Data are based on outpatient, nonemergency department serum creatinine level closest to dialysis initiation date. Error bars indicate 95% CIs.
Error bars indicate 95% CIs.
aP < .001 for trend.
See Table 2 for unadjusted P values.
aP < .05 for trend.
bP < .01 for trend.
eTable. Unadjusted and Adjusted Associations for Calendar Year Trend in the Annual Risk of Initiating Elective Dialysis, by Index eGFR Category at Kaiser Permanente Northern California, 2001-2018, by 3-Year Interval
eFigure 1. Mean (95% Confidence Interval) Estimated Glomerular Filtration Rate at Dialysis Initiation Among 4,865 Patients Initiating Dialysis Within 1 Year Between 2001-2018, by 3-Year Interval: Elective vs Urgent Dialysis and Peritoneal vs Hemodialysis
eFigure 2. One-Year Risk of Initiating Dialysis (Overall) by Calendar Period, by Index Estimated Glomerular Filtration Rate, 2001-2018
Customize your JAMA Network experience by selecting one or more topics from the list below.
Hsu C, Parikh RV, Pravoverov LN, et al. Implication of Trends in Timing of Dialysis Initiation for Incidence of End-stage Kidney Disease. JAMA Intern Med. 2020;180(12):1647–1654. doi:10.1001/jamainternmed.2020.5009
How are temporal changes in the level of estimated glomerular filtration rate (eGFR) at which long-term dialysis is initiated among at-risk patients associated with the number of patients with end-stage kidney disease in the population?
This cohort study of 983 122 individuals in the initial 3-year interval (2001-2003) to 1 844 317 individuals in the final interval (2016-2018) used data from a large, integrated health care delivery system in Northern California and found that an increase in the number of individuals starting dialysis with an eGFR of 10 to 24 mL/min/1.73 m2 was associated with changes over time in the likelihood of receiving dialysis at this eGFR independent of the number of people in the underlying population who had this same eGFR. Estimated incidence of new end-stage kidney disease cases would have been 16% lower with no changes in system-level practice patterns or other factors besides timing of long-term dialysis initiation.
The timing of long-term dialysis initiation is associated with the number of individuals with end-stage kidney disease.
In the last 2 decades, there have been notable changes in the level of estimated glomerular filtration rate (eGFR) at which patients initiate long-term dialysis in the US and around the world. How changes over time in the likelihood of dialysis initiation at any given eGFR level in at-risk patients are associated with the population burden of end-stage kidney disease (ESKD) has not been not well defined.
To examine temporal trends in long-term dialysis initiation by level of eGFR and to quantify how these patterns are associated with the number of patients with ESKD.
Design, Setting, and Participants
Retrospective cohort study analyzing data obtained from a large, integrated health care delivery system in Northern California from 2001 to 2018 in successive 3-year intervals. Included individuals, ranging in number from as few as 983 122 (2001-2003) to as many as 1 844 317 (2016-2018), were adult members with 1 or more outpatient serum creatinine levels determined in the prior year.
Main Outcomes and Measures
One-year risk of initiating long-term dialysis stratified by eGFR levels. Multivariable logistic regression was performed to assess temporal trends in each 3-year cohort with adjustment for age, sex, race, and diabetes status. The potential change in dialysis initiation in the final cohort (2016-2018) was estimated using the relative difference between the standardized risks in the initial cohort (2001-2003) and the final cohort.
In the initial 3-year cohort, the mean (SD) age was 55.4 (16.3) years, 55.0% were women, and the prevalence of diabetes was 14.9%. These characteristics, as well as the distribution of index eGFR, were stable across the study period. The likelihood of receiving dialysis at eGFR levels of 10 to 24 mL/min/1.73 m2 generally increased over time. For example, the 1-year odds of initiating dialysis increased for every 3-year interval by 5.2% (adjusted odds ratio, 1.052; 95% CI, 1.004-1.102) among adults with an index eGFR of 20 to 24 mL/min/1.73 m2, by 6.6% (adjusted odds ratio, 1.066; 95% CI, 1.007-1.130) among adults with an eGFR of 16 to 17 mL/min/1.73 m2, and by 5.3% (adjusted odds ratio, 1.053; 95% CI, 1.008-1.100) among adults with an eGFR of 10 to 13 mL/min/1.73 m2, adjusting for age, sex, race, and diabetes. The incidence of new cases of ESKD was estimated to have potentially been 16% (95% CI, 13%-18%) lower if there were no changes in system-level practice patterns or other factors besides timing of initiating long-term dialysis from the initial 3-year interval (2001-2003) to the final interval (2016-2018) assessed in this study.
