inc. All rights reserved.

Conflict of interest: The authors have no conflicts of interest to declare. ⁎ Corresponding author at: Institute for Health Research, Kaiser Permanente Colorado, 10065 E. Harvard Ave., Suite 300, Denver, CO 80231. Tel.: +1 303 614 1396; fax: +1 303 614 1305. E-mail address: [email protected] (E.B. Schroeder). 1 Present addresses: Department of Medicine, University of Washington, Seattle, Washington. 2 Present addresses: The Argus Group, Hamilton, Bermuda. 3 Present addresses: Department of Medicine, Lutheran Medical Center, Brooklyn, New York.

1. Introduction Diabetes is a leading cause of chronic kidney disease (CKD) (de Boer et al., 2011; US Renal Data System, 2013). CKD occurs in 20–40% of individuals with diabetes (American Diabetes Association, 2014; de Boer et al., 2011; Garg, Kiberd, Clark, Haynes, & Clase, 2002; Nathan et al., 2009; National Kidney Foundation, 2007), and even small degrees of renal impairment are associated with increased cardiovascular disease risk, cardiovascular mortality, and health care costs (de Boer

http://dx.doi.org/10.1016/j.jdiacomp.2015.04.007 1056-8727/$© 2015 Elsevier Inc. All rights reserved.

Please cite this article as: Schroeder, E.B., et al., Prevalence of chronic kidney disease among individuals with diabetes in the SUPREME-DM Project, 2005–2011, Journal of Diabetes and Its Complications (2015), http://dx.doi.org/10.1016/j.jdiacomp.2015.04.007

2

E.B. Schroeder et al. / Journal of Diabetes and Its Complications xxx (2015) xxx–xxx

et al., 2009, 2011; Ninomiya et al., 2009; Tuttle et al., 2014; US Renal Data System, 2013). The U.S. Renal Data System (USRDS) annually compiles information about CKD prevalence in the U.S., including CKD prevalence among individuals with diabetes. Its data primarily come from four sources: 1) Medicare claims, 2) laboratory test results and self-reported health information from NHANES, 3) claims from the Truven Health Marketscan Database, and 4) claims and laboratory test results from the Clinformatics DataMart (US Renal Data System, 2013). Methods of estimating CKD prevalence have differed depending on the data source used. Studies using Medicare and Marketscan databases have relied entirely on claims data (diagnosis and procedure codes), since these databases do not include laboratory test results. Studies using NHANES are based on cross-sectional laboratory test results and self-reported health information, but do not capture claims data. Only the Clinformatics DataMart database contains both claims and laboratory data on the non-end-stage renal disease population, and to date the USRDS use of these laboratory data has been limited (US Renal Data System, 2013). Previous publications have found poor concordance between CKD ascertainment methods based on diagnosis codes and those based on laboratory test results (Ferris et al., 2009; Grams et al., 2011; Kern et al., 2006; Stevens et al., 2005; Winkelmayer et al., 2005). However, the implications of this lack of concordance for estimation of CKD prevalence among individuals with diabetes have not been explored. The electronic health records of large health care delivery systems offer a complementary source of data for estimation of trends in disease prevalence as well as comparative effectiveness research to prevent or delay the development of CKD or end-stage renal disease. We used data from the SUrveillance, PREvention, and ManagEment of Diabetes Mellitus (SUPREME-DM) DataLink, a database that includes over one million individuals with diabetes from 11 U.S. healthcare delivery systems, to examine CKD prevalence and trends among individuals with diabetes from 2005 to 2011, describe the agreement among different CKD ascertainment methods, and compare CKD prevalence and trends among pre-specified clinical and demographic subgroups (age, gender, hypertension, and cardiovascular disease).

2. Material and methods 2.1. Setting and data sources SUPREME-DM is a consortium of 11 member organizations of the HMO Research Network. Collectively, SUPREME-DM includes data from approximately 16 million adults in 10 states from 2005 to 2011. Health systems in SUPREME-DM include Geisinger Health System (Pennsylvania), Group Health (Washington), HealthPartners (Minnesota), Henry Ford Health System (Michigan), Kaiser Permanente regions in Colorado, Northern California, Southern California, Hawaii, Georgia, Northwest (Oregon and Washington), and Marshfield Clinic (Wisconsin). Research institutions embedded in these health systems have developed a distributed virtual data warehouse that contains information on demographics, outpatient pharmacy dispensing, laboratory tests and laboratory results, and diagnosis and procedure codes from outpatient and inpatient health care encounters from their electronic health record and administrative data systems (Hornbrook et al., 2005). The dataset developed within SUPREME-DM, the DataLink, is the largest and clinically detailed privately-insured diabetes patient cohort ever assembled in the U.S (Nichols et al., 2012, 2015). This study was approved by the Kaiser Permanente Colorado Institutional Review Board (IRB), and each participating site either ceded oversight to the Kaiser Permanente Colorado IRB or received approval from their local IRB.

