eFigure 1. Study Cohort Selection
eFigure 2. Percentage of Hemodialysis Patients Initiating Dialysis with an Arteriovenous Access Over Time
eTable 1. Parameter Estimates and 95% Confidence Intervals of Segmented Regression Models
eTable 2. Results of Sensitivity Analyses
eTable 3. Results of Segmented Regression Models in the Complete Case Scenario (n = 22,140)
Customize your JAMA Network experience by selecting one or more topics from the list below.
Ferguson TW, Garg AX, Sood MM, et al. Association Between the Publication of the Initiating Dialysis Early and Late Trial and the Timing of Dialysis Initiation in Canada. JAMA Intern Med. 2019;179(7):934–941. doi:10.1001/jamainternmed.2019.0489
Is the dissemination of the Initiating Dialysis Early and Late randomized clinical trial associated with changes in the proportion of patients who initiate dialysis early in Canada?
In this interrupted time series analysis involving 28 468 patients with incident dialysis, a statistically significant immediate decrease in the proportion of early dialysis initiations was observed after trial publication, and a change in the initiation trend was noted between the pretrial and posttrial periods. No statistically significant differences were observed in acute inpatient dialysis initiations, the proportion of patients receiving home dialysis as the initial modality, or the proportion of arteriovenous access creation at hemodialysis initiation.
The Initiating Dialysis Early and Late trial appeared to be associated with an immediate and sustained change in the timing of dialysis initiation in Canada.
Published in 2010, the Initiating Dialysis Early and Late (IDEAL) randomized clinical trial, which randomized patients with an estimated glomerular filtration rate (GFR) between 10 and 15 mL/min/1.73 m2 to planned initiation of dialysis with an estimated GFR between 10 and 14 mL/min/1.73 m2 (early start) or an estimated GFR between 5 and 7 mL/min/1.73 m2 (late start), concluded that early initiation was not associated with improved survival or clinical outcomes.
To assess the association between the IDEAL trial results and the proportion of early dialysis starts over time.
Design, Setting, and Participants
This interrupted time series analysis used data from the Canadian Organ Replacement Register to study adult (≥18 years of age) patients with incident chronic dialysis between January 1, 2006, and December 31, 2015, in Canada, which has a universal health care system. Patients from the province of Quebec were excluded because its privacy laws preclude submission of deidentified data without first-person consent. The patients included in the study (n = 28 468) had at least 90 days of nephrologist care before starting dialysis and a recorded estimated GFR at dialysis initiation. Data analyses were performed from November 2016 to January 2019.
Main Outcomes and Measures
The primary outcome was the proportion of early dialysis starts (estimated GFR >10.5 mL/min/1.73 m2), and the secondary outcomes included the proportions of acute inpatient dialysis starts, patients who started dialysis using a home modality, and patients receiving hemodialysis who started with an arteriovenous access. Measures included the trend prior to the IDEAL trial publication, the change in this trend after publication, and the immediate consequence of publication.
The final cohort comprised 28 468 patients, of whom 17 342 (60.9%) were male and the mean (SD) age was 64.8 (14.6) years. Before the IDEAL trial, a statistically significant increasing trend was observed in the monthly proportion of early dialysis starts (adjusted rate ratio, 1.002; 95% CI, 1.001-1.004; P = .004). After the IDEAL trial, an immediate decrease was observed in the proportion of early dialysis starts (rate ratio, 0.874; 95% CI, 0.818-0.933; P < .001), along with a statistically significant change in trend between the pretrial and posttrial periods (rate ratio, 0.994; 95% CI, 0.992-0.996; P < .001). No statistically significant differences were found in acute inpatient dialysis initiations, the proportion of patients receiving home dialysis as the initial modality, or the proportion of arteriovenous access creation at hemodialysis initiation after the IDEAL trial publication.
Conclusions and Relevance
The publication of the IDEAL trial appeared to be associated with an immediate and meaningful change in the timing of dialysis initiation in Canada.
