EMPA-KIDNEY (The Study of Heart and Kidney Protection With Empagliflozin): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for EMPA-KIDNEY (The Study of Heart and Kidney Protection With Empagliflozin) (NCT03594110), a randomized phase 3 study. The primary outcome is Interventional Part: Time to First Occurrence of Kidney Disease Progression or Cardiovascular Death ('as Adjudicated').

2. Clinical Question

3. Trial Design

EMPA-KIDNEY was a phase 3, randomized, double-blind, parallel-group, placebo-controlled clinical trial sponsored by Boehringer Ingelheim. Its official title describes a multicentre international study. The registry record reports an enrollment of 6609 and lists chronic kidney disease as the condition under study.

Participants were assigned to empagliflozin 10 mg once daily or matching placebo once daily. The population included patients with chronic kidney disease at risk of progression, with or without diagnosed diabetes mellitus. Both groups received treatment on top of standard of care. The record identifies the randomised set as the primary endpoint population, including all randomised participants whether treated or not.

The recorded study start was 2019-01-31, and the primary completion date was 2022-07-05. The record status is completed. The primary endpoint timeframe ran from randomisation to the final follow-up visit in the interventional part, up to 1136 days. The record also describes post-trial observational follow-up of a subset for cardio-renal outcomes after the interventional part; details of that follow-up are not reported in the registry record.

The registry lists secondary outcomes covering heart failure hospitalization, cardiovascular death, all-cause hospitalization and death, and kidney outcomes. Some were specified for the interventional part and others for the overall study. The record does not describe the sequence or hierarchy used to test those secondary outcomes. Nor does it report further details about how randomisation was generated or concealed. The stated double masking and placebo comparison identify the broad design, but the registry record does not specify which roles or parties were masked.

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: All subjects were screened for eligibility prior to participation in the trial. Subjects attended a specialist site which ensured that they (the subjects) strictly met all inclusion and none of the exclusion criteria. Subjects were not to be allocated to a treatment group if any of the entry criteria were violated.

Interventional Part

Reasons for not completing this period:

Non-interventional Part

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Randomised set (RS) included all randomised participants, whether treated or not.

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: Randomised set (RS) included all randomised participants, whether treated or not.

6. Primary Endpoint

The registry lists the primary outcome as Interventional Part: Time to First Occurrence of Kidney Disease Progression or Cardiovascular Death ('as Adjudicated'), measured over From the day of randomisation to the day of the final follow-up visit in the interventional part of the trial, up to 1136 days..

The primary endpoint was time to first occurrence of kidney disease progression or adjudicated cardiovascular death during the interventional part. Kidney disease progression could be defined by end stage kidney disease, a sustained decline in estimated glomerular filtration rate to <10 mL/min/1.73m^2, renal death, or a sustained decline of ≥40% in estimated glomerular filtration rate from randomisation. End stage kidney disease was defined as initiation of maintenance dialysis or receipt of a kidney transplant.

Combining kidney progression with cardiovascular death captures events from two clinically important outcome domains in one endpoint. A time-to-first-event definition focuses on the time until a participant experiences the first qualifying event; it does not count later events for that primary outcome. The registry reports the endpoint as an incidence rate: the number of participants with a first event is related to patient-years at risk, with the rate expressed per 100 patient-years at risk.

Rates account for differing observed time at risk across participants, whereas the sponsor’s Cox analysis reports a hazard ratio comparing the event experience over time between groups. Those summaries answer related but distinct questions. The registry record does not report separate results for the kidney and cardiovascular components within the primary endpoint, so the combined result does not establish which component contributed more to the overall finding.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

How the primary event rate is defined. The registry defines the primary result as the number of participants experiencing a first event of kidney disease progression or adjudicated cardiovascular death, scaled by patient-years at risk and expressed per 100 patient-years at risk. Patient-years at risk combine each participant’s time at risk within a treatment group. This type of rate can account for variation in follow-up time: participants need not all contribute the same duration of observation. It is not a simple proportion of participants with events, and it should not be interpreted as a fixed-time probability.