Conclusions and Relevance
The present results underscore the importance the timing of initiating long-term dialysis has on the size of the population of individuals with ESKD.
In the last 2 decades, there have been notable changes in the estimated glomerular filtration rate (eGFR) at which long-term dialysis is initiated in the US and around the world.1 For example, according to the nationally comprehensive US Renal Data System registry, the proportion of new patients with end-stage renal disease in the US with an eGFR of 10 to 15 mL/min/1.73 m2 at the start of dialysis increased from 10% in 1996 to a peak of 28% in 2010.2 Although several previous publications have examined dialysis initiation over time, a key remaining knowledge gap is the distribution of eGFR in the underlying at-risk population from which the end-stage kidney disease (ESKD) cases are derived. Thus, secular trends in the likelihood of dialysis initiation at any given eGFR level could not be defined in previous studies, and the implications of potential trends in changes in the size of the at-risk population on the number of patients with ESKD in the population could not be estimated.
Nationally, the societal and patient burden of ESKD is high, with Medicare fee-for-service spending for beneficiaries with ESKD rising from $33.8 billion to $35.4 billion between 2015 and 2016.2 Furthermore, national rates of morbidity and death are very high surrounding initiation of long-term dialysis.3-5 Initiation of long-term dialysis can also be associated with declines in cognitive function and functional status6,7 and worsening cardiac structure.8 The July 2019 Advancing American Kidney Health initiative issued by the White House targeted as a major national priority reducing the number of new ESKD cases by 25% by 2030.9
Because ESKD is operationally defined as receiving long-term dialysis or a kidney transplant, we hypothesized that the number of patients with incident ESKD can vary considerably depending on decisions between patients and physicians about the choice and timing of kidney replacement therapy, and particularly of initiating dialysis. Consistent with this, after publication of the Initiating Dialysis Early and Late (IDEAL) trial—a randomized clinical trial showing no significant difference in death or other outcomes with earlier (eGFR of 9.0 mL/min/1.73 m2) vs later (eGFR of 7.2 mL/min/1.73 m2) initiation of dialysis10—a rapid drop in early dialysis initiation was observed in Canada.11 Although early dialysis initiation was defined in that study as dialysis initiation at eGFR higher than 10.5 mL/min/1.73 m2, there was no information regarding how the number of patients with chronic kidney disease (CKD) in the underlying population with eGFR higher than 10.5 mL/min/1.73 m2 may have changed during the relevant period.
To date, no study has quantified how secular patterns in the likelihood of dialysis initiation at any given eGFR level can impact the number of patients with incident ESKD in a population over time. We addressed this important knowledge gap by examining temporal trends in dialysis initiation by level of kidney function within a large, community-based population to provide information relevant for policies supporting the national advancement of kidney health.
Kaiser Permanente Northern California (KPNC) is a large, integrated health care delivery system currently providing comprehensive care for more than 4.5 million members. Its membership is sociodemographically diverse and highly representative of the local and statewide population.12 Nearly all aspects of care are captured through KPNC’s integrated electronic health record system, with access to clinical data elements from all practice settings (eg, laboratory test results and vital signs) as well as previously validated algorithms that incorporate relevant diagnostic or procedure codes and pharmacy dispenses. This study was approved by the KPNC institutional review board, which also waived the requirement for obtaining informed consent because this study was a retrospective data-only analysis of linked data that were deidentified before analysis.
We assembled cohorts of all adult members (18 years or older) with at least 1 valid outpatient serum creatinine measurement from 2001 through 2018 grouped into 3-year intervals (ie, 2001-2003, 2004-2006, 2007-2009, 2010-2012, 2013-2015, and 2016-2018), which provided more stable and precise estimates of ESKD rates. Individuals with 12 or more months of continuous health plan membership were included in 3-year cohorts on the first eligible January 1st within each 3-year interval. Approximately 15% of eligible patients were excluded in each cohort owing to insufficient prior continuous membership. Patients with a prior kidney transplant or who were already undergoing long-term dialysis were also excluded (approximately 5% of each cohort).