2.2. Study participants We used inpatient and outpatient diagnosis codes, laboratory, and pharmacy data elements included in the DataLink to identify adults with diabetes. Specifically, we considered diabetes identification in the DataLink as the earlier of one inpatient diagnosis (ICD-9-CM 250.x, 357.2, 366.41, 362.01–362.07, either primary or secondary) or any combination of two of the following events, using the date of the first event in the pair as the identification date: 1) HbA1c N 6.5%; 2) fasting plasma glucose N 126 mg/dl; 3) random plasma glucose N 200 mg/ dl; 4) outpatient diagnosis code (same codes as for inpatient); 5) any anti-hyperglycemic medication dispensing. When the two events were from the same source (e.g. two outpatient diagnoses or two elevated laboratory values), we required them to occur on separate days no more than two years apart. Two dispensings of metformin or thiazolidinediones with no other indication of diabetes were not included because these agents could be used for diabetes prevention or to treat polycystic ovarian syndrome. Criteria ascertained during periods of pregnancy were excluded. Information on diabetes status was available from 2005 through 2011, and from as early as 2000 at some sites. Once an individual was identified as having diabetes, they remained in the diabetes cohort until the date of censoring (the earliest of disenrollment from a participating health plan for greater than 90 days, death, or December 31, 2011). For this study, the eligible source population for each calendar year consisted of individuals enrolled in a participating health plan from January 1 through December 31 of that year, without any enrollment gaps greater than 90 days. The index date was defined as the latest of the diabetes diagnosis date (if diagnosed after January 1, 2005) or January 1st of the first full year of enrollment in 2005–2011 when an individual was 20 years or older, the age that is traditionally used by the Centers for Disease Control and Prevention to differentiate adults and youth with diabetes (Centers of Disease Control and Prevention, 2014). Individuals with diabetes were included in these analyses if they: 1) were at least 20 years old by January 1st, 2011, 2) were enrolled at the diabetes diagnosis date and had at least one day of enrollment after the index date, 3) had at least one full calendar year of enrollment, and 4) had at least two laboratory measurements of hemoglobin A1c, fasting glucose, or random glucose or two blood pressures recorded within 365 days before or after the index date. This last criterion was designed to exclude people receiving care at locations where their data were not being captured in the participating plans' electronic health records. 2.3. Study variables We calculated CKD prevalence separately for each calendar year. Following the USRDS methodology, we did not carry CKD definitions forward into future years. We examined the following distinct methods of CKD ascertainment, separately and in combination. 1) Diagnosis codes. Following USRDS claims data methodology, we required at least one inpatient ICD-9 diagnosis code or two outpatient ICD-9 diagnosis codes that indicate kidney disease. We used the USRDS eligibility codes: 016.0, 095.4, 189.0, 189.9, 223.0, 236.91, 250.4, 271.4, 274.1, 283.11, 403.x1, 403.x0, 404.x2, 404.x3, 404.x0, 404.x1, 440.1, 442.1, 447.3, 572.4, 580–588, 591, 642.1, 646.2, 753.12–753.17, 753.19, 753.2, and 794.4 (US Renal Data System, 2013). 2) Impaired estimated glomerular filtration rate (eGFR). Using all available ambulatory serum creatinine results, we estimated GFR using the CKD-EPI equation (Levey et al., 2009). To meet the definition for impaired eGFR, we required at least two eGFR b 60 mL/min/ 1.73 m 2 separated by 91–365 days without any intervening values N 60 mL/min/1.73 m2. At least one of the eGFR's b 60 mL/ min/1.73 m2 was required to be in the given calendar year; the other could be in the same year or the preceding year.

Please cite this article as: Schroeder, E.B., et al., Prevalence of chronic kidney disease among individuals with diabetes in the SUPREME-DM Project, 2005–2011, Journal of Diabetes and Its Complications (2015), http://dx.doi.org/10.1016/j.jdiacomp.2015.04.007

E.B. Schroeder et al. / Journal of Diabetes and Its Complications xxx (2015) xxx–xxx

3) Albuminuria. Using all available ambulatory spot urine albumin creatinine ratios, urine protein dipsticks, and spot urine protein creatinine ratios, we identified abnormal values: albumin creatinine ratio N 30 mg/g, dipstick protein 1 + or greater, and protein creatinine ratio N 50 mg/g (Hallan et al., 2012). To define a population with persistent albuminuria, we estimated an albuminuria persistence rate (i.e. the proportion of people with a single abnormal urine test who have a second abnormal urine test). This was done by randomly selecting one abnormal urine lab between 2005 and 2010 for each person with at least one abnormal urine value. We then identified a cohort of people with a second urine test 91–365 days after the randomly selected abnormal test. We calculated the persistence rate as the percentage of people with a first abnormal test whose second test was also abnormal, which was 49.3%. We conducted sub-group analyses based on age, gender, hypertension, and cardiovascular disease. Age was defined as of January 1st of the calendar year of interest. Hypertension and cardiovascular disease (not including congestive heart failure) were defined as at least one inpatient ICD-9 diagnosis code or two outpatient ICD-9 diagnosis codes within the calendar year (hypertension: 362.11, 401.x–405.x, 437.2; CVD: 404.x1, 410–414, 420–421, 423–424, 426– 427, 429, 430–438, 440–444, 447, 451–453, 557, 785.0–785.3, V42.2, V45.0, V45.81, V45.82, V53.3) (US Renal Data System, 2013). Hypertension and cardiovascular disease status were determined separately using data for a given calendar year, and were not carried forward to future years.

2.4. Statistical analysis Due to lack of systematic collection of race information, particularly in earlier study years, 9% of the cohort was missing race data. Because race (African American or non-African American) is necessary for the CKD-EPI formula, we conducted multiple imputations. For individuals with missing race information, this imputation created five replicate demographic observations indicating whether the individual was African American or not, using site-specific frequencies of the populated race data. Using these data, separate determinations of eGFR based CKD status were made for each replicate on every year of follow-up for these individuals (Little & Rubin, 2002). We generated descriptive statistics for each calendar year, as well as for all unique individuals. We then calculated the CKD prevalence per calendar year for different CKD ascertainment methods. We also present the weighted average prevalence, which was calculated using the yearly prevalences weighted by the size of the cohort in each year. We examined baseline characteristics by presence or absence of laboratory results (serum creatinine and urine albumin) by calendar year. For subgroup analysis (by age, gender, hypertension, and cardiovascular disease), we used the combined CKD criteria (any of diagnosis code, impaired eGFR, or albuminuria criteria). Generalized estimating equations analyses were used to model overall and stratified linear yearly trends of CKD prevalence. Separate models were estimated for each of the previously described individual and combined CKD criteria. These models used the log link for Poisson distributed counts of CKD events per health care system (n = 11) and stratification variable (age, gender, hypertension, and cardiovascular disease) by year, adjusted for the cohort size in each year. The correlation of yearly prevalences within each site was accounted for by a first order autoregressive covariance structure. Models were adjusted for study site (health care system) and produced prevalences that were age- and sex-standardized to the 2010 U.S. Census population. Using these models, we compared whether there was a trend over time, whether the prevalence was the same between subgroup (test of coincidence or difference in overall average), and