Dialysis as a treatment for kidney failure has been widely and successfully applied for more than 4 decades; however, the optimal time to initiate dialysis remains elusive. Once dialysis is initiated, renal recovery is rare and patients experience high risks of morbidity and mortality as well as a diminished quality of life.1 In addition, early initiation of dialysis is costly and may not provide optimal value from a health system perspective.2 From 1990 to 2010, physicians began to promote dialysis initiation at higher levels of estimated glomerular filtration rate (GFR), partly because of a desire to maintain physical and cognitive functions before dialysis initiation and possibly because of other health system incentives.3,4
In August 2010, the Initiating Dialysis Early and Late (IDEAL) randomized clinical trial was published. The trial randomized participants with predialysis chronic kidney disease to planned initiation of dialysis in the estimated GFR (Cockroft-Gault) range of 10 to 14 mL/min/1.73 m2 (early start) compared with initiation of dialysis when the estimated GFR was in the range of 5 to 7 mL/min/1.73 m2 (late start).5 (To convert estimated GFR to milliliter per second per square meter, multiply by 0.0167.) The trial concluded that earlier initiation of dialysis was not associated with a statistically significant difference in survival (hazard ratio of death, 1.04; 95% CI, 0.83-1.30) or other clinical outcomes, including cardiovascular events and infections.5
Despite issues of crossover between the treatment arms and concerns about increasing suboptimal starts with delayed initiation, the findings from the IDEAL trial were consistent with those of subsequent and previous studies in the observational literature and confirmed no advantages of early dialysis initiation for either patient or health system outcomes.6-12 As such, the treatment guidelines for dialysis initiation recommended deferral until the estimated GFR decreases below 6 mL/min/1.73 m2 or until symptoms arise.13
Adoption of randomized trial findings and clinical practice guidelines into routine clinical practice may take years. Barriers and delays in knowledge translation include a lack of awareness or confidence in study findings, difficulty implementing findings, and financial incentives that can discourage adoption. Conversely, successful rapid implementation of guidelines and best care practices may be associated with increased effectiveness and efficiency for the health care system through the improvement of quality and appropriateness of patient care and through the reduction of clinical practice variations.14
This study aimed to identify the association between the IDEAL trial publication5 and the proportion of patients who initiate dialysis early (estimated GFR>10.5 mL/min/173m2) in Canada.6,8,15 The study used data collected from the Canadian Organ Replacement Register (CORR).
Information on dialysis starts was collected from CORR, a government-mandated and validated dialysis registry that tracks information on organ replacement therapy in Canada and stores information on organ transplant and dialysis activity from treatment initiation until death.16 We excluded patients from the province of Quebec because its privacy laws preclude submission of deidentified data without first-person consent; as such, data are largely delayed or incomplete. Information collected in CORR includes patient demographics (age and sex), clinical characteristics (body mass index, comorbidities, and origin of kidney failure), laboratory measurements at dialysis initiation, whether nephrologist care was provided prior to dialysis and the duration of predialysis care, date of dialysis initiation, location of treatment, dialysis modality at initiation (eg, home- or facility-based dialysis), and dialysis access.16 We used the last available estimated GFR prior to dialysis initiation in mL/min/1.73 m2, which was estimated using the Modification of Diet in Renal Disease equation.17 The setting of first dialysis treatment (eg, outpatient or inpatient start) was obtained by linking CORR to the Discharge Abstract Database18 to determine hospitalizations.
The primary outcome in this study was the change in the proportion of early dialysis starts, defined as having an estimated GFR at dialysis initiation greater than 10.5 mL/min/1.73 m2. Secondary outcomes included the change in the proportion of urgent or emergent dialysis starts, defined as first dialysis treatment as an acute inpatient (1776 patients from Manitoba were excluded from this analysis because of differences in health card number encryption that did not allow for linkage between CORR and the Discharge Abstract Database); the proportion of patients starting dialysis using a home modality (home peritoneal dialysis or home hemodialysis); and the proportion of patients receiving hemodialysis starting dialysis with an arteriovenous fistula (AVF) or arteriovenous grafts (AVG) compared with a central venous catheter.