The rate estimate is accompanied by a 95% confidence interval. In general, a confidence interval conveys the precision of an estimate under the statistical procedure used to construct it. A narrower interval indicates greater precision than a wider interval on the same scale, though the interval alone does not resolve questions about bias, endpoint definition, or clinical relevance. The registry record does not state the method used to calculate the confidence intervals for the rates.

How to read the Cox comparison. The sponsor reports a Cox regression analysis, with the comparison expressed as a hazard ratio. In general, a Cox model compares the event hazard between groups over time and can incorporate information about when events occur, rather than reducing follow-up to a single event proportion. A hazard ratio below 1 indicates a lower estimated hazard in the numerator group relative to the comparison group, according to the model’s coding. It is not a direct estimate of an absolute difference in event probability. The registry record does not report model covariates, checks of the proportional-hazards assumption, or other fitting details.

The reported confidence interval for the hazard ratio is 99.83%, rather than the 95% confidence intervals shown for the group rates. These intervals have different stated coverage and refer to different statistics. The comparison also has a reported p-value of < 0.001. In general, a p-value measures how incompatible the observed data are with a specified null hypothesis under the assumptions of the analysis; it is not the probability that the null hypothesis is true, nor a measure of treatment effect size.

Why event counts can be over-dispersed. General explanation: for a count outcome, a simple Poisson model assumes that the variance of the count is tied to its mean. Real clinical event counts may vary more than that assumption allows, for example when individuals differ in underlying risk or exposure time. This extra variability is called over-dispersion. A model that fails to account for it can produce confidence intervals that are too narrow. The primary endpoint is a time-to-first-event rate and the sponsor reports Cox regression; the registry record does not say that an over-dispersed count model was used. The general issue is therefore a statistical consideration, not a claim about the sponsor’s analysis.

What the secondary re-calculations mean. Recalculated comparisons based on posted summary statistics are simple, unadjusted analyses. They do not reproduce the sponsor’s fitted model and may not account for its covariates, repeated measurements, missing-data handling, or other design features. For a continuous outcome, a mean difference expresses the contrast between group means on the original outcome scale. Its confidence interval indicates the range of values compatible with that unadjusted calculation under its assumptions. If an interval includes zero, the data do not distinguish the estimated contrast from zero at that interval’s stated level, but that alone does not prove the groups are identical.

General explanation: repeated measurements on the same participant are correlated, so a model for repeated measures needs to represent that dependence. A mixed model can use measurements across time while accounting for within-participant correlation, subject to its specified structure and assumptions. The registry lists mixed models for one secondary analysis, but does not report the full model specification. The accompanying unadjusted re-calculation should therefore be viewed as a separate descriptive check, not a replacement for the sponsor’s model-based result.

Planning illustrations are not observed power. The planning calculations use a two-sided significance level of 0.05, equal allocation between arms, and a normal approximation. They illustrate sample sizes for detecting stated fractions of an observed effect; time-to-event illustrations use a total number of events, while other illustrations use participants per arm. These are planning illustrations derived from observed effects, not the trial’s original design justification and not observed power. The registry record does not report the trial’s sample-size justification or the assumptions used by the investigators when planning enrollment.

In general, sample-size calculations depend on assumptions about the effect to detect, outcome variability or event frequency, allocation, and statistical testing. Changing those assumptions changes the planning result. The illustrations do not establish that the trial was adequately or inadequately powered for every listed outcome. They also should not be read as a probability that a particular result would be statistically significant after the trial was completed.

10. Analyses Reported by the Sponsor

For the primary endpoint, the registry reports the sponsor’s Cox regression comparison as a hazard ratio of 0.72, with a 99.83% confidence interval from 0.59 to 0.89 and p < 0.001. No separate recalculated primary hazard ratio is reported. The posted group incidence rates are descriptive rate estimates, not a re-calculation of the Cox model comparison. Because these statistics use different methods and summarize different features of the data, the rate estimates should not be substituted for the sponsor’s hazard ratio.

For the secondary continuous outcomes with detailed results, the registry also reports sponsor analyses using mixed models or linear regression, alongside simple unadjusted re-calculations from summary statistics. Those re-calculations do not reproduce the sponsor’s fitted analyses. For the fluid-overload sub-study, the recalculated unadjusted comparison and sponsor-reported comparison both indicate a negative mean difference, with their stated intervals excluding zero. For the kidney cortical T1 mapping sub-study, both analyses report an interval that includes zero. These broad patterns are similar, although the estimates and intervals are not identical.