3

whether the trend differed between subgroups (test of parallelism or difference in slope). Analyses were conducted using SAS version 9.2. 3. Results Fig. 1 shows the cohort construction. The number of unique individuals was 879,312, with 3,824,655 person–years of observation (Table 1). While the size of the population increased over time, the covariate distribution remained similar. Median follow-up time was 5.0 years. The population was 48% female. The annual mean age ranged between 61.4 years in 2005 and 62.1 years in 2011. Forty-five percent of these individuals were ethnic or racial minorities, and 9% had unknown or missing race and therefore had race imputed. Seventy one percent of the population had been diagnosed with hypertension, and 35% with cardiovascular disease. The percent of individuals with at least one serum creatinine per year increased over time, from 87.1% in 2006 to 91.0% in 2011 (P b 0.001 for test for trend) (Table 2). Among those with a serum creatinine, mean eGFR ranged from 68.8 ml/min/1.72 m 2 in 2007 to 71.7 ml/min/ 1.72 m2 in 2011 (P = 0.01 for test for trend). Individuals without at least one serum creatinine in a calendar year were younger, more likely to be White, and less likely to have diagnosed hypertension or cardiovascular disease (Supplementary Table 1). The percent of individuals with at least one measure of urine albumin ranged from 72 to 73% for all years except 2011, when it was 76% (Table 1). Of individuals who had one abnormal urine measure, 65–67% had a second urine collected in the next 91–365 days. Individuals without at least one measure of urine albumin in a calendar year were more likely to be less than 45 or greater than 75 years of age, less likely to be Hispanic, more likely to be White, and less likely to have diagnosed hypertension or cardiovascular disease (Supplementary Table 1). Fig. 2 and Table 2 show the number and percent of individuals meeting each method of CKD ascertainment by year, unadjusted (Fig. 2A), age- and sex-standardized to the 2010 U.S. Census population (Fig. 2B), or restricted to individuals at least 65 years of age (Fig. 2C). The weighted average CKD prevalence was 20.0% using diagnosis codes, 17.7% using impaired eGFR alone, and 11.9% using albuminuria alone. Overall, 32.7% met at least one of the three CKD criteria. After age- and sex-standardization to the 2010 U.S. Census population, the CKD prevalence based on diagnosis codes rose from 10.7% in 2005 to 14.3% in 2011 (an average relative increase of 4.1%

DataLink: Individuals with Diabetes N=1,229,898

At least 20 years old by 1/1/2011 N=1,220,670 (1% excluded)

Enrolled at the diabetes diagnosis date And at least one day of enrollment after the index date N=1,183,711 (3% excluded)

At least one full calendar year of enrollment N=965,809 (18% excluded)

2 lab or blood pressure measurements +/- 365 days of the index date N=879,312 (9% excluded) Fig. 1. SUPREME-DM DataLink, inclusion and exclusion criteria.

Please cite this article as: Schroeder, E.B., et al., Prevalence of chronic kidney disease among individuals with diabetes in the SUPREME-DM Project, 2005–2011, Journal of Diabetes and Its Complications (2015), http://dx.doi.org/10.1016/j.jdiacomp.2015.04.007

4

E.B. Schroeder et al. / Journal of Diabetes and Its Complications xxx (2015) xxx–xxx

Table 1 Characteristics of the SUPREME-DM population by year, 2005–2011. 2005

2006

2007

2008

2009

2010

2011

Unique overall

Total, n

457,419

489,969

519,999

550,037

576,033

601,391

629,807

879,312

Sex, n (%) Female Male

222,428 (49) 234,991 (51)

238,519 (49) 251,450 (51)

253,738 (49) 266,261 (51)

269,113 (49) 280,924 (51)

281,675 (49) 294,358 (51)

293,830 (49) 307,561 (51)

308,068 (49) 321,739 (51)

425,014 (48) 454,298 (52)

Agea, n (%) 0–19 20–44 45–54 55–64 65–74 75–84 85+ Age, mean (SD)

0 (0) 47,674 (10) 90195 (20) 127,134 (28) 113,255 (25) 66,899 (15) 12,262 (3) 61.4 (13.2)

0 (0) 50,857 (10) 95,693 (20) 137,512 (28) 119,052 (24) 72,381 (15) 14,474 (3) 61.5 (13.2)

0 (0) 54,478 (10) 101,518 (20) 148,036 (28) 122,805 (24) 76,295 (15) 16,867 (3) 61.4 (13.3)

0 (0) 57,899 (11) 106,430 (19) 157,734 (29) 127,898 (23) 80,909 (15) 19,167 (3) 61.5 (13.4)

0 (0) 59,777 (10) 109,327 (19) 164,867 (29) 135,711 (24) 84,840 (15) 21,511 (4) 61.7 (13.4)

0 (0) 60,867 (10) 111,567 (19) 171,513 (29) 144,208 (24) 89,283 (15) 23,953 (4) 61.9 (13.4)