All analyses were performed from November 2016 to January 2019 with SAS software, version 9.4 (SAS Institute Inc). Descriptive statistics for the study population were provided, with continuous variables presented as means (SDs) for normally distributed variables and medians (interquartile ranges [IQRs]) for nonnormal variables, and categorical variables presented as percentages. Comparisons between the pretrial and posttrial periods were performed, depending on the distribution of the outcome. For example, continuous variables were compared using unpaired, 2-tailed t tests or Mann-Whitney tests when appropriate, whereas categorical variables were compared using the χ2 test.
For evaluation of primary and secondary outcomes, generalized estimating equation models were applied with a log link, negative binomial distribution, and an autoregressive covariance structure. Models were adjusted for patient-level demographics (age, sex, body mass index) and differences in comorbid conditions. Missing data in adjusted models were imputed using PROC MI in the SAS software with 5 imputations. The final model was structured as an interrupted time series analysis with outcomes as proportions determined at monthly intervals. The model included the trend before the IDEAL trial publication, an evaluation of the change in this trend after publication, and the immediate consequence of publication.
The pretrial period lasted 56 months, from January 1, 2006, to August 31, 2010. After publication of the trial in August 2010,5 a 6-month grace period was included in the model, between September 1, 2010, and February 28, 2011, to allow for passive dissemination of the publication; these 6 months were excluded from the baseline analysis. The posttrial period was 58 months long, from March 1, 2011, to December 31, 2015.
Results of the regression analyses were presented as adjusted rate ratios. The rate ratio for the variable representing the pretrial trend can be interpreted as the monthly change in the slope of the considered outcome. The rate ratio for the change in trend represents the change in the slope of the outcome between the pretrial and posttrial periods. For the immediate effect variable, the rate ratio represents the decrease in proportion immediately observed after publication and the grace period.19,20 All results were presented with CIs, assuming α = .05.
We performed several sensitivity analyses. We evaluated the outcome of removing the 6-month grace period, such that September 1, 2010, was considered the beginning of the posttrial period, and then including an additional grace period before publication of the trial to account for earlier dissemination (eg, conference presentations), such that March 1, 2010, was considered the end of the pretrial period. We increased the length of nephrologist care required before starting dialysis for inclusion in the study to 120 days in the pretrial period and to 180 days in the posttrial period to evaluate the association with nephrologist involvement. We performed a subgroup analysis in patients who initiated dialysis as elective outpatients. A further sensitivity analysis was performed on the estimated GFR threshold used to define early initiation (baseline 10.5 mL/min/1.73 m2) with values of 9.5 mL/min/1.73 m2 and 11.5 mL/min/1.73 m2. For adjusted models, we analyzed the complete case, excluding imputations.
In total, 43 472 patients with incident dialysis were recorded in CORR between January 1, 2006, and December 31, 2015. Of these, 13 725 patients were excluded for having fewer than 90 days of nephrologist care before initiating dialysis, 981 were excluded for missing estimated GFR data at dialysis initiation, and 298 were excluded for having a recorded estimated GFR at dialysis initiation higher than 30 mL/min/1.73 m2. The final cohort comprised 28 468 patients (eFigure 1 in the Supplement), of whom 17 342 (60.9%) were male and the mean (SD) age was 64.8 (14.6) years. Compared with the pretrial population, patients in the posttrial period were more likely to be male; had lower serum hemoglobin, lower serum albumin, and higher serum phosphate levels at initiation of dialysis; had more days of predialysis care; had higher body mass index; and had more comorbid conditions (Table 1).
