#NephJC Chat
Tuesday Feb 8th 9 pm Eastern
Wednesday Feb 9th 9 pm IST
BMJ 2021 Nov 29;375:e066306. doi: 10.1136/bmj-2021-066306.
Timing of dialysis initiation to reduce mortality and cardiovascular events in advanced chronic kidney disease: nationwide cohort study
Edouard L Fu 1, Marie Evans 2, Juan-Jesus Carrero 3, Hein Putter 4, Catherine M Clase 5, Fergus J Caskey 6, Maciej Szymczak 7, Claudia Torino 8, Nicholas C Chesnaye 9, Kitty J Jager 9, Christoph Wanner 10, Friedo W Dekker 11, Merel van Diepen 11
PMID: 34844936
Introduction
For a long time, the traditional measures for starting dialysis were based on symptoms. Once uremia appears, initiate dialysis. In the late 1980s and then in the 1990s, observational studies appeared which seemed to suggest that starting dialysis earlier might be beneficial (Bonomini et al, Nephron 1986; Tattersall et al, Am J Nephrol 1995) . After all, if dialysis is safe, and improves uremic milieu, why not start earlier? GFR reporting (especially with MDRD eGFR, and CKD staging) helped us realize there was a larger proportion of CKD patients with GFR below 20 than previously thought. Trends in initiation did change, and we began to start dialysis earlier. There were other observational studies however, that did suggest caution (eg Clark et al, CMAJ 2011). Patients who started dialysis seemed to have worse outcomes when compared to patients who started dialysis late. These reports were susceptible to several biases though:
Sicker patients, say those with heart failure, may start dialysis earlier due to volume concerns
Comparing patients who started early, with those who were well enough to be alive and start dialysis later, is susceptible to immortal time bias
Not accounting for the time of initiation would also lead to lead time bias.
Appendix Figures 1 - 3 from Fu et al, BMJ 2021
These aspects would apply to a meta-analysis from 2012, which included all prior observational studies, and reported slightly worse outcomes with early dialysis (HR 1.05; 95% CI, 1.02-1.08) (Susantitaphong et al, AJKD 2012). Hence these observational studies were shrugged away, and early dialysis initiation continued. Until the IDEAL trial (Initiating Dialysis Early and Late) came along.
Proportion starting dialysis at: ≥10 ml/min per 1.73 m2. (From Rosansky et al, Kidney Int 2009)
In 2010, the IDEAL trial (Cooper et al, NEJM 2010) in NEJM became one of the largest randomized controlled trials to delve into this topic. Adults with CKD and eGFR between 10-15 in Australia and New Zealand were split into two groups, one with a planned dialysis start of eGFR 10-14 (early group) and the other with a planned dialysis start of eGFR 5-7 (late group), and compared based on the primary outcome of death. The trial reported no difference between the groups, putting the metaphorical brakes on the ever earlier initiation trend in the prior decades. Guidelines quickly followed, recommending that dialysis should only be started when at lower GFRs, or when symptoms appeared, with an intent to defer/delay (eg Nesrallah et al, CMAJ 2014).
While there was no difference in death between the groups, the IDEAL trial did report difficulty in recruiting patients, with a total of only around 800 participants recruited over the course of 8 years. Additionally, there was a relatively narrow range of eGFR values that were examined and the actual mean GFRs at which dialysis was initiated was 12 (early) versus 9.8 (late) with a smaller separation of 2.2 ml/min (calculated with Cockroft-Gault). There was also a dichotomy between early and late - would there be smaller GFR intervals at which a benefit might be seen? Even now, GFR at initiation does vary across countries, from ~ 5 in Taiwan to about 11 in the US. Trials in this space are notoriously difficult, can we go back to observational studies and avoid the flaws/biases described above?
The authors of the study in question sought to perform this Herculean task. In order to combat the typical biases that arise with such an approach, the design criteria of a clinical trial was mimicked by using the “cloning, censoring and weighting” model, the target trial emulation method (Hernan et al, J Clin Epi 2016). Traditional methods of adjusting for biases use either propensity score matching, and/or regression based methods to adjust for underlying confounding. Let’s see how these new-fangled statistical approaches perform.
The study
Question: What is the optimal time at which to initiate dialysis in CKD patients?