Our primary outcome was initiation of long-term dialysis within the first eligible year in each 3-year interval, ascertained through a comprehensive, manually validated health plan ESKD treatment registry.13 Patients are entered into the registry after manual review to confirm that receipt of dialysis was not temporary or only for acute kidney injury. We removed cases of preemptive kidney transplant (3% of ESKD cases). Initial dialysis modality (hemodialysis vs peritoneal dialysis) was ascertained from registry files. We defined urgent dialysis starts as being preceded by an episode of dialysis-requiring acute kidney injury13 within 28 days before long-term dialysis initiation. All cases not meeting this definition were considered elective starts.
Age, sex, and self-reported race were obtained from electronic health records to calculate eGFR and to account for potential changes in the source population. Targeted comorbidities in the study population were defined using validated International Classification of Diseases, Ninth Revision and Tenth Revision diagnostic codes, laboratory results, and receipt of medications using electronic health record–based data that were cleaned and linked at the individual-patient level into the KPNC Virtual Data Warehouse as previously described and validated.14-17 We ascertained outpatient laboratory test results, including the CKD Epidemiology Collaboration equation–based eGFR and proteinuria.18
For each 3-year cohort, we calculated the 1-year risk of initiating long-term dialysis, overall and by urgent vs elective starts, directly standardized to the 2001 to 2003 KPNC population by age, sex, and index eGFR (ie, most recent outpatient measurement before cohort entry). We estimated a potential change in dialysis initiation in the final cohort (2016-2018) using the relative difference between the standardized risks and associated 95% CIs in the initial cohort (2001-2003) and the final cohort.
We next stratified the study population into a priori selected index eGFR levels of 60 to 150, 30 to 59, 25 to 29, 20 to 24, 18 to 19, 16 to 17, 14 to 15, 10 to 13, 6 to 9, and 0 to 5 mL/min/1.73 m2. Within each eGFR stratum, we calculated the 1-year risk of receiving long-term dialysis across each 3-year interval. All crude temporal trends were evaluated using Cochrane-Armitage tests for trend. To assess multivariable-adjusted temporal trends, we evaluated the significance and odds ratios (95% CIs) of an ordinal 3-year cohort term in a logistic regression model adjusting for age, sex, race, and diabetes status and used a generalized estimating equations approach to account for correlations between repeated individuals across calendar cohorts.
To enhance comparability with previous studies, we also report the mean (SD) eGFR at dialysis initiation in each time interval, using the most recent outpatient eGFR measurement before dialysis initiation.2 This eGFR value is separate from the index eGFR value used for assembly of calendar year cohorts.
All data were analyzed using SAS, version 9.4 (SAS Institute Inc). A 2-sided P < .05 was considered statistically significant.
Between 2001 and 2018, the size of the eligible population increased consistent with growth in overall KPNC enrollment and more frequent outpatient serum creatinine testing per patient (Table 1). In the first 3-year interval (2001-2003) the mean (SD) age in the eligible population was 55.4 (16.3) years, 55.0% were women, and the prevalence of diabetes was 14.9%. These characteristics as well as the distribution of index eGFR were stable across the study period. The prevalence of documented hypertension and dyslipidemia increased since 2001, although this was largely due to increased detection and changing systemwide practice guidelines. In later years, we also observed a higher prevalence of Asian/Pacific Islander and Hispanic patients.
Overall, the temporal trend in mean eGFR at dialysis initiation mirrored that seen in the general US population,19 with a progressively higher mean eGFR through the first decade of the 21st century, followed by a plateau (Figure 1). Mean (SD) eGFR at dialysis initiation for patients starting long-term dialysis was 12.4 (13.1) mL/min/1.73 m2 in 2001 to 2003, rose to 16.3 (17.7) mL/min/1.73 m2 in 2010 to 2012, and did not further increase through 2016 to 2018. These temporal trends were similar for elective starts and for all patients undergoing peritoneal dialysis or hemodialysis (eFigure 1 in the Supplement).
