Health Effects of oraL Protein Supplements in HD (The HELPS-HD Trial): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Health Effects of oraL Protein Supplements in HD (The HELPS-HD Trial) (NCT02933151), a randomized na study. The primary outcome is Mortality.

2. Clinical Question

This was a pragmatic question about two nutritional care protocols as implemented in participating hemodialysis facilities, rather than a comparison in which every individual patient was separately randomized. The intervention was a protocol for supplement use during dialysis. The contrast was therefore between facility policies: an intensive protocol for all patients and usual care with specified eligibility groups.

The registry record describes incident and prevalent patients, and it uses different mortality assessment starting points for those groups. Mortality assessment began at day 0 of study initiation for prevalent patients and day 120 for incident patients, because incident patients received supplements during their first 120 days of hemodialysis. Follow-up varied among participants.

The question is about the mortality outcome associated with assignment to these protocols in the trial setting. It does not ask whether an individual patient's nutritional status changed, and the record does not report a primary outcome estimate for hospitalization. Hospitalization is listed as a secondary outcome, but its results are not reported in the registry record.

3. Trial Design

HELPS-HD was a randomized, open-label, parallel, cluster-randomized pragmatic trial. The facility, rather than the individual patient, was assigned to follow the intensive or usual care nutritional supplement protocol. The registry record describes 105 participating facilities in the analytic dataset and an enrollment of 10457.

In the usual care protocol, facilities continued to provide oral intradialytic protein-based supplements to patients with serum albumin below 3.5 g/dL and to incident hemodialysis patients during the first months of care. Under the intensive protocol, facilities followed a policy of providing supplements to all hemodialysis patients. The supplement is described as a food containing about 15 g of protein. The record does not provide further details about the supplement product or protocol implementation.

The trial was open-label, with no masking. This is consistent with comparing facility care protocols that would be apparent to facilities and patients, although the registry record does not describe any specific procedures for limiting the influence of that knowledge on care or outcome assessment. The primary outcome was all-cause mortality. Hospitalization is listed as a secondary outcome.

The trial began on 2017-01-23, and primary completion is recorded as 2020-08-31. The intervention was discontinued on April 13, 2020 because of a Covid-related prohibition on eating in facilities. Event follow-up concluded on August 13, 2020. Follow-up time varied among participants, with the maximum described as approximately 3.5 years.

The registry record does not report the allocation sequence method, the number of facilities assigned to each protocol, or details of the cluster-level analysis beyond the reported model. The record also does not report a sample-size justification or other protocol-level statistical planning details.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Before assignment: There were 10457 patients consented in participating facilities. Because all incident patients, regardless of facility allocation, received an identical nutritional supplement protocol for the first 120 days, incident patients are excluded from analyses for the 1st 120 days. This results in an analytic dataset of 9896 individuals that serves as the basis for Table 1 and all primary analyses.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Incident and prevalent hemodialysis patients at 105 participating hemodialysis facilities in the analytic dataset

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

Population: Primary analyses incorporate a 120 day lag for incident dialysis patients as the usual care and intensive protocols were identical during the time of hemodialysis initiation, such that 9896 participants were in the analytic dataset (4763 in usual care and 5133 in intensive nutritional supplement protocols) received the intervention of a total of 10457 participants allocated to a trial intervention. There were 561 incident hemodialysis patients who did not remain in the DCI facility past day 120.

6. Primary Endpoint

The registry lists the primary outcome as Mortality, measured over Intervention discontinued on April 13, 2020 due to Covid-related prohibition on eating in facilities. Event follow-up concluded on August 13, 2020. Follow-up time varied among participants, with the maximum of approximately 3.5 years..

The primary endpoint was all-cause mortality, recorded as a number of participants. The registry description includes withdrawal from dialysis in the death classification. Mortality was assessed from day 0 of study initiation for prevalent patients and from day 120 for incident patients, because all incident patients received nutritional supplements during their first 120 days of hemodialysis.

Mortality is a direct, clinically important outcome for a trial asking whether a facility nutritional supplement protocol is associated with mortality. It does not depend on a subjective rating of symptoms or a laboratory surrogate. That does not make the comparison free of design or measurement concerns: the registry record uses different assessment starting points for prevalent and incident patients, and follow-up time varied among participants.

The reported analysis uses a Cox proportional hazards model, which expresses a comparison in terms of the rate of experiencing the event over follow-up, rather than simply comparing the raw numbers of deaths. This approach can accommodate differing follow-up times under its modeling assumptions. The registry record does not describe how the model handled clustering of patients within facilities, how censoring was defined, or whether the proportional hazards assumption was assessed.

The primary endpoint is distinct from the adverse-event listings. The registry record notes that adverse events and serious adverse events were not specifically classified, given the minimal-risk intervention and pragmatic design. The death counts listed with the adverse-event information should therefore be read as mortality counts, not as a complete adverse-event profile.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

What does the hazard ratio describe?