A total of 10 323 patients (36.3%) started dialysis during the study period with an estimated GFR at initiation higher than 10.5 mL/min/1.73 m2. The proportion of early dialysis starts was 39.0% (95% CI, 38.1%-39.9%) in the pretrial period and was 34.0% (95% CI, 33.3%-34.7%) in the posttrial period. We observed a statistically significant increasing trend in the monthly proportion of early dialysis starts in the pretrial period (adjusted rate ratio, 1.002; 95% CI, 1.001-1.004; P = .004). A statistically significant decrease in the proportion of early starts was observed immediately after the pretrial and grace periods (rate ratio, 0.874; 95% CI, 0.818-0.933; P < .001) (Table 2). In addition, a statistically significant change in trend was observed between the pretrial and posttrial periods (rate ratio, 0.994; 95% CI, 0.992-0.996; P < .001) (Figure 1).
A total of 7166 patients (26.9%) started dialysis as acute inpatients during the study period. The proportion of acute inpatient starts was 24.0% (95% CI, 23.2%-24.8%) in the pretrial period and was 28.9% (95% CI, 28.2%-29.6%) in the posttrial period. No statistically significant trend in the proportion of acute inpatient initiations was observed in the pretrial period (adjusted rate ratio, 1.000; 95% CI, 0.999-1.002; P = .73), no statistically significant immediate consequence was observed (adjusted rate ratio, 1.092; 95% CI, 0.998-1.195; P = .06), and no statistically significant change in trend between the pretrial and posttrial periods was observed (adjusted rate ratio, 1.002; 95% CI, 1.000-1.005; P = .08) (Figure 2) (Table 2).
A total of 7066 patients (24.8%) started dialysis with a home modality during the study period. The proportion of patients initiating dialysis with a home modality was 26.8% (95% CI, 26.0%-27.6%) in the pretrial period and was 23.1% (95% CI, 22.5%-23.8%) in the posttrial period. No statistically significant trend in the proportion of dialysis initiations with a home modality was observed in the pretrial period (adjusted rate ratio, 1.001; 95% CI, 0.999-1.003; P = .32), no statistically significant immediate consequence was observed (adjusted rate ratio, 0.916; 95% CI, 0.824-1.019; P = .11), and no statistically significant change in trend between the pretrial and posttrial periods was observed (adjusted rate ratio, 0.997; 95% CI, 0.994-1.000; P = .05) (Figure 3) (Table 2).
A total of 21 054 patients (74.0%) started hemodialysis during the study period. Of these patients, 5497 (26.1%) began therapy with an AVF or AVG dialysis access. The proportion of patients initiating hemodialysis with an AVF or AVG access was 27.6% (95% CI, 26.6%-28.5%) in the pretrial period and was 25.1% (95% CI, 24.3%-25.9%) in the posttrial period. We observed a statistically significant decrease in the proportion of hemodialysis patients initiating with an AVF or AVG in the pretrial period (adjusted rate ratio, 0.997; 95% CI, 0.994-1.000; P = .02). No statistically significant immediate consequence was observed after the pretrial and grace periods (adjusted rate ratio, 1.013; 95% CI, 0.885-1.160; P = .85), and no statistically significant change in trend between the pretrial and posttrial periods was observed (adjusted rate ratio, 1.003; 95% CI, 0.999-1.007; P = .10) (eFigure 2 in the Supplement).
An overview of the segmented regression models is provided in Table 2. Parametric estimates for the segmented regression models are provided in eTable 1 in the Supplement.
Results of the sensitivity analyses for the proportion of early starts are provided in eTable 2 in the Supplement. The conclusions from all of the sensitivity analyses were similar to those from the baseline analysis. A total of 22 140 patients had no missing data and were included in the complete case scenario analysis. Results of the complete case scenario analysis are provided in eTable 3 in the Supplement. The complete case analysis also found conclusions similar to those in the baseline analysis.
In this interrupted time series study of 28 468 individuals who started dialysis in Canada from 2006 to 2015, the IDEAL trial publication was associated with an immediate and sustained change in the timing of dialysis initiation. The percentage of patients who started dialysis early in the pretrial period was 39.0%, decreasing to 34.0% in the posttrial period. We also confirmed that this change in dialysis initiation practice had no sustained association with acute inpatient dialysis initiation, home dialysis as the initial modality, or incident AVF or AVG use. Together, these findings suggest that the IDEAL trial was associated with an immediate and sustained change in the timing of dialysis initiation in Canada (excluding Quebec).