In our study population, the number of eligible patients who started dialysis was 847 (0.086% of 983 122 at risk) in 2001 to 2003, 917 (0.074% of 1 241 537) in 2004 to 2006, 1011 (0.072% of 1 411 690) in 2007 to 2009, 1013 (0.068% of 1 482 883) in 2010 to 2012, 1092 (0.068% of 1 607 737) in 2013 to 2015, and 1224 (0.066% of 1 844 317) in 2016 to 2018. The proportion of elective ESKD cases was 85.6% in 2001 to 2003, 89.2% in 2004 to 2006, 94.1% 2007 to 2009, 74.9% in 2010 to 2012, 87.2% in 2013 to 2015, and 84.0% in 2016 to 2018. In addition, the proportion of new ESKD cases initiating peritoneal dialysis was 13.7% in 2001 to 2003, 14.2% in 2004 to 2006, 11.0% in 2007 to 2009, 13.4% in 2010 to 2012, 21.7% in 2013 to 2015, and 20.8% in 2016 to 2018. As recently reported,20 KPNC has made a concerted systemwide effort to increase peritoneal dialysis as the preferred initial modality since approximately 2008.
The likelihood of long-term dialysis initiation within 1 year increased from 0.086% (95% CI, 0.080%-0.092%) in 2001 to 2003, to 0.103% (95% CI, 0.099%-0.107%) in 2016 to 2018 (P < .001 for trend) (Figure 2). When examining the annual risk of initiating long-term dialysis by index eGFR between 2001 and 2018, the most prominent secular trends were noted in those with an index eGFR of 20 to 24 mL/min/1.73 m2 (from 3.2% to 5.1%; P = .001), those with an index eGFR of 18 to 19 mL/min/1.73 m2 (from 8.3% to 11.0%; P = .03), those with an index eGFR of 16 to 17 mL/min/1.73 m2 (from 12.4% to 17.9%; P = .005), and those with an index eGFR of 10 to 13 mL/min/1.73 m2 (from 34.2% to 40.3%; P = .03) (Figure 3; eFigure 2 in the Supplement). After adjustment for age, sex, race, and diabetes, statistically significant temporal increases in 1-year odds of initiating dialysis persisted among eGFR categories of 20 to 24, 16 to 17, and 10 to 13 mL/min/1.73 m2 (Table 2). Among patients with an index eGFR of 20 to 24 mL/min/1.73 m2, the 1-year odds of initiating dialysis increased for every 3-year interval by 5.2% (adjusted odds ratio, 1.052; 95% CI, 1.004-1.102); among patients with an eGFR 16 to 17 mL/min/1.73 m2, the 1-year odds of initiating dialysis increased by 6.6% (adjusted odds ratio, 1.066; 95% CI, 1.007-1.130); and among patients with an eGFR of 10 to 13 mL/min/1.73 m2, the 1-year odds of initiating dialysis increased by 5.3% (adjusted odds ratio, 1.053; 95% CI, 1.008-1.100) (Table 2). By contrast, among patients with very low index eGFR (≤9 mL/min/1.73 m2), we observed stable or decreased unadjusted and adjusted odds of initiating dialysis over time. These patterns were similar among the subset of patients with elective dialysis starts (eTable in the Supplement).
Overall, we estimated that incidence of ESKD could have potentially been 16% (95% CI, 13%-18%) lower if there were no changes in system-level practice patterns or other factors besides timing of initiation of long-term dialysis from the initial 3-year interval (2001-2003) to the final 3-year interval (2016-2018) assessed in this study.
In this study using a large, integrated health care delivery system that facilitated accurate determination of the size and eGFR distribution of the underlying source population, we estimated that incidence of ESKD could have potentially been 16% (95% CI, 13%-18%) lower if there were no changes in system-level practice patterns or other factors besides timing of initiation of long-term dialysis from the initial 3-year interval (2001-2003) to the final 3-year interval (2016-2018) assessed in this study.