Design
Nationwide observational cohort study
Comparing dialysis initiation strategies for a range of eGFR values, between 4 and 19 ml/min/1.73m2, in increments of 1, with the range of 6-7 taken as the reference group.
Target trial approach, summarized in the figure below.
Study population and sample size
30,180 patients in the National Swedish Renal Registry of patients referred to Nephrologists. Enrolment in the registry is mandatory in Sweden when patients reach an eGFR <30 mL/min/1.73 m2, but some clinics may start reporting them earlier. Subsequent outpatient visits to nephrology care (on average two to three a year per patient) are registered until death or emigration. Nearly all (96%) nephrology clinics in Sweden report to the Swedish Renal Registry, and the estimated national coverage is >75% for patients referred to nephrologists with chronic kidney disease G4-5.
Important consideration: They used the CKD-EPI equation and since information on ethnicity is not available in Sweden by law, all participants were assumed to be of white, European ancestry for purposes of eGFR calculation
Inclusion criteria
18 years of age or older
eGFR between 10-20
One prior measurement of eGFR between 10-30 as confirmation
No history of kidney transplant
At least one available measurement of each of the following:
Systolic BP
Diastolic BP
Calcium
Phosphate
Albumin
Hemoglobin
Outcomes
Five year all cause mortality
Five year major adverse cardiovascular events (composite of cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke)
Funding: Support from the Swedish Research Council, Center of Innovative Medicine Karolinska, and Stockholm City Council, although they 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; or the decision to submit the manuscript for publication.
Results
Out of 30,180 patients on the registry, 10,290 were included, with reasons for exclusion outlined below. These were then ‘cloned’ to provide three equal groups who were ‘assigned’ to start dialysis at different GFR cut-offs.
Table 1 (with all 3 groups showing exact same characteristics due to the underlying cloning methods) has the details of the baseline characteristics. Median age was 73, about 36% were women, and 42% had diabetes. Median eGFR was 16.8, and almost 70% of the study population had an eGFR between 15 and 20.
3,822 patients initiated dialysis during a median follow-up of 3.1 years, notably with a 60/40 split between HD and PD as the initial modality of choice. The figure below shows the actual variation of GFR at the start of dialysis in Sweden during this period.
The observed risk of mortality progressively decreased from an eGFR of 19 to 15, and then progressively increased the later dialysis was initiated, until eGFR of 4. When using the mean eGFR at start of dialysis in the US as a reference point, in other words the eGFR 11-12 group, the mortality risk varied from 2.8% higher in later initiation of dialysis to 3.1% lower in earlier initiation of dialysis.
The largest mortality benefit observed was when the reference group of eGFR 6-7 was compared to the group with eGFR 15-16.
Notably, however, the maximum difference in five-year restricted mean survival time was only 1.6 months (95% CI 1.0-2.0). In order to achieve this, patients would need to initiate dialysis 47.9 (95% CI 46.2-49.6) months, or 4 years, earlier than the reference group. This is demonstrated in the figure below.
Trade-off between additional survival time (five year restricted mean survival time difference) and time that dialysis has to be initiated earlier, for dialysis initiation strategies with estimated glomerular filtration rate (eGFR) values between 4 and 19 mL/min/1.73 m2 in increments of 1 mL/min/1.73 m2, with 6-7 mL/min/1.73 m2 as reference.
Positive value indicates longer survival and earlier dialysis start compared with reference group. In other words, about 5 months are needed for eGFR to decline 1 mL/min/1.73 m2
For MACE on the other hand, the absolute risk was lowest between eGFR 18 to 11, and then progressively increased until eGFR of 4. When using the eGFR 11-12 reference group again, risk differences varied between an increase of 4.7% for the eGFR 4-5 group and a decrease of 0.2% for the eGFR 12-13 group.
Subgroup analysis
The subgroup analysis (shown below) shows pretty consistent benefit across all subgroups though one wonders about the effect across gender and diabetes.