To our knowledge, this study is the first to examine long-term trends in the timing of dialysis initiation since the publication of the IDEAL trial and the corresponding clinical practice guidelines.5,13 The Arbor Research Investigators recently examined the timing of dialysis initiation in the United States Renal Data System database from 1995 to 2012 and found that the trend toward earlier initiation peaked in 2009 and leveled off in 2012.21 Although Li et al21 confirmed a steady rise in early initiation in the 1990s and 2000s, they did not observe a meaningful decrease in early initiation after 2010, as was seen in our data. We believe several possible explanations exist for these differences.
First, our data included 58 months after the publication of the IDEAL trial and may have more accurately captured the decrease in estimated GFR at initiation. Second, in 2014, the Canadian Society of Nephrology issued dedicated clinical practice guidelines on the timing of dialysis initiation. These guidelines strongly recommended adopting an intent-to-defer strategy for dialysis initiation.13 In contrast, the Kidney Disease Outcomes Quality Initiative clinical practice guidelines for hemodialysis (published in 2015) did not recommend deferral and were more conservative in the strength of their recommendations.22 Furthermore, retrospective studies of the merits of early dialysis initiation occurred early in Canada,6 and the appropriate timing of initiation was debated at annual meetings of the Canadian Society of Nephrology. These factors that captured physician interest, combined with knowledge-translation activities such as audit and feedback on data reports, may be associated with the immediate and meaningful change in dialysis initiation seen in the Canadian health care system. Third, certain health care system factors, which are not present in a universal health system such as that in Canada, may have affected the timing in the United States, in which cost and reimbursement structures for dialysis services are different. These differences were highlighted in an analysis of dialysis timing in the US Department of Veterans Affairs health system, which showed a lower estimated GFR at initiation for patients in the Veterans Affairs system than for Medicare beneficiaries, and particularly those who were older or at higher mortality risk.3
We did not notice a sustained increase in acute inpatient dialysis initiations, a decrease in home modality as the initial modality, or a decrease in permanent dialysis access insertion at dialysis initiation after the IDEAL trial dissemination. Critics of the trial and the clinical practice guidelines have suggested that early initiation may be a surrogate for cachexia or comorbidity23,24 and that delaying initiation may be associated with more hospitalizations and fewer initiations with the use of home-based therapies, although this change was not observed in the IDEAL study itself.5 A change in acute inpatient dialysis initiations was noted in the posttrial period; however, this finding was neither statistically significant nor sustained in adjusted models. Longer-term follow-up and an examination of hospitalizations prior to dialysis initiation may be needed to confirm the safety of the intent-to-defer strategy.
These findings have implications for patients, physicians, and policymakers. Patients, clinicians, and health system pressures all play a role in the dialysis initiation decision, and although patient factors are dominant, the findings suggest that a health system intervention can change the outcome. The temporal changes observed in the posttrial period suggest that clinicians who are aware and agree with the IDEAL results and corresponding guidelines can initiate change in their practice. In addition, a companion economic study of the IDEAL trial found that reducing early initiation of dialysis was associated with lower dialysis costs without any statistically significant differences in quality of life, further supporting an investment in knowledge translation in this area.25 From a research perspective, our approach provides a template for using national registry data to study the implications of future landmark trials and events for dialysis practice patterns.
The strengths of this study include broad national coverage of all dialysis starts with predialysis nephrologist care over a 10-year period. As such, the findings provide an accurate representation of the association of the IDEAL trial with nephrology practice in Canada. We were also able to link the data with national administrative databases containing hospitalizations and acute dialysis starts. As such, we could assess the association between the change in practice and unintended adverse outcomes. These linkages are not possible outside of universal health care settings.
