#NephJC Chat
Tuesday, August 5th, 2025, 9 pm Eastern on Bluesky
J Am Soc Nephrol. 2025 Jul 10. doi: 10.1681/ASN.0000000765. Online ahead of print.
Higher versus Lower Phosphate Targets for Patients Undergoing In-Center Hemodialysis: A Randomized Controlled Trial
Daniel Edmonston, Tamara Isakova, Laura M Dember, Sophia Waymyers, Davy Andersen, Kevin E Chan, Hrishikesh Chakraborty, Myles Wolf
PMID: 40638247
Introduction
Expert opinion, rather than rigorous experimentation, has dominated guideline recommendations on phosphorus control in ESRD patients for decades. Secondary and tertiary hyperparathyroidism, marked by alterations in PTH, calcium, and phosphorus, are some of the most common metabolic alterations encountered in patients on dialysis. Multiple large observational studies, Dialysis Outcomes and Practice Patterns Study (DOPPS; Marcelo Barreto Lopes et al, Nephrol Dial Transplant, 2020), the COSMOS study (Cannata-Andía et al, Nefrologia, 2008), and the post hoc analysis of HEMO study (Ron Wald et al. Am J Kidney Dis. 2008) have reported that higher phosphate levels are linked with worse outcomes. Elevated serum phosphate, at least in these observational studies, has been strongly associated with vascular calcification and increased risk of cardiovascular events and mortality. But these observational studies are quite confounded - since the same factors associated with a ‘better’ controlled phosphate (e.g. dietary adherence to salt, volume, and phosphate, adherence to dialysis time/schedule, adherence to all pills, socioeconomic/educational status) are also associated with better clinical outcomes. The kind of trials that have been done are of one phosphate binding agent versus another, showing some are better or worse at lowering phosphate, without asking, should we even lower phosphate at all? A meticulous meta-analysis of these trials reports that ‘there was no evidence that any drug class lowered mortality or cardiovascular events when compared to placebo’ (Palmer et al AJKD 2016). Despite this weak evidence base, the KDIGO guidelines recommend lowering phosphate levels toward the normal range as an essential part of managing patients on dialysis. To maintain phosphate levels within the normal range, patients often require strict dietary modifications, adequate hemodialysis, and the use of phosphate binders. These strategies to lower phosphate can have adverse consequences, including increased calcium load and vascular calcification, gastrointestinal irritation, and a significant pill burden. They also add a huge cost burden estimated at a staggering $1.5 billion (2015) in the US alone (St Peter et al, AJKD 2018). The dietary restrictions and pill burden likely also reduce patients’ quality of life.
Table from Hiremath, AJKD blog 2021
Thus, despite being a big part of managing hemodialysis, no large RCT has ever shown that pushing phosphate lower with any intervention actually improves survival…mostly because no large RCTs have ever been completed.
While pilot trials like TARGET (Wald et al, CJASN 2017) and SPIRiT (UK) (Bhargava et al, BMC Nephrology 2019) demonstrated that achieving durable phosphate separation is feasible in hemodialysis patients, the HiLo trial (US) was designed to directly test whether liberal versus intensive phosphate targets translate into meaningful differences in mortality and hospitalization. The other ongoing trial is PHOSPHATE (international multi-center Pragmatic Randomised Trial of High Or Standard PHosphAte Targets in End-stage Kidney Disease; NephTrials summary), which will take a couple of years more, so we will discuss what HiLo recently reported.
Fig 1. Factors in favor of and against more aggressive reduction of serum phosphate levels, from Edmonston DL, et al, AJKD, 2021
The Study
Methods
HiLo was a randomized, pragmatic, multicentre, open-label trial designed to compare clinical outcomes between two phosphate targets in patients receiving maintenance hemodialysis. Dialysis facilities were stratified by size (above or below the median facility census) and then randomized at the facility level to either a high phosphate target or low phosphate target. All patients within a given facility who consented to the trial were randomized to the same phosphate target.
Inclusion criteria
Adults 18 years of age and older receiving maintenance in-centre hemodialysis three times per week
Exclusion criteria
In-centre nocturnal hemodialysis
History of calciphylaxis
Active pregnancy
Randomization: Cluster to Individual
Initial plan: The HiLo trial started with cluster randomization, which included entire clinics assigned to either a high or low phosphate target to keep treatment consistent and avoid overlap within units. The sample size was determined using a simulation with DaVita and TiME trial (Dember LM, et al, JASN, 2019).
Several assumptions were made for these simulations
Mortality: 15% (Hi group) vs. 12.8% (Lo group) → 16% relative reduction.
Hospitalizations: 2/year (Hi) vs. 1.89/year (Lo); 35% reflecting a zero-inflated distribution.
Intraclass correlation: 0.001 (mortality), 0.003 (hospitalizations).
General assumptions: 2–4 year follow-up, with a 5% annual loss, and proportional hazards for mortality.
Enrollment: started March 2020, paused for COVID-19 (April–November 2020).