The optimal time to initiate long-term dialysis in patients with kidney disease remains unclear,10,21-27 with a high degree of variability in practice patterns over time and across health systems.1 Among 11 215 older US veterans with sustained eGFR lower than 15 mL/min/1.73 m2, Tamura et al28 observed that patients exclusively using fee-for-service Medicare for nephrology care had 28% higher frequency of dialysis initiation compared with those exclusively using Department of Veterans Affairs nephrology care. Based on US Renal Data System data, the percentage of patients with incident kidney failure who initiated dialysis at an eGFR of 10 mL/min/1.73 m2 or higher increased steadily from the mid-1990s (13%) until 2010 (43%), when it remained stable or slightly declined.2 Similar temporal trends have been observed in the Veterans Affairs health system29 and in other countries, such as the United Kingdom.30
We hypothesize that the level of eGFR at which long-term dialysis was started during the 2 decades internationally has been influenced by a variety of system-, physician-, and patient-level factors.1 On a national level, perspectives from opinion leaders31 combined with high-profile, consensus-based clinical practice guidelines32,33 contributed to an emphasis on recommending a “healthy/timely” dialysis,1 which included efforts to promote the placement of arteriovenous fistulas at higher eGFR levels.34 This may have contributed to patients accepting dialysis at relatively higher eGFR levels during the present study period, especially if they had a mature arteriovenous fistula and symptoms that are challenging to manage medically. In addition, because KPNC promoted the use of peritoneal dialysis as the initial modality during the study period, the need for a significant amount of time to train patients and caregivers may have also contributed, in part, to our observed patterns of dialysis starts at a higher eGFR. The concurrent expansion of educational efforts to raise awareness of kidney disease and empower patients with CKD regarding their options may have affected patterns in timing of initiation of dialysis. Of note, as a fully integrated health care delivery system, KPNC had no financial incentives to initiate dialysis earlier vs later.
An alternative possibility that would also lead to the identical observation of an increase in the number (and percentage) of individuals starting dialysis with an eGFR of 10 to 24 mL/min/1.73 m2 is that there are relatively more people in the underlying source population who had an eGFR of 10 to 24 mL/min/1.73 m2 in more recent years, with no change over time in the likelihood of receiving dialysis at this eGFR.29 A recent interrupted time series analysis provided circumstantial evidence that the progressive increase in mean eGFR level at the start of dialysis may likely be driven more by secular trends in practice patterns or other factors and not by secular trends in the distribution of eGFR in the underlying at-risk population. Ferguson et al11 showed that after publication of the IDEAL trial in 2010, which found no systematic clinical benefit for initiation of earlier vs later dialysis, there was an immediate decrease in the proportion of early dialysis starts (ie, at eGFR >10.5 mL/min/1.73 m2) in Canada. However, Ferguson et al did not have data on the underlying at-risk Canadian population. By contrast, a major strength of our study is that we captured the underlying at-risk population whose distribution of eGFR was known.
Few prior studies have attempted to quantify the public health and cost implications of changes in timing of dialysis initiation. Using US Renal Data System data and observed rates of eGFR loss from a CKD cohort, O’Hare et al35 estimated that long-term dialysis was initiated nationally a mean of approximately 5 months earlier (and approximately 8 months earlier for those aged ≥75 years) in 2007 vs 1997, leading to more than $1.5 billion of cost annually. However, that study did not consider the association of changes in timing of dialysis initiation with the number of incident ESKD cases, which our present analysis showed is notable. To put the observed relative 16% change between 2001 and 2018 in context, this is approximately two-thirds of the target relative 25% reduction in new ESKD cases by 2030 called for in the 2019 White House Advancing American Kidney Health initiative.9 In addition, if there was a 16% smaller ESKD population, the estimated cost difference could be up to $2.2 billion because Medicare Part B spending for total outpatient dialysis–related services was $13.7 billion in 2016 to 2017.36 This cost is likely an underestimate because it does not include indirect costs of care (eg, patient and caregiver travel time; costs associated with creating and maintaining functioning dialysis access; potential loss of employment; and costs associated with caregiver and patient burnout).