Sensitivity Analyses
In order to validate their statistical methods, the authors first performed a supporting analysis using the eGFR thresholds set forth in the IDEAL trial, i.e eGFR 10-14 as the early group and eGFR 5-7 as the late group. They found the early group to have a 3.3% (1.3-5.3) lower 5-year mortality risk and a 3.6% (1.0-6.0) lower 5-year risk of MACE. The achieved eGFR in each group of the IDEAL trial was 7.2 and 9.0 in early and late, respectively, and those achieved in this study were 6.0 and 8.3, respectively. The authors say these were congruent with those in the IDEAL trial. In the IDEAL trial, during a median follow-up period of 3.6 years, 152 of 404 patients in the early-start group (37.6%) and 155 of 424 in the late-start group (36.6%) died (hazard ratio with early initiation, 1.04; 95% CI, 0.83 to 1.30; P=0.75).
The authors then used traditional analytical approaches to introduce immortal time bias as in previous observational studies and demonstrated that early dialysis initiation was associated with worse outcomes, the opposite of the primary results.
Similarly, when they started follow-up at dialysis initiation, which introduced selection/survivor bias and lead time bias, the hazard ratio for early initiation was 1.58 (1.37 to 1.83) compared with late initiation, thus demonstrating the effect of incorrect analytical techniques.
Discussion
In this observational study using target trial emulation methods, the authors report that earlier dialysis start is associated with a small clinical benefit, which was robust in all the sensitivity analyses they performed. The authors seem to have been successful in mimicking a clinical trial, as the findings were somewhat consistent with those in the IDEAL trial.
While there does appear to be some benefit to earlier dialysis initiation, the question must be asked about how much is that benefit worth? For only a 1.6 month difference in mortality, patients would have to initiate dialysis 4 years earlier. This is not only associated with a heavy burden on the patient, but also on any caretakers they have, and the healthcare system as a whole.
Does this mean we should start dialysis at a GFR of 15, instead of aiming to start lower and/or when symptoms occur? Before we change current practice, let’s consider:
Are these results believable?
Are these conclusions generalizable?
What could explain these findings?
Should we change our practice as a result of this study?
Limitations
Among the limitations identified by the authors are
lack of information on the specific reason for starting dialysis, introducing an element of residual confounding. Those who started at a higher eGFR level may have done so due to severe fluid overload or uremic symptoms which in and of themselves may portend a worse prognosis, and hence affect the results. This analysis only reports initiation according to eGFR, which is not the only consideration for initiation
Additionally, the data the authors obtained did not include information on quality of life or other factors that may affect dialysis accessibility such as living in a remote area. Initiating dialysis early is a privilege that not many countries in the developing world enjoy, unfortunately, due to lack of resources or lack of access.
Creatinine based eGFR estimation is not great - especially being not as precise for the minute differences by eGFR reported in this study. (eg an eGFR of 10 might be in reality anywhere from 7 to 13).
Around 40% of the patients in Sweden in this analysis started dialysis on PD. We know otherwise that patients with higher GFR or more residual function are likely to start PD. Whether PD provides a survival advantage over HD is a more vexing question (with similar issues of lead time bias and confounding), but this is a consideration to keep in mind.
Another possible limitation is that patients start at different GFRs for a reason. This is not random variation, and no amount of statistical skulduggery can remove this selection bias aspect. Patients with worse clinical symptoms might start dialysis earlier, needless to say, which can be accounted for by adjusting for measured clinical variables. However, there can be other factors which lead to an earlier start. Patients with slower progression (e.g. ADPKD) might allow for a planned early start, and someone who is progressing fast (e.g. diabetic nephropathy) might start late by eGFR standards alone. Access to care might also impact these factors - though possibly less so in a system like Sweden.
Hence one should be very cautious in generalizing these findings outside Sweden, for reasons mentioned above, including the high PD at initiation. It would be helpful to replicate these analyses in other countries to see if these findings remain true.
Lastly, it is difficult to explain why earlier dialysis initiation would be helpful. There is badness involved with dialysis initiation: complications with access creation, faster residual renal function loss, higher societal cost, apart from the drudgery of performing dialysis. Could the considerations of urea kinetics be true after all?
Conclusions
The authors themselves do not suggest we should start dialysis early - the small increment noted in the possible quantity of life of a few months has to be balanced against the many years of dialysis needed to accrue this. The major finding of this study, is possibly the neat explanation of lead time and immortal time bias which dissolves with the target trial emulation methodology used here.
Summary by
Michael Turk
Swapnil Hiremath