After enrolling 544 patients (12% of 4400 target) issues emerged:
Baseline phosphate imbalance (0.9 mg/dL higher in Hi group compared to the Lo group)
Unequal consent rates (38% Hi group vs. 63% Lo group)
Risk of bias from post-randomization consent
The Data and Safety Monitoring Board (DSMB) advised switching from cluster to individual randomization.
New simulation: 4000 individually randomized patients → >95% power.
Randomization: 1:1, site-level stratification, random blocks.
Re-enrollment resumed: February 23, 2022.
Only 249 additional patients were recruited.
November 2023: DSMB recommended early termination due to low enrollment and insufficient phosphate separation between arms.
Fig 3. Randomization in HiLo trial, from Edmonston DL, et al. AJKD, 2021
Intervention:
1. Hi group: Target serum phosphate ≥ 6.5 mg/dl (2.1mmol/L)
2. Lo group: Target serum phosphate < 5.5 mg/dl (1.8 mmol/L) (higher than KDIGO guideline target to reflect real-world practice)
Goal: Maintain at least a 1.0 mg/dl separation between the groups
Data Collection
All trial data were collected from the dialysis electronic health record (EHR) through monthly data transfers from DaVita to the Duke Clinical Research Institute.
Data elements included:
Demographic details and comorbid conditions
Medications and vascular access type
Date of dialysis initiation
Dialysis treatment parameters
Routine laboratory data included:
Monthly: Serum phosphate, calcium, hemoglobin, and albumin
Quarterly: Parathyroid hormone levels
Dates of hospitalizations
Dates of death, transplantation, or transfer to another dialysis facility
Outcomes
The primary outcome was designed to assess a hierarchical composite outcome of all-cause mortality, followed by all-cause hospitalization.
Secondary outcomes included all-cause mortality alone and all-cause hospitalization alone.
Statistical Analysis
To handle the switch from cluster to individual randomization, a partial clustering approach was used. For analysis of this combined cohort, two new “clusters” were created for the individually randomized Hi and Lo groups.
These were analyzed together with the original cluster-randomized patients.
Primary hierarchical composite outcome was evaluated using the Finkelstein and Schoenfeld method to calculate a win ratio.
Each Hi group patient was compared to each Lo group patient hierarchically for time to mortality and frequency of hospitalization.
Win ratio calculation: The win ratio is the total wins divided by total losses for the Hi group. A win ratio >1 suggests better outcomes for the Hi group. A 95% confidence interval was calculated with adjustments for clustering.
Secondary outcomes
All-Cause Mortality was analyzed using Cox proportional hazards models, adjusted for treatment group, age, and baseline phosphate, with a sandwich estimator to account for clustering.
All-cause Hospitalization rate was modeled using a negative binomial distribution, adjusting for the same factors and clustering.
Study Oversight
The HiLo Steering Committee (academic investigators, DaVita Clinical Research, and NIDDK) designed the protocol and oversaw the trial. The Duke Clinical Research Institute was the data coordinating center. The NIDDK-appointed DSMB monitored enrollment, labs, safety, and outcomes. The HiLo trial was funded by the National Institutes of Health’s National Institute of Diabetes and Digestive and Kidney Diseases (NIH/NIDDK) with no industry sponsorship. One author disclosed honoraria from Akebia and Ardelyx; no other authors reported conflicts of interest.
Results
Study Population and Phosphate Separation
Between March 2020 and November 2023, 793 participants were enrolled, with 352 randomized to the Hi group and 441 to the Lo group. Randomization occurred in two phases: the first 544 patients were cluster randomized by the dialysis facility (228 Hi, 316 Lo), followed by 249 patients who were individually randomized (124 Hi, 125 Lo) ( figure 1). Interestingly, patients were individually consented after cluster assignment, and not all eligible patients within each cluster were enrolled. Attrition was mostly due to death, transfer, or transplant, with minimal withdrawal due to phosphate targets.
Figure 1. Flow diagram for the cluster, individual, and combined study cohorts, from Edmonston et al, 2025
Participants had a mean age of 60 years, 41% were women, and 41% were Black. Baseline phosphate was higher in the Hi group (6.5 vs 5.8 mg/dL), with the imbalance more pronounced in the cluster-randomized cohort. Over the trial, the time-averaged phosphate difference between groups was 0.8 mg/dL (0.3 mmol/L). However, phosphate separation was less marked when analyzed as change from baseline.
Table 1. Baseline characteristics of the combined study cohort, from Edmonston et al, 2025
Primary Outcome
Over a median follow-up of 1.4 years, there were 57 deaths (16%) in the Hi group and 96 deaths (21%) in the Lo group. Total hospitalizations were similar: 720 in the Hi group vs 685 in the Lo group.
The hierarchical composite outcome of all-cause death and hospitalization yielded a win ratio of 0.97 (95% CI 0.55–1.71; p = 0.91), indicating no significant difference.
