What Can We Really Conclude About the FHN Trials and Mortality on Frequent Hemodialysis?
In 2015, investigators in the Frequent Hemodialysis Network (FHN) Trial Group reported long-term mortality patterns in each of the Daily and Nocturnal Trials. In a Journal of the American Society of Nephrology Express publication, Chertow and colleagues reported patterns during long-term follow-up of participants in the FHN Daily Trial. The trial enrolled 245 patients who were randomly assigned to receive in-center hemodialysis for either:
- 6 sessions weekly, at about 2 ½ hours per session (N = 125)
- 3 sessions weekly, at about 3 ½ hours per session (N = 120)
During a median follow-up of 3.6 years, there were 54 deaths:
- 20 with frequent hemodialysis
- 34 with usual hemodialysis
In survival analysis, the intention-to-treat mortality hazard ratio for frequent versus usual hemodialysis was 0.54. In the September issue of the American Journal of Kidney Diseases, Rocco and colleagues reported patterns during long-term follow-up of participants in the FHN Nocturnal Trial. The trial enrolled 87 patients who were randomly assigned to receive either:
- Nocturnal home hemodialysis for 6 sessions weekly (N = 45)
- Hemodialysis, typically at home, for 3 sessions weekly (N = 42)
During a median follow-up of 3.7 years, there were 19 deaths:
- 14 with frequent nocturnal hemodialysis
- 5 with usual hemodialysis
In survival analysis, the intention-to-treat mortality hazard ratio for frequent versus usual hemodialysis was 3.88. The long-term mortality effects in both trials were statistically significant.
Significant findings in important randomized clinical trials require scrutiny. In the case of the FHN Trials, we see vexing data: opposing mortality effects with two frequent hemodialysis schedules. Are the FHN Trials truly in opposition? Do the trials definitively show that mortality risk with frequent hemodialysis is different than with conventional hemodialysis? For several important reasons, the answer is no. Actually, the trials leave uncertainty about mortality effects.
The FHN Trials randomly assigned patients to either frequent or conventional hemodialysis. The randomized clinical trial is the ideal study design for identifying treatment effects, so long as the trial is large and randomization balances risk factors between treatment groups. However, interpreting mortality effects in the FHN Trials is challenging, because the trials were small, there were differences between treatment groups, and adherence to assigned treatment was low.
First, each of the FHN Trials was relatively small. Sample size is a key determinant of statistical power. Thus, small trials can have limited power to detect treatment effects on some outcomes, especially time-to-event outcomes. In the 2007 report of the design of the FHN Nocturnal Trial, Suri and colleagues wrote, “Although it might have been ideal to evaluate the effects of frequent HD on mortality and other “hard” end points, this would require over 1500 subjects per trial and several years of accrual and follow-up.” What is less appreciated about small trials is their tendency to lead us astray. Significant results in small trials are more likely to be false than significant results in large trials; this is a mathematical fact. This phenomenon can be attributed partially to the possibility of observing atypical outcomes in one treatment group.
In the FHN Nocturnal Trial, for example, the death rate was only 3.2 cases per 100 patient-years in conventional hemodialysis patients. Of the 5 deaths that were observed, 3 were due to cancer. In other words, with almost 4 years of follow-up, there were 2 non-cancer deaths in conventional hemodialysis patients. We are dealing with an atypical result, as the all-cause mortality rate is far below what has been reported in diverse dialysis patient populations, including US in-center hemodialysis and peritoneal dialysis patients, as well as short daily hemodialysis patients. Actually, the all-cause mortality rate is on par with mortality in kidney transplant patients.
Another challenge with small trials is achieving balance in baseline characteristics of treatment groups. The purpose of randomization is to balance risk factors across treatment groups. In its simplest form, randomization can be operationalized with a coin flip. Imagine that we were to flip a fair coin 10 times. Although we expect to see 5 heads and 5 tails, on average, the probability of seeing such a result in 10 flips is only 25%. Such is the effect of randomization. In the long run, we expect randomization to balance risk factors. In the short run, randomization may leave imbalances.
In the Nocturnal Trial, there were quite a few differences between treatment groups and many of those differences predisposed daily nocturnal treatment to higher risk of death:
- Mean age was 2.3 years lower in nocturnal hemodialysis patients.
- Mean hemoglobin at the start of the trial was 0.3 grams lower in nocturnal hemodialysis patients than in conventional hemodialysis patients.
- The prevalence of atrial fibrillation was 13% in nocturnal hemodialysis patients. In contrast, prevalence was exactly 0% in conventional hemodialysis patients.
- Median vintage was 1.3 years in the nocturnal hemodialysis group, but only 0.5 years in the conventional hemodialysis group. This difference manifests clinically as a difference in urine volume. Whereas 40% of nocturnal hemodialysis patients had urine volume over 500 mL per day, 55% of conventional hemodialysis patients had such volume.
- In the subset of trial subjects who were US citizens, the prevalence of Medicaid coverage was 31% in nocturnal hemodialysis patients, but only 16% in conventional hemodialysis patients.
In the Daily Trial, there were also some difference between treatment groups, despite sample size that was nearly 3 times lager than in the Nocturnal Trial:
- Mean age was 3.1 years lower in short daily hemodialysis patients. This likely predisposed short daily treatment to lower risk of death.