A hazard ratio compares the event rate over time between groups as represented by a Cox proportional hazards model. It is not a ratio of the total death counts and is not itself an absolute risk difference. The reported estimate is 1.03. The registry record does not make clear which protocol is the numerator and which is the reference in the analysis, so the estimate's direction cannot be attached to a named protocol from the result alone.

The Cox model is commonly used when participants have different lengths of follow-up and the outcome is the time until an event. It uses information about event timing and follow-up, rather than treating every participant as though they were observed for the same duration. Its interpretation depends on assumptions, including how the hazard relationship is modeled over time. Whether those assumptions were checked is not reported in the registry record.

How should the confidence interval be read?

The 95% confidence interval runs from 0.92 to 1.14. In general, a confidence interval expresses uncertainty in an estimated quantity under the statistical model and its assumptions. For a hazard ratio, 1 represents no relative difference in the modeled hazards. Because the interval includes 1, it spans estimates on either side of that reference point. The interval is not a guarantee that the true value lies within its limits, and it does not encompass uncertainties caused by possible design or data limitations.

The interval also shows why the point estimate should not be read in isolation. An estimate is a single summary of the data, while its interval indicates the range of values reasonably compatible with the data under the model. The registry record supplies no p-value, so statistical interpretation should focus on the reported estimate and interval without inventing a test result.

Why does cluster randomization matter?

In a cluster-randomized design, groups such as facilities are assigned to protocols, and patients within a facility share that assignment. Their outcomes may be more alike than outcomes from patients in different facilities because they share care settings and practices. This dependence affects the amount of information in the data and the uncertainty of treatment comparisons. A simple analysis that treats every patient as independently randomized can give misleading uncertainty estimates.

The registry record identifies the trial as cluster randomized and reports only the sponsor's own analysis. It does not describe the model's facility-level adjustment, the handling of within-facility correlation, or other details needed to evaluate the cluster analysis fully. A re-analysis using only aggregated individual-level counts would not recreate the sponsor's cluster-aware analysis.

What is the role of follow-up time?

Participants had variable follow-up, and the maximum was approximately 3.5 years. In a time-to-event analysis, participants who have not experienced the event by the end of observation contribute information up to their last observed time, subject to the model's censoring assumptions. The registry record does not report the censoring rules or detail how variable follow-up was handled. It specifies different mortality assessment starting points for incident and prevalent patients, which is important context for the risk period represented by the endpoint.

The intervention stopped on April 13, 2020 because of a Covid-related prohibition on eating in facilities, and event follow-up ended on August 13, 2020. The record does not report a separate analysis of the period before or after discontinuation, or a sensitivity analysis addressing the interruption.

What can planning calculations tell readers?

In general, sample-size calculations estimate the number of participants or events that a future study would need under stated assumptions about the effect to detect, allocation, significance level, and statistical model. They are planning illustrations, not observed power and not evidence that the completed trial was adequately or inadequately powered. The planning assumptions described for this analysis include a two-sided significance level of 0.05, equal allocation, a normal approximation, and an effect defined as a stated fraction of the observed effect. For time-to-event illustrations, the assumption specifies a total number of events; for other illustrations, it specifies participants per arm.

The registry record does not report the trial's own sample-size justification. A planning illustration based on the observed effect is not a substitute for the original design rationale, and it cannot establish what the investigators anticipated when planning the study.

What is not available for independent assessment?

Details about model covariates, facility-level variance or correlation, proportional hazards diagnostics, censoring, missing outcome data, and sensitivity analyses are not reported in the registry record. The record gives the result and its confidence interval but not enough model detail to reproduce the analysis or evaluate every assumption. These omissions limit statistical assessment; they do not, by themselves, show that a particular analysis choice was inappropriate.

10. Exact Estimates for Each Arm

No two-arm comparison is shown for this trial, but each arm's result can still be summarized with exact interval estimates computed from the posted counts. The Clopper-Pearson interval is exact and conservative; the Wilson interval is a common closer-to-nominal alternative; the Bayesian credible interval uses a uniform Beta(1, 1) prior. For small groups the three can differ noticeably.