Among the win ratio comparisons:
21% were driven by mortality
40% by hospitalization frequency
39% resulted in a tie
Figure 3. Primary Hierarchical Composite Outcome as Assessed by the Win Ratio: As the primary outcome of the trial, this figure directly presents the win ratio, its confidence interval, and the proportions of "wins," "losses," and "ties" attributed to mortality and hospitalization, from Edmonston et al 2025
Secondary Outcomes
All-Cause Mortality
Annualized mortality was 11 deaths/100 person-years in Hi vs 13/100 person-years in Lo. Adjusted hazard ratio for all-cause mortality was 0.76 (95% CI 0.48–1.20), suggesting a non-significant trend favoring the Hi group.
In the individually randomized cohort, the trend reversed: HR 2.70 (95% CI 0.52–13.94), though with wide confidence intervals. Mortality was more than 10 times higher in the cluster-randomized cohort than in the individually randomized cohort (supplemental table 3 and 4)
Supplementary table 3. Outcomes in cluster randomized patients, from Edmonston et al, 2025
Supplementary table 4. Outcomes in individually randomized participants, from Edmonston et al 2025
All-Cause Hospitalization
The Hi group had a higher hospitalization rate:
134 vs 96 hospitalizations per 100 person-years in the Lo group
Mean hospitalizations: 1.55 vs 1.09 per patient-year in the Lo group
This corresponded to an incidence rate ratio of 1.36 (95% CI 1.11–1.67), suggesting more frequent hospitalizations in the Hi group.
Findings were consistent across both cluster and individual randomization strata.
Adverse events
Table 3. Adverse events, from Edmonston et al, 2025
Lab-based adverse events occurred in over 80% of participants in both arms. Hypophosphatemia was more common with the low target (5% vs 2%), while hyperphosphatemia predominated in the high target group (72% vs 58%). Hypercalcemia, iron overload, and secondary hyperparathyroidism were similar across groups.
Discussion
Observational literature has long painted phosphate as a major contributor to morbidity and mortality in dialysis patients. The study by Block et al. (JASN 2004) study linked higher phosphate levels to increased mortality, and a subsequent systematic review of observational studies (Palmer et al. JAMA, 2011) quantified this risk — an 18% increase in death per 1 mg/dL rise in serum phosphate. “Phosphate is a poison to humanity” became a common refrain, echoed in figure captions and conference slides proclaiming phosphate the root of all evil. These mere associations remained untested in RCTs—until the HiLo trial valiantly, but futilely, tried to address this gap.
Lowering serum phosphate in maintenance hemodialysis patients did not improve clinical outcomes in this underpowered randomized trial. The primary composite outcome of all-cause mortality and hospitalization showed no significant difference between high- and low-phosphate target groups, despite a time-averaged phosphate separation of 0.8 mg/dL. There were some trends, but it is hard to read the tea leaves from such an underpowered trial. These findings suggest that simply targeting lower phosphate may not translate into a huge benefit as the ‘phosphate is poison’ proponents propose.
Interestingly, real-world phosphate levels have drifted away from KDIGO’s recommended <5.5 mg/dL target, as shown in a DOPPS special report (Guedes et al.). From 2002–2012, phosphate levels declined across regions, but by 2021, over half of U.S. patients had phosphate >5.5 mg/dL despite ongoing phosphate binder use. This trend underscores growing clinical uncertainty and variability — reinforcing the need for outcome-driven trials like HiLo and PHOSPHATE to enhance guidance based upon better data. Unfortunately, due to the HiLo trial falling significantly short on recruitment, it was significantly underpowered to make any conclusions, apart from the dogged resistance of the US dialysis community to accept that enrolling patients in such a trial is a worthy endeavour. Deja vu, the TIME trial (Dember et al, JASN 2019)? Sadly, well meaning but misguided guidelines and quality measures based on confounded observational data seem to be a hindrance rather than a help in performing practice affirming/changing clinical trials.
Limitations
Underpowered and prematurely terminated:
The trial was stopped early with only 793 of the planned 4400 patients enrolled, limiting statistical power to detect meaningful differences.Inadequate phosphate separation:
Despite targeting a ≥1.0 mg/dL difference, the actual separation achieved was only ~0.8 mg/dL, weakening the biological contrast between arms.Bias from cluster randomization and post-randomization consent:
Post-randomization consent led to imbalances in baseline phosphate and enrollment rates, compromising internal validity. This was later corrected by switching to individual randomization. ‘Too little too late. ’ This has been acknowledged by the authors as well.Missing key data and lack of adjudication:
The study did not collect phosphate binder dosing, dietary intake, or adherence data. Hospitalizations and deaths were unadjudicated, with no cause-specific outcomes, which is key to understanding a study, especially during the pandemic.Limited generalizability and disrupted implementation:
Conducted in a single dialysis organization and during the COVID-19 pandemic, the trial had to deal with operational challenges and excluded home dialysis patients — limiting applicability and intervention fidelity.
Conclusion
Despite a thoughtful, pragmatic design, real-world recruitment challenges and insufficient phosphate separation limited HiLo’s impact on answering the questions surrounding the clinical advantages of phosphate-lowering. Until adequately powered trials like the PHOSPHATE report, the case for aggressive phosphate lowering remains firmly in the realm of expert opinion.