0 (0) 62,349 (10) 113,782 (18) 179,895 (29) 153,564 (24) 93,989 (15) 26,228 (4) 62.1 (13.4)

3,457 (0) 147,810 (17) 221,075 (25) 243,272 (28) 173,946 (20) 76,810 (9) 12,942 (1) 57.1 (13.4)

Race/Ethnicity, n (%) American Indian/Alaska Native Asian African American Hispanic–African American Hispanic–Non-African American Multiple Native Hawaiian or Other Pacific Islander White Unknown or not reported

1,294 (0) 4,8912 (11) 51,317 (11) 1,007 (0) 85,200 (19) 8,274 (2) 3,430 (1) 220,743 (48) 37,242 (8)

1,421 (0) 53,641 (11) 54,670 (11) 1,046 (0) 93,977 (19) 8,973 (2) 3,833 (1) 234,979 (48) 37,429 (8)

1,530 (0) 58,764 (11) 57,735 (11) 1,083 (0) 102,802 (20) 9,675 (2) 4,188 (1) 247,456 (48) 36,766 (7)

1,640 (0) 64,139 (12) 60,923 (11) 1,142 (0) 111,385 (20) 10,350 (2) 4,509 (1) 260,029 (47) 35,920 (7)

1,742 (0) 68,804 (12) 64,118 (11) 1,187 (0) 118,724 (21) 10,928 (2) 4,789 (1) 271,448 (47) 34,293 (6)

1,873 (0) 74,003 (12) 67,373 (11) 1,222 (0) 126,884 (21) 11,572 (2) 5,098 (1) 283,157 (47) 30,209 (5)

2,019 (0) 79,558 (13) 70,015 (11) 1,263 (0) 136,122 (22) 12,262 (2) 5,583 (1) 293,605 (47) 29,380 (5)

2,720 (0) 98,022 (11) 92,215 (10) 1,726 (0) 181,750 (21) 15,576 (2) 7,100 (1) 402,748 (46) 77,455 (9)

Comorbiditiesb, n (%) Hypertension Cardiovascular disease

214,766 (47) 96,046 (21)

245,676 (50) 103,842 (21)

276,002 (53) 114,357 (22)

291,772 (53) 119,170 (22)

312,163 (54) 126,842 (22)

320,636 (53) 134,561 (22)

323,909 (51) 140,018 (22)

620,653 (71) 308,125 (35)

Laboratory availability, n (%) At least 1 serum creatinine Mean eGFR At least one albuminuria/proteinuria lab Follow-up timec (years) mean (SD), median (IQR)

404,890 (89) 69.7 331,468 (72) 5.7 (2.0), 7.0 (4.2–7.0)

426,528 (87) 69.0 359,106 (73) 6.0 (1.6), 7.0 (5.2–7.0)

458,216 (88) 68.8 379,481 (73) 6.1 (1.3), 7.0 (5.4–7.0)

486,929 (89) 70.0 399,767 (73) 6.1 (1.3), 7.0 (5.1–7.0)

511,992 (89) 71.0 415,407 (72) 5.9 (1.4), 6.9 (4.9–7.0)

544,058 (90) 71.2 439,141 (73) 5.6 (1.7), 6.7 (4.2–7.0)

573,033 (91) 71.7 478,069 (76) 5.2 (2.1), 6.3 (3.2–7.0)

849,855 (97) 70.3 788,283 (90) 4.7 (2.1), 5.0 (2.8–7.0)

Abbreviations: IQR, inter-quartile range; eGFR, estimated glomerular filtration rate; SD, standard deviation; SUPREME-DM, SUrveillance, PREvention, and ManagEment of Diabetes Mellitus. a For individual calendar years, age was determined as of January 1st of that year. For “Unique overall,” age was determined as of the diabetes index date, and therefore could be less than 19 if an individual was first diagnosed prior to age 19, and then subsequently aged into the cohort. b For individual calendar years, comorbidities were determined using diagnosis codes from that calendar year. For the “Unique overall” column, we denote whether individuals met the comorbidity diagnosis criteria during one or more of the calendar years. c Follow-up time was calculated from the diabetes index date through the date of censoring (earliest of disenrollment, death, or December 31, 2011).

per year, P b 0.001). The CKD prevalence based on eGFR fell from 9.7% in 2005 to 8.6% in 2011 (an average relative decrease of 2.3% per year, P b 0.001). The CKD prevalence based on albuminuria or any of the three criteria remained stable over the observation period (P = 0.4 and P = 0.8, respectively). Over half (56.8%) of individuals meeting any of the three CKD criteria met only one of those criteria (Fig. 3). Over one-third (34.4%) of individuals who met the diagnosis code criteria had no supporting laboratory values. Conversely, nearly half (48.7%) of individuals who met one or more of the laboratory criteria did not have a diagnosis code indicating CKD. Among individuals meeting at least one CKD criteria, older individuals had higher CKD prevalence than younger individuals (Supplementary Table 2). Prevalence did not differ between men and women. Individuals with cardiovascular disease or hypertension had a higher CKD prevalence than individuals without cardiovascular disease or hypertension. There were no differences between these subgroups in the trend in CKD over time (P N 0.1). 4. Discussion In this cohort of nearly 900,000 individuals with diabetes, we found that the estimates of the prevalence and trends of CKD among individuals with diabetes from 2005 to 2011 depended on the method of CKD ascertainment. Overall, our data do not consistently suggest a