- Mean predialysis blood pressure was higher in short daily hemodialysis patients than in conventional hemodialysis patients. Mean systolic blood pressure was 1.3 millimeters higher in short daily hemodialysis patients.
- Whereas 72% of short daily hemodialysis patients produced less than 50 milliliters per day of urine, only 60% of conventional hemodialysis patients produced minimal urine.
Adjustment might correct imbalances, but the feasibility of adjustment is uncertain. Each adjustment factor would require 5 to 10 deaths. Therefore, limited adjustment in the Daily Trial is possible. However, the investigators presented no adjusted analysis. In the Nocturnal Trial, with only 19 deaths, adjustment was essentially infeasible.
Another challenge with interpreting mortality patterns in the FHN Trials is low adherence to assigned treatment after the end of each trial. The pre-specified duration of each trial was 12 months. After each trial ended, changes in dialysis schedule were common. Chertow and colleagues reported that at Month 14 after randomization in the Daily Trial, only 28% of patients who were assigned to frequent hemodialysis received at least 4 sessions weekly. Rocco and colleagues reported deviations in both groups of the Nocturnal Trial. Ironically, about 40% of conventional hemodialysis patients converted to nocturnal hemodialysis during the first year after the trial ended.
This is an important feature to consider as we interpret the Nocturnal Trial. During the 12 months after randomization, only 3 deaths were observed: 2 in nocturnal hemodialysis patients and 1 in conventional hemodialysis patients. The “action” occurs during the first year after the trial ended. During that year, 9 deaths were observed in patients who were originally assigned to nocturnal hemodialysis, whereas only 1 death was observed in patients who were originally assigned to conventional hemodialysis. On an intention-to-treat basis, this is the pivotal result regarding mortality. The question remains, however, whether this result shows that nocturnal hemodialysis is the causal factor that engendered 9 deaths, instead of 1 death. Rocco and colleagues reported death counts and rates during the first year after the trial ended, as a function of both randomly assigned treatment and actually delivered treatment.
Let's start with those patients who were originally assigned to conventional hemodialysis and divide them into patients who actually received conventional hemodialysis during the first year after the trial ended and other patients who actually received nocturnal hemodialysis during the same interval. In patients who actually received conventional hemodialysis, there was 1 death and the death rate was 3.7 cases per 100 patient-years. In patients who actually received nocturnal hemodialysis, there were no deaths. Thus, this comparison favors nocturnal hemodialysis, although an absolute difference of 1 death is hardly meaningful.
More interesting is the observed mortality in patients who were originally assigned to nocturnal hemodialysis. In patients who continued to undergo nocturnal hemodialysis during the first year after the trial ended, there were 4 deaths and the death rate was 21.5 deaths per 100 patient-years. In contrast, in patients who reverted to conventional hemodialysis, there were 5 deaths and the death rate was 45.7 deaths per 100 patient-years. This is a large difference that favors nocturnal hemodialysis. Admittedly, this analysis is not definitive. However, it suggests that intention-to-treat and as-treated analyses of the FHN Nocturnal Trial lead to very different conclusions about mortality. Until we know the characteristics of patients that switched hemodialysis treatment regimens during extended follow-up and apply appropriate methodology to account for such characteristics, these patterns and the utility of intention-to-treat analysis will remain enigmatic.
In light of these issues, Rocco and colleagues issue a note of caution: “In summary, our observation of higher mortality among individuals randomly assigned to the nocturnal hemodialysis group compared to the conventional hemodialysis group raises concerns regarding the long-term effects of frequent nocturnal hemodialysis, but should be interpreted cautiously due to a surprisingly low (0.03 per patient-year) mortality for individuals randomly assigned to the conventional group, low statistical power for the mortality comparison, and the high rate of hemodialysis prescription changes.”
Although the FHN Trials may not be a definitive source of information about the mortality effects of frequent hemodialysis, the trials should not be ignored. What can FHN teach us? In fact, FHN is a high-quality source of information about 12-month effects of frequent hemodialysis on outcomes that were more adequately powered than mortality. From that perspective, the Daily and Nocturnal Trials offer consistent evidence about the effects of frequent hemodialysis.
Across the two trials, the 12-month effects of frequent hemodialysis on left ventricular mass similarly favored the frequent schedule. The effects on the Beck Depression Inventory score were also similar. The effects on serum phosphorus were qualitatively similar, although the magnitude of the effect of nocturnal hemodialysis was relatively larger, likely owing to the time-dependent kinetics of phosphate clearance. The effects on predialysis systolic blood pressure were similar, as were the effects on the number of antihypertensive agents.
Thus, the physiologic effects of frequent hemodialysis appeared to be consistent with short daily and nocturnal schedules. This is a key finding of FHN. Other details of the FHN Trials should also be heeded. For example, in patients with marked residual renal function, the prescription of more frequent hemodialysis should be carefully considered. The results of the FHN Nocturnal Trial suggest that the effects of longer session duration in the home setting merit additional study; mean session duration with conventional hemodialysis was 4 hours and 16 minutes! Ultimately, the FHN Trials provide valuable data, especially during the 12 months of each trial.