R code and output

arm    <- c("Usual Care Protocol", "Intensive Protocol")
n      <- c(4763, 5133)
events <- c(1756, 1856)
for (i in seq_along(n)) {
  ci <- binom.test(events[i], n[i])$conf.int          # Clopper-Pearson exact interval
  cr <- qbeta(c(0.025, 0.975), 1 + events[i], 1 + n[i] - events[i])   # Bayesian, uniform Beta(1,1) prior
  wi <- (function(x, n) { p <- x / n; z <- qnorm(0.975); d <- 1 + z^2 / n
    c((p + z^2 / (2 * n) - z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d, (p + z^2 / (2 * n) + z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d) })(events[i], n[i])
  cat(sprintf("ARM|%s|%d|%d|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f\n", arm[i], n[i], events[i], 100 * events[i] / n[i],
              100 * ci[1], 100 * ci[2], 100 * wi[1], 100 * wi[2], 100 * cr[1], 100 * cr[2]))
}

ARM  Usual Care Protocol  4763  1756  36.9  35.5  38.3  35.5  38.2  35.5  38.2
ARM  Intensive Protocol  5133  1856  36.2  34.8  37.5  34.9  37.5  34.9  37.5

11. Analyses Reported by the Sponsor

The registry record reports the sponsor's Cox proportional hazards analysis: an estimate of 1.03 with a 95% confidence interval from 0.92 to 1.14. A p-value is not reported. This is the analysis to use when describing the trial's posted mortality result.

No simple unadjusted re-calculation is presented alongside that estimate. The trial randomized facilities, not individual patients, so a re-analysis that treated each patient's outcome as an independent randomized observation could misrepresent the design and produce misleading uncertainty. Aggregate death counts alone are not enough to reproduce a cluster-aware time-to-event analysis, particularly with variable follow-up.

The registry record does not report enough detail to independently reproduce the sponsor's model, including the mapping of analysis identifiers to protocol names and the handling of facility-level clustering. The reported hazard ratio therefore should be presented as the sponsor's estimate without assigning its direction to a named arm beyond what the record states. The confidence interval provides the reported measure of statistical precision; the registry record does not provide a separate p-value or a competing recalculated result.

12. Multiplicity, Missing Data and Analysis Populations

The registry record lists all-cause mortality as the primary outcome and hospitalization as a secondary outcome. One analysis is reported. How the investigators controlled the chance of false-positive findings across outcomes, or whether they prespecified a hierarchy for interpreting them, is not reported in the registry record. The hospitalization result is also not reported in the registry record.

Multiplicity matters when a study evaluates multiple outcomes, subgroups, time points, or models. In general, considering many possible comparisons can increase the chance that at least one appears unusual by chance. Whether that issue applies to the trial's complete analysis plan cannot be determined from the registered information. No additional outcome analyses or subgroup results should be inferred from the listed primary result.

Missing-data methods are not reported in the registry record. The record does not describe whether mortality status was missing for any participants, how such cases were handled, or whether imputation or sensitivity analyses were used. It also does not report censoring rules beyond stating that follow-up time varied. These details matter because time-to-event estimates rely on assumptions about the observation process and censoring.

The analytic population is described as incident and prevalent hemodialysis patients at 105 participating facilities. The registry record does not explain how that analytic population relates to overall enrollment of 10457, nor does it report participant flow, exclusions, or the number with complete outcome information. The group counts listed with the adverse-event information are not a substitute for a full analysis-population accounting.

Analysis population definitions, missing-data handling, and multiplicity plans are important components of a statistical analysis plan. Their absence from the registry record limits the ability to assess how all planned comparisons were managed, but it should not be taken as proof that no such procedures existed.

13. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 5% reporting threshold.

The adverse-event information lists zero serious events and zero other events affected in each protocol group. It also lists 1732 deaths in usual care and 1827 deaths in the intensive protocol. These death counts are presented separately from the serious and other event counts and should be interpreted as mortality information, not as a complete summary of adverse events.

The registry record explicitly says that adverse events and serious adverse events were not specifically classified. It gives the rationale as the minimal-risk intervention, described as a food containing about 15 g of protein in some participants in usual care facilities and all participants in intensive protocol facilities, together with the pragmatic trial design. No serious-event terms are listed.

Consequently, the zero affected counts do not establish that no adverse events occurred. They indicate only that none were entered in those specific event-count fields in the registry record, alongside the statement that such events were not specifically classified. The record does not report the procedures for identifying or collecting adverse events, so these fields cannot support a detailed safety comparison.

The mortality comparison is reported through the Cox model, while the listed death counts are raw counts without adjustment for follow-up time or clustering. They should not be treated as adverse-event rates or used to replace the primary time-to-event analysis.

14. What the Result Does, and Does Not, Mean

The posted result describes a facility-level comparison of two nutritional supplement protocols among hemodialysis patients. The reported Cox proportional hazards estimate is 1.03, with a 95% confidence interval from 0.92 to 1.14. Because the interval includes 1, the reported estimate is compatible with no relative difference as well as values on either side of that reference. The registry record does not report a p-value, and the result should not be described using an unreported significance test.

The analysis identifier-to-protocol mapping is not stated in the reported result, so the direction of the hazard ratio cannot be assigned to the intensive or usual care protocol from this record alone. The raw death counts do not resolve that limitation: they do not account for time at risk or the cluster-randomized design. The sponsor's model is therefore the relevant reported comparison, while its full specifications are not available in the registry record.