major change in CKD prevalence among individuals with diabetes in the U.S. between 2005 and 2011. Furthermore, there was poor concordance between the CKD ascertainment methods. Note that none of our estimates represents a true “gold standard,” but are instead representative of the data that are available using administrative and electronic health record data. We found a higher CKD prevalence based on diagnosis codes, and a lower laboratory-based CKD prevalence than the USRDS (Coresh et al., 2007; de Boer et al., 2011; US Renal Data System, 2013). The USRDS reports CKD prevalence among those with diabetes of 19.8% based on 2011 Medicare claims data (fee-for-service, age 65 +), 5.6% from Truven Health MarketScan (ages 20–64) and 6.0% from Clinformatics DataMart (ages 20–64) (US Renal Data System, 2013). Our diagnosis code-based prevalence in 2011, restricted to individuals aged 65 or older, was 33.9%, which may reflect coding practices in our health systems. Using 2005–2010 NHANES data and no albuminuria persistence rate, the USRDS reports an impaired eGFR-based prevalence of 19.3%, albuminuria-based prevalence of 29.9%, and combined (either impaired eGFR or albuminuria) prevalence of 40.1% (US Renal Data System, 2013). De Boer, using 2005–2008 NHANES data and an albuminuria persistence rate of 78%, found an impaired eGFR-based prevalence of 17.7%, albuminuria-based prevalence of 23.7%, and a combined rate of 34.5% (de Boer et al., 2011). Our age- and sex-standardized CKD prevalence using impaired eGFR was 9.3%,

Please cite this article as: Schroeder, E.B., et al., Prevalence of chronic kidney disease among individuals with diabetes in the SUPREME-DM Project, 2005–2011, Journal of Diabetes and Its Complications (2015), http://dx.doi.org/10.1016/j.jdiacomp.2015.04.007

E.B. Schroeder et al. / Journal of Diabetes and Its Complications xxx (2015) xxx–xxx

5

Table 2 Chronic kidney disease prevalence by year, according to method of ascertainment, unadjusted, age and sex standardized to the 2010 United States Census population, and restricted to age 65 and older, 2005–2011. Method of chronic kidney disease ascertainment

Unadjusted Diagnosis codes Impaired eGFR Albuminuria Diagnosis codes or impaired eGFR Impaired eGFR or albuminuria Diagnosis codes, impaired eGFR, or albuminuria Age and sex standardized to 2010 U.S. Census population Diagnosis codes Impaired eGFR Albuminuria Diagnosis codes or impaired eGFR Impaired eGFR or albuminuria Diagnosis codes, impaired eGFR, or albuminuria Restricted to age N 65 Diagnosis codes Impaired eGFR Albuminuria Diagnosis codes or impaired eGFR Impaired eGFR or albuminuria Diagnosis codes, impaired eGFR, or albuminuria

2005

2006

2007

2008

2009

2010

2011

2005–2011 Weighted average

N (%)

N (%)

N (%)

N (%)

N (%)

N (%)

N (%)

%

64,695 (14.1) 81,439 (17.8) 57,297 (12.5) 110,969 (24.3)

82,091 (16.8) 90,804 (18.5) 63,285 (12.9) 127,288 (26.0)

103,123 (19.8) 98,554 (19.0) 60,967 (11.7) 143,956 (27.7)

113,462 (20.6) 97,676 (17.8) 64,718 (11.8) 148,910 (27.1)

127,302 (22.1) 98,881 (17.2) 66,707 (11.6) 159,007 (27.6)

134,791 (22.4) 102,969 (17.1) 69,978 (11.6) 166,760 (27.7)

137,660 (21.9) 107,547 (17.1) 74,065 (11.8) 171,960 (27.3)

20.0 17.8 11.9 26.9

121,610 (26.6) 142,394 (31.1)

134,098 (27.4) 160,042 (32.7)

139,042 (26.7) 172,785 (33.2)

141,672 (25.8) 179,970 (32.7)

144,720 (25.1) 150,890 (25.1) 190,327 (33.0) 199,532 (33.2)

158,264 (25.1) 207,084 (32.9)

25.9 32.7

48,896 44,396 51,581 72,617

59,265 (12.1) 48,858 (10.0) 55,869 (11.4) 82,285 (16.8)

72,671 52,742 53,833 93,292

77,732 51,904 57,546 95,585

86,100 (14.9) 51,647 (9.0) 59,603 (10.3) 101,817 (17.7)

89,140 (14.8) 52,835 (8.8) 62,151 (10.3) 104,836 (17.4)

90,210 (14.3) 54,453 (8.6) 65,769 (10.4) 106,856 (17.0)

13.7 9.3 10.6 17.2

86,158 (18.8) 105,913 (23.2)

93,556 (19.1) 117,035 (23.9)

95,193 (18.3) 124,788 (24.0)

98,002 (17.8) 129,833 (23.6)

99,858 (17.3) 103,256 (17.2) 137,022 (23.8) 142,014 (23.6)

107,944 (17.1) 146,749 (23.3)

17.9 23.6

36,404 61,750 27,177 73,829 76,662 85,355

48,740 (23.7) 69,211 (33.6) 30,788 (15.0) 85,049 (41.3) 85,468 (41.5) 96,970 (47.1)

63,440 (29.4) 74,951 (34.7) 29,647 (13.7) 95,798 (44.4) 89,772 (41.6) 105,672 (48.9)

71,819 (31.5) 74,850 (32.8) 31,526 (13.8) 99,740 (43.8) 91,239 (40) 110,376 (48.4)

82,060 (33.9) 76,499 (31.6) 32,779 (13.5) 107,090 (44.2) 93,881 (38.8) 117,659 (48.6)

92,682 (33.9) 84,658 (30.9) 37,744 (13.8) 119,743 (43.7) 104,813 (38.3) 131,986 (48.2)

30.0 32.3 13.9 43.1 39.7 47.9

(10.7) (9.7) (11.3) (15.9)