This study has some limitations. First, we assumed that the IDEAL trial was the major catalyst for change in dialysis initiation practice, and we attributed the change to the trial results and publication. Other events during the same period, however, could be involved in the association. In fact, a guideline on the timing of dialysis and several retrospective studies that were consistent with the IDEAL trial were published in the posttrial era and could be associated with the sustained decrease seen in this study. Nonetheless, it is difficult to attribute the early and meaningful change in initiation to these other events, given that the timing of the change corresponds to the publication of the IDEAL trial.
Second, we studied only patients who had evidence of nephrologist care at least 90 days prior to dialysis initiation. We focused on these patients to exclude those who were not referred, and therefore initiated dialysis suboptimally,26 because the results of the IDEAL trial would not apply to this population. Whether there is a trend in timing for all dialysis initiations remains unknown. In addition, we studied only patients who eventually started dialysis and therefore had to survive to reach dialysis initiation. If delaying or deferring dialysis initiations was associated with increased morbidity or mortality in stage 5 chronic kidney disease (not receiving dialysis), we would not have captured these outcomes in this population. Studies of the association of the IDEAL trial with patient and health system outcomes in a stage 5 chronic kidney disease population before dialysis initiation are needed to address this evidence gap.
Third, this study occurred in a universal health care system. Thus, the applicability of the findings to privately funded health care settings, in which financial incentives could affect dialysis initiation, is unknown.
This study found that publication of the IDEAL trial appeared to be associated with an immediate and sustained change in the timing of dialysis initiation in Canada, excluding Quebec. No statistically significant sustained differences were found in acute inpatient dialysis initiations, the proportion of home-based modality as the initial modality, or the proportion of arteriovenous access construction for hemodialysis.
Accepted for Publication: January 10, 2019.
Corresponding Author: Navdeep Tangri, MD, Seven Oaks General Hospital, 2300 McPhillips St, Ste 2LB10, Winnipeg, Manitoba R2V 3M3, Canada (email@example.com).
Published Online: May 28, 2019. doi:10.1001/jamainternmed.2019.0489
Author Contributions: Dr Tangri and Mr Ferguson 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: Ferguson, Sood, Komenda, Naimark, Nesrallah, Dixon, Manns, Tangri.
Acquisition, analysis, or interpretation of data: Ferguson, Garg, Sood, Rigatto, Chau, Nesrallah, Soroka, Beaulieu, Alam, Kim, Dixon, Manns, Tangri.
Drafting of the manuscript: Ferguson, Sood, Chau, Naimark, Manns, Tangri.
Critical revision of the manuscript for important intellectual content: Ferguson, Garg, Sood, Rigatto, Komenda, Nesrallah, Soroka, Beaulieu, Alam, Kim, Dixon, Tangri.
Statistical analysis: Ferguson, Dixon, Tangri.
Obtained funding: Manns, Tangri.
Administrative, technical, or material support: Sood, Chau, Komenda, Soroka, Beaulieu, Alam, Kim, Tangri.
Supervision: Komenda, Manns, Tangri.
Other - study intervention and practice guideline: Nesrallah.
Conflict of Interest Disclosures: Dr Komenda reported honoraria from Boehringer Ingelheim Canada Ltd, Otsuka Pharmaceutical Inc, and AstraZeneca Inc as well as membership on the scientific advisory board for NxStage Medical Inc. Dr Alam reported honoraria from Otsuka Pharmaceutical Inc. Dr Kim reported support from Astellas Pharma Canada Inc. Dr Tangri reported grants from Research Manitoba and from Canadian Institute of Health Research (CIHR) during the conduct of the study, grants from Astra Zeneca Inc, personal fees from Otsuka Pharmaceutical Inc, and personal fees and other support from Tricida Inc outside of the submitted work. No other disclosures were reported.
Funding/Support: This study was supported in part by a grant from the Dr Adam Linton Chair in Kidney Health Analytics and a Clinician Investigator Award from the Canadian Institutes of Health Research (Dr Garg); funding from the Jindal Research Chair for the Prevention of Kidney Disease (Dr Sood); and funding from the CIHR New Investigator Award and a CIHR Foundation Award from the Canadian Institute of Health Research (Dr Tangri).
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.