The result does not establish that the protocols have identical effects. A confidence interval that includes the no-difference value is not proof of equivalence, and the record does not describe a non-inferiority or equivalence question. Nor does the result establish a particular effect for individual patients, facilities outside the analytic dataset, or outcomes other than mortality.

The record lists hospitalization as a secondary outcome, but its results are not reported. It also does not report details of missing-data handling, multiplicity procedures, or sensitivity analyses. Those gaps restrict interpretation beyond the single posted mortality estimate.

This is a statistical description of the registered result, not medical advice. It identifies what the registry record reports and the limits of drawing conclusions from that estimate alone. The trial's pragmatic setting and facility-level assignment are relevant to understanding the comparison, but the record does not supply enough detail to assess all assumptions or to generalize the result beyond its stated population and care context.

15. Why This Trial Matters Statistically

The trial addresses a practical question about how nutritional supplement protocols are organized in hemodialysis facilities. Its contrast is between a policy applying supplements to all patients and usual care that targets specified patient groups. This makes the facility-level design important: the assignment concerns a care protocol, not an individually assigned supplement in isolation.

Statistically, the design illustrates why the unit of randomization matters. Patients within a facility share the assigned protocol, so an analysis must respect the clustering rather than assume independent individual assignment. The registry record identifies the design as cluster randomized and reports a Cox proportional hazards estimate, while leaving some model details unreported. Recognizing the design prevents raw patient counts from being mistaken for a complete treatment comparison.

The outcome also highlights the role of time-to-event methods when follow-up differs among participants. The record reports variable follow-up, a maximum of approximately 3.5 years, and a mortality assessment period with different starting points for incident and prevalent patients. These features make the hazard-based analysis more relevant than a comparison of raw counts alone, while also making model and censoring details important for assessment.

The reported estimate and interval communicate both a point estimate and uncertainty. The interval includes 1, and the record does not report a p-value or map the analysis identifiers to the named protocols. Those limits matter as much as the point estimate when describing the evidence. The registry record also shows why a trial's statistical interpretation depends on information beyond a single result: analysis populations, missing-data procedures, multiplicity plans, and handling of cluster structure affect how fully a reader can evaluate the finding.

16. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

Several trial-specific details needed to assess the mortality analysis are not reported in the registry record. The result does not map the analysis identifiers to the intensive and usual care protocols, so the hazard ratio's direction cannot be assigned to a named protocol. The record also does not give a p-value, model covariates, facility-level correlation handling, or diagnostics for the proportional hazards assumption.

Follow-up varied among participants, and mortality assessment began at different points for prevalent and incident patients. Although the endpoint description explains those starting points, censoring rules and the treatment of incomplete follow-up are not reported. The intervention was discontinued on April 13, 2020 because of a Covid-related prohibition on eating in facilities, while event follow-up concluded on August 13, 2020. The registry record does not state how the intervention interruption was handled in the analysis or whether sensitivity analyses assessed its implications.

The registry record describes an analytic population at 105 facilities and overall enrollment of 10457, but does not provide participant flow or explain how the analysis population relates to enrollment. Missing-data methods, multiplicity procedures, and results for the listed secondary outcome of hospitalization are not reported in the registry record. The record also does not provide the trial's sample-size justification.

Finally, the intervention was open-label and assigned by facility. The record does not describe masking or procedures to limit potential effects of knowledge of assignment on care or outcome assessment; masking is recorded as none. These are limits on what can be evaluated from the registry record, not evidence by themselves that the reported estimate is biased.

17. Related Tutorials

18. Related Statistical Calculators

19. Sources

• ClinicalTrials.gov: Health Effects of oraL Protein Supplements in HD (The HELPS-HD Trial) — NCT02933151.
• Registry record last updated 2024-11-05; status last verified 2024-11.
• Results first posted 2024-11-05.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

20. Record Summary

HELPS-HD was an open-label, parallel, cluster-randomized pragmatic trial comparing intensive and usual care nutritional supplement protocols at hemodialysis facilities. The analytic population is described as incident and prevalent hemodialysis patients at 105 participating facilities. Overall enrollment is recorded as 10457.

The primary outcome was all-cause mortality, with withdrawal from dialysis classified as death. The registry record reports a Cox proportional hazards estimate of 1.03 and a 95% confidence interval from 0.92 to 1.14. It does not report a p-value or identify which named protocol is the numerator in the reported comparison. Hospitalization is listed as a secondary outcome, but its result is not reported in the registry record.

Follow-up varied among participants. The intervention was discontinued on April 13, 2020, and event follow-up concluded on August 13, 2020. The registry record does not report several details needed to reproduce or fully assess the analysis, including missing-data handling, multiplicity procedures, and the model's handling of clustering.