(18.9) (32.1) (14.1) (38.4) (39.8) (44.4)

(14.0) (10.1) (10.4) (17.9)

(14.1) (9.4) (10.5) (17.4)

89,053 (34.6) 80,617 (31.3) 35,184 (13.7) 114,307 (44.4) 99,260 (38.6) 125,516 (48.8)

Abbreviations: eGFR, estimated glomerular filtration rate.

our albuminuria rate (with adjustment using a persistence rate) was 10.6%, and our combined rate with these two measures was 17.9%. Another study using data from the Department of Veteran Affairs and private health care delivery systems estimated CKD prevalence using serum creatinine and diagnosis data. They identified a higher rate of missing creatinine measurements than in our study, ranging between 24 and 63% depending on the delivery system. Their final prevalence estimates were based on imputations using NHANES data rather than direct measurement (Shahinian et al., 2013). There are several potential reasons why our laboratory-based prevalence rates are lower than those from NHANES: 1) both our numerators and denominators are constructed differently, since we did not use a single set of cross-sectional laboratory results. In order to remain consistent with the diagnosis code-based method and because the timing of laboratory values varies in a community-dwelling population, we required a full year of enrollment in order to be considered in a given calendar year, and we considered all laboratory measurements from that calendar year. 2) While testing rates were high (each year, greater than 87% and 72% had at least one serum creatinine or urine measure, respectively), they were not universal. Thus, there is some degree of under ascertainment due to incomplete laboratory testing. However, individuals without laboratory measurements tended to be at lower risk for CKD. Furthermore, even with complete laboratory testing, it is highly unlikely that our rates would have been as high as in NHANES. 3) We used an internally derived albuminuria persistence rate that is lower than that previously reported (de Boer et al., 2011; Deeb et al., 2012; Pugliese et al., 2011; Saydah et al., 2013). 4) These insured populations may have a lower rate of CKD than the general U.S. population due to better health care access and use of reno-protective therapy such as ACE inhibitors. In addition, it should be noted that eGFR in our population increased over time. This could reflect a change in the composition of the population being tested each year, or serum creatinine assay changes over time (de Boer et al., 2011; Matsushita et al., 2012).

Previous studies have also found poor concordance between laboratory measurements and diagnosis codes for CKD, although many of these studies took place in the hospital setting or relied on a single serum creatinine measurement. Using eGFR as the gold standard, a variety of ICD-9 codes had sensitivities of 11–42% and specificities of 93–99% (Ferris et al., 2009; Grams et al., 2011; Kern et al., 2006; Stevens et al., 2005; Winkelmayer et al., 2005). Given the inclusiveness for the USRDS ICD-9 codes for CKD, which include diagnosis such anatomical kidney disease, it is not surprising that the concordance between laboratory-based and diagnosis-based methods was low. In our study and others, the results of urine albumin tests or serum creatinine measurements varied within individuals over time. This variability can be due to many factors (American Diabetes Association, 2014; Bellomo, Kellum, & Ronco, 2012; Joffe et al., 2010; Miller & Bruns, 2009; Saydah et al., 2013; Stevens, Coresh, Greene, & Levey, 2006; Tuttle et al., 2014). Persistence rates of albuminuria (i.e. the proportion of people with a single abnormal urine test who have a second abnormal urine test) have ranged from 57 to 95%, while our internally derived persistence rate was 49% (de Boer et al., 2011; Deeb et al., 2012; Pugliese et al., 2011; Saydah et al., 2013). Albuminuria persistence rates are likely to be higher in research studies (Deeb et al., 2012). In a population with two eGFR assessments separated by at 90 days, only 56% of individuals with a first eGFR b 60 ml/min/ 1.73 m 2 had a second eGFR also b 60 ml/min/1.73 m 2 (DeVille et al., 2012). This intrapersonal variability in laboratory values has also been a challenge in cross-sectional surveillance efforts such as NHANES (Saydah et al., 2013). Strengths of this study include the ability to combine and compare results from diagnosis codes and laboratory measurements, the large database, and the diversity of membership with regards to geographic location and age. Limitations include the lack of standardization of the timing of laboratory measurements, missing laboratory measurements (either because they were not performed or because results

Please cite this article as: Schroeder, E.B., et al., Prevalence of chronic kidney disease among individuals with diabetes in the SUPREME-DM Project, 2005–2011, Journal of Diabetes and Its Complications (2015), http://dx.doi.org/10.1016/j.jdiacomp.2015.04.007

6

E.B. Schroeder et al. / Journal of Diabetes and Its Complications xxx (2015) xxx–xxx

A) 50% 45%

Dx + GFR 24.5%

Dx 21.2%

40% 35% 30%

Dx

25%

GFR

20% 15% 10% 5%

Dx+ ACR 7.2%

ACR Dx or GFR GFR or ACR

ACR 17.7%

Any

GFR 17.9% Dx + GFR + ACR 8.7% GFR + ACR 2.8%

0%

Fig. 3. Overlap between the different chronic kidney disease ascertainment methods. Weighted averages from the SUPREME-DM DataLink, 2005–2011. Abbreviations: ACR = albuminuria definition; Dx = diagnosis code definition; GFR = impaired estimated glomerular filtration rate definition.

B) 50% 45% 40% 35% 30% 25% 20%

Dx GFR ACR Dx or GFR

15% 10%

individuals. However, our work demonstrates how health care systems can use clinically-derived data to identify and track their trends in CKD prevalence and serves as a useful adjunct to the USRDS data. In the future, this registry could be used for monitoring evolving patterns of care; examining variations in patterns of care, cost of care, and outcomes; and comparative effectiveness research examining the safety and relative efficacy of various treatment strategies.