Our study has several important implications. Given the lack of strong evidence that earlier initiation of long-term dialysis has net clinical benefit,10,21-27 careful evaluation of contemporary dialysis initiation practices is needed,35 along with ongoing improved shared decision-making efforts to personalize the approach with patients.34 Dialysis is currently recommended when kidney function is “low” and a patient has symptoms or signs consistent with kidney failure. However, some symptoms and signs (eg, fatigue or pruritus) may be nonspecific and caused by comorbid conditions, whereas others (eg, anorexia or hypoalbuminemia) can change over time and may improve spontaneously without dialysis therapy. In most cases, long-term dialysis is initiated not to treat an immediately life-threatening condition (eg, uremic pericarditis or refractory hyperkalemia). As noted previously, the aggressive promotion of earlier arteriovenous fistula placement in patients with CKD may also affect patient and physician expectations about the timing of initiating dialysis. Thus, decisions about when to start dialysis are not always fully objective, and policy decisions that have encouraged systematic earlier dialysis initiation can be adjusted. For example, O’Hare et al29 suggested reimposing scrutiny by the Health Care Financing Administration and renal networks of an upper eGFR limit at which elective dialysis initiation would trigger a review. Following publication of the IDEAL trial, there have also been calls for starting dialysis later.37 The United Kingdom National Institute for Health and Care Excellence now recommends waiting until an eGFR of 5 to 7 mL/min/1.73 m2 or the presence of symptoms affecting daily life before starting dialysis.38 Furthermore, the Japanese Society of Nephrology guidelines recommend patients “endure under conservative treatment until the GFR is <8 mL/min/1.73 m2 even when the symptoms of renal failure are observed.”39(p97)
Our results also support the perspective that practice patterns or other factors beyond just population CKD burden may impact the number of new ESKD cases. Thus, incidence of ESKD (as currently defined) in a population may not be an entirely optimal quality metric to reflect the success or failure of renoprotective efforts, and policy makers and public health officials should consider other metrics as well.40
Our study was limited in that we only assessed the association between the timing of dialysis initiation and incident ESKD rates within an integrated health care delivery system in California. However, the KPNC membership is sociodemographically diverse and highly representative of the local and statewide population.12 Supporting the external validity of our findings is the observation that the temporal trend in mean eGFR at the start of dialysis among KPNC members—progressively increasing through the first decade of this century (Figure 1)—matches that reported by the US Renal Data System,2 US Department of Veterans Affairs,29 Canadian Organ Replacement Register,11 and United Kingdom Renal Registry.30 We did not have information on the specific system-, physician-, and patient-level reasons that may have contributed to the observed trends, nor did we analyze how changing the timing and threshold for dialysis initiation may have affected clinical outcomes (including in frail25 or older41 persons). However, previous studies do not provide convincing evidence that systematically starting dialysis earlier confers net benefits,10,26,27,42 including the only large randomized trial performed to date.10
In conclusion, choices regarding the timing of dialysis initiation should be made on a patient-by-patient basis to maximize the net benefit for individual patients with kidney disease, including taking into account the results of large randomized clinical trials, such as the IDEAL trial.10 Our results underscore the importance the timing of initiating long-term dialysis has on the size of the population of patients with ESKD.
Accepted for Publication: August 4, 2020.
Corresponding Author: Chi-yuan Hsu, MD, MSc, Division of Nephrology, Department of Medicine, University of California, San Francisco, 533 Parnassus Avenue, Box 0532, U-400, San Francisco, CA 94143 (email@example.com).
Published Online: October 12, 2020. doi:10.1001/jamainternmed.2020.5009
Author Contributions: Dr Go and Mr Parikh had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Hsu, Tan, Go.
Acquisition, analysis, or interpretation of data: Hsu, Parikh, Pravoverov, Zheng, Glidden, Go.
Drafting of the manuscript: Hsu, Parikh, Go.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Hsu, Parikh, Glidden, Go.
Obtained funding: Hsu.
Administrative, technical, or material support: Hsu, Parikh, Pravoverov, Zheng, Tan, Go.
Supervision: Pravoverov, Tan, Go.
Conflict of Interest Disclosures: Dr Hsu is the medical director of the San Francisco Wellbound dialysis unit. He reported receiving research funding from Satellite Healthcare Inc to support an investigator-initiated research project and grants from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) during the conduct of the study. Dr Zheng reported being a senior partner of The Permanente Medical Group, outside the submitted work. Dr Go reported receiving grants from NIDDK during the conduct of the study. No other disclosures were reported
Funding/Support: This study was supported by award K24 DK092291 from NIDDK and funding from the Kaiser Permanente Northern California Division of Research.
Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.