GFR or ACR Any

4.1. Conclusions

5% 0%

C) 50% 45% 40% 35% 30%

Dx

25%

GFR

20%

ACR

15%

Dx or GFR

10%

GFR or ACR

5%

We used a unique diabetes cohort to track CKD prevalence from 2005 to 2011. Our conclusions about CKD prevalence and its trend over time differed according to the CKD ascertainment method, highlighting the necessity of using multiple sources of data (including both diagnosis codes and longitudinal laboratory assessments) to accurately estimate and track CKD prevalence. Compared to the USRDS, CKD prevalence was higher among individuals with diabetes based on diagnosis codes in the Medicare-aged population, and lower based on laboratory measurements. For populations with a high rate of laboratory testing, we recommend using laboratory-based CKD criteria, as these are not subject to changes in diagnostic coding and allow more precise longitudinal assessment. However, for populations with lower rates of laboratory testing, a combination of diagnostic codes and laboratory testing may provide a more complete picture.

Any

0%

Fig. 2. Chronic kidney disease prevalence by year, according to method of chronic kidney disease ascertainment, unadjusted, age and sex standardized to the 2010 United States Census population, and restricted to age 65 and older, 2005–2011. A. Unadjusted. B. Age- and sex-standardized to the 2010 United States Census population. C. Restricted to age 65 and older. Abbreviations: NHANES, National Health and Nutrition Examination Survey; USRDS, United States Renal Data System. a Weighted average from 2005 to 2011. b Data from NHANES 2005–2008, from de Boer et al. (2011). c Data from NHANES 2005–2010, from the 2013 USRDS Annual Data Report (US Renal Data System, 2013). d Data from United States Medicare population, age 65 and older, from the 2013 USRDS Annual Data Report (US Renal Data System, 2013).

were not in our database), use of multiple accredited clinical chemistry laboratories, the short time frame limiting our ability to detect trends over time, and missing race data. Importantly, our results may not be generalizable to other health systems or uninsured

Acknowledgements We gratefully acknowledge the project management of Andrea Paolino, MA. We thank the programmers and site investigators at Kaiser Permanente Colorado, Northwest, Northern California, Southern California, Georgia, and Hawaii, Geisinger Health System, Marshfield Clinic Research Foundation, HealthPartners Institute for Education and Research, Henry Ford Health System, and Group Health Research Institute. This project was supported by grant number R01HS019859 from the Agency for Healthcare Research and Quality (AHRQ). The study sponsor had no role in the study design; collection, analysis, and interpretation of data; writing the report; and the decision to submit the report for publication. The content is solely the responsibility of the authors and does not necessarily represent the official views of AHRQ. Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.jdiacomp.2015.04.007.

Please cite this article as: Schroeder, E.B., et al., Prevalence of chronic kidney disease among individuals with diabetes in the SUPREME-DM Project, 2005–2011, Journal of Diabetes and Its Complications (2015), http://dx.doi.org/10.1016/j.jdiacomp.2015.04.007

E.B. Schroeder et al. / Journal of Diabetes and Its Complications xxx (2015) xxx–xxx

References American Diabetes Association (2014). Standards of medical care in diabetes–2014. Diabetes Care, 37(Suppl. 1), S14–S80. Bellomo, R., Kellum, J. A., & Ronco, C. (2012). Acute kidney injury. Lancet, 380, 756–766. Centers of Disease Control and Prevention (2014). National Diabetes Statistics Report: Estimates of diabetes and its burden in the United States, 2014. Atlanta, GA: US Department of Health and Human Services. Coresh, J., Selvin, E., Stevens, L. A., Manzi, J., Kusek, J. W., Eggers, P., et al. (2007). Prevalence of chronic kidney disease in the United States. JAMA, 298, 2038–2047. de Boer, I. H., Katz, R., Cao, J. J., Fried, L. F., Kestenbaum, B., Mukamal, K., et al. (2009). Cystatin C, albuminuria, and mortality among older adults with diabetes. Diabetes Care, 32, 1833–1838. de Boer, I. H., Rue, T. C., Hall, Y. N., Heagerty, P. J., Weiss, N. S., & Himmelfarb, J. (2011). Temporal trends in the prevalence of diabetic kidney disease in the United States. JAMA, 305, 2532–2539. Deeb, A., Zaoui, P., Le, P. S., Tartry, D., Lantelme, P., Ducher, M., et al. (2012). Are triplicate urine samples necessary to assess albuminuria? Nephron Clinical Practice, 122, 80–85. DeVille, J., Smith, D. H., Johnson, E. S., Yang, X., Petrik, A. F., Weiss, J. R., et al. (2012). Predicting chronic kidney disease outcomes: Are two estimated glomerular filtration rates better than one? Population Health Management, 15, 113–118. Ferris, M., Shoham, D. A., Pierre-Louis, M., Mandhelker, L., Detwiler, R. K., & Kshirsagar, A. V. (2009). High prevalence of unlabeled chronic kidney disease among inpatients at a tertiary-care hospital. American Journal of the Medical Sciences, 337, 93–97. Garg, A. X., Kiberd, B. A., Clark, W. F., Haynes, R. B., & Clase, C. M. (2002). Albuminuria and renal insufficiency prevalence guides population screening: Results from the NHANES III. Kidney International, 61, 2165–2175. Grams, M. E., Plantinga, L. C., Hedgeman, E., Saran, R., Myers, G. L., Williams, D. E., et al. (2011). Validation of CKD and related conditions in existing data sets: A systematic review. American Journal of Kidney Diseases, 57, 44–54. Hallan, S. I., Matsushita, K., Sang, Y., Mahmoodi, B. K., Black, C., Ishani, A., et al. (2012). Age and association of kidney measures with mortality and end-stage renal disease. JAMA, 308, 2349–2360. Hornbrook, M. C., Hart, G., Ellis, J. L., Bachman, D. J., Ansell, G., Greene, S. M., et al. (2005). Building a virtual cancer research organization. Journal of the National Cancer Institute Monographs, 12–25. Joffe, M., Hsu, C. Y., Feldman, H. I., Weir, M., Landis, J. R., & Hamm, L. L. (2010). Variability of creatinine measurements in clinical laboratories: Results from the CRIC study. American Journal of Nephrology, 31, 426–434. Kern, E. F., Maney, M., Miller, D. R., Tseng, C. L., Tiwari, A., Rajan, M., et al. (2006). Failure of ICD-9-CM codes to identify patients with comorbid chronic kidney disease in diabetes. Health Services Research, 41, 564–580. Levey, A. S., Stevens, L. A., Schmid, C. H., Zhang, Y. L., Castro, A. F., III, Feldman, H. I., et al. (2009). A new equation to estimate glomerular filtration rate. Annals of Internal Medicine, 150, 604–612. Little, R. J., & Rubin, D. B. (2002). Statistical analysis with missing data. John Wiley & Sons, Inc (New York). Matsushita, K., Mahmoodi, B. K., Woodward, M., Emberson, J. R., Jafar, T. H., Jee, S. H., et al. (2012). Comparison of risk prediction using the CKD-EPI equation and the

7

MDRD study equation for estimated glomerular filtration rate. JAMA, 307, 1941–1951. Miller, W. G., & Bruns, D. E. (2009). Laboratory issues in measuring and reporting urine albumin. Nephrology, Dialysis, Transplantation, 24, 717–718. Nathan, D. M., Zinman, B., Cleary, P. A., Backlund, J. Y., Genuth, S., Miller, R., et al. (2009). Modern-day clinical course of type 1 diabetes mellitus after 30 years' duration: The diabetes control and complications trial/epidemiology of diabetes interventions and complications and Pittsburgh epidemiology of diabetes complications experience (1983-2005). Archives of Internal Medicine, 169, 1307–1316. National Kidney Foundation (2007). KDOQI clinical practice guidelines and clinical practice recommendations for diabetes and chronic kidney disease. American Journal of Kidney Diseases, 49, S1–S180. Nichols, G. A., Desai, J., Elston, L. J., Lawrence, J. M., O'Connor, P. J., Pathak, R. D., et al. (2012). Construction of a multisite DataLink using electronic health records for the identification, surveillance, prevention, and management of diabetes mellitus: The SUPREME-DM Project. Preventing Chronic Disease, 9, E110. Nichols, G. A., Schroeder, E. B., Karter, A. J., Gregg, E. W., Desai, J., Lawrence, J. M., et al. (2015). Trends in diabetes incidence among 7 million insured adults, 2006-2011: The SUPREME-DM Project. American Journal of Epidemiology, 181, 32–39. Ninomiya, T., Perkovic, V., de Galan, B. E., Zoungas, S., Pillai, A., Jardine, M., et al. (2009). Albuminuria and kidney function independently predict cardiovascular and renal outcomes in diabetes. Journal of the American Society of Nephrology, 20, 1813–1821. Pugliese, G., Solini, A., Fondelli, C., Trevisan, R., Vedovato, M., Nicolucci, A., et al. (2011). Reproducibility of albuminuria in type 2 diabetic subjects. Findings from the Renal Insufficiency And Cardiovascular Events (RIACE) study. Nephrology, Dialysis, Transplantation, 26, 3950–3954. Saydah, S. H., Pavkov, M. E., Zhang, C., Lacher, D. A., Eberhardt, M. S., Burrows, N. R., et al. (2013). Albuminuria prevalence in first morning void compared with previous random urine from adults in the national health and nutrition examination survey, 2009-2010. Clinical Chemistry, 59, 675–683. Shahinian, V. B., Hedgeman, E., Gillespie, B. W., Young, E. W., Robinson, B., Hsu, C. Y., et al. (2013). Estimating prevalence of CKD stages 3-5 using health system data. American Journal of Kidney Diseases, 61, 930–938. Stevens, L. A., Coresh, J., Greene, T., & Levey, A. S. (2006). Assessing kidney function– measured and estimated glomerular filtration rate. New England Journal of Medicine, 354, 2473–2483. Stevens, L. A., Fares, G., Fleming, J., Martin, D., Murthy, K., Qiu, J., et al. (2005). Low rates of testing and diagnostic codes usage in a commercial clinical laboratory: Evidence for lack of physician awareness of chronic kidney disease. Journal of the American Society of Nephrology, 16, 2439–2448. Tuttle, K. R., Bakris, G. L., Bilous, R. W., Chiang, J. L., de Boer, I. H., Goldstein-Fuchs, J., et al. (2014). Diabetic kidney disease: A report from an ADA Consensus Conference. Diabetes Care, 37, 2864–2883. US Renal Data System (2013). USRDS 2013 Annual Data Report: Atlas of chronic kidney disease and end-state renal disease in the United States. Bethesda, MD: US Renal Data System. Winkelmayer, W. C., Schneeweiss, S., Mogun, H., Patrick, A. R., Avorn, J., & Solomon, D. H. (2005). Identification of individuals with CKD from Medicare claims data: A validation study. American Journal of Kidney Diseases, 46, 225–232.

Please cite this article as: Schroeder, E.B., et al., Prevalence of chronic kidney disease among individuals with diabetes in the SUPREME-DM Project, 2005–2011, Journal of Diabetes and Its Complications (2015), http://dx.doi.org/10.1016/j.jdiacomp.2015.04.007