The registry does not flag a discrepancy for these comparisons. The sponsor’s analysis remains the relevant model-based result because it reflects the analysis method recorded for each outcome; the simple re-calculations are supplementary checks, not replacements. Full model specifications and details needed to independently reproduce the sponsor analyses are not reported in the registry record.

11. Multiplicity, Missing Data and Analysis Populations

The registry record describes a primary endpoint and a range of secondary outcomes, but it does not report a multiplicity strategy. In particular, it does not say whether secondary hypotheses were tested in a hierarchy, whether p-values were adjusted, or whether a prespecified family of comparisons was used. When multiple hypotheses are tested, the chance of at least one small p-value can increase. Without a stated approach, secondary p-values should not be interpreted as though they all provide independent confirmatory evidence.

The record also does not report whether interim analyses were conducted, whether any formal stopping boundaries were used, or whether the analysis plan was amended. These details are not reported in the registry record; their absence from the record does not establish what was or was not done. A detailed statistical analysis plan is not reported in the registry record.

The primary endpoint population is identified as the randomised set, including all randomised participants whether treated or not. That definition clarifies the intended analysis population, but does not by itself explain how missing outcome information, withdrawals, treatment discontinuation, or censoring were handled. Those methods are not reported in the registry record. For time-to-event analyses, censoring rules and assumptions can affect the interpretation of the estimated event experience, so the absence of those details limits independent assessment.

For secondary outcomes, the registry record does not provide full analysis-population definitions or missing-data procedures for each outcome. It also does not report the detailed model specifications needed to evaluate every analysis. These omissions matter particularly for sub-studies, where the participants contributing measurements may differ from the full randomised set. The record lists a post-trial observational follow-up of a subset, but does not describe its analysis methods or how those findings relate to the interventional analyses.

These reporting gaps do not invalidate the posted results, but they limit the ability to evaluate robustness, reproduce analyses, and determine how strongly secondary findings were protected against multiplicity. Conclusions should therefore stay close to the reported primary comparison and distinguish it from secondary or exploratory analyses.

12. 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.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: acute kidney injury (3.5%), corona virus infection (3.2%), blood potassium increased (2.6%), blood creatinine increased (1.7%), ischaemic cardiomyopathy (1.4%).

The registry lists serious adverse event counts and other adverse event counts by assigned group, along with deaths. In the placebo group, the record lists 3305 participants at risk, 1167 with serious adverse events, 262 with other adverse events, and 353 deaths. In the empagliflozin 10 mg group, it lists 3304 at risk, 1089 with serious adverse events, 228 with other adverse events, and 314 deaths. These are registry counts, not adjusted comparisons of safety risk.

The listed serious adverse event terms include acute kidney injury (3.5%), corona virus infection (3.2%), blood potassium increased (2.6%), blood creatinine increased (1.7%), ischaemic cardiomyopathy (1.4%), and cardiac failure (1.3%). These terms describe reported events; the registry entries quoted do not establish that an event was caused by study treatment. The record does not provide the event definitions, exposure-adjusted analyses, severity distributions, or detailed assessment of relatedness needed for a fuller safety comparison.

Counts alone do not account for follow-up time or differences in participants’ baseline risks. They also do not establish whether an apparent difference between groups is statistically reliable. The safety figures should therefore be read as descriptive registry reporting, alongside the limitations in event attribution and analysis detail.

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

The primary result describes a comparison between randomized groups for a defined composite endpoint during the interventional part of EMPA-KIDNEY. The posted incidence rates and the sponsor’s Cox analysis point in the same direction: the estimated rate and hazard are lower in the empagliflozin group than in the placebo group. The sponsor reports a confidence interval for the hazard ratio that excludes 1 and p < 0.001. These findings indicate a statistically clear group difference under the reported analysis.

That statistical result does not, by itself, state how important the difference is to an individual, nor does it establish an effect for every component of the composite. The primary outcome combines kidney disease progression and adjudicated cardiovascular death. The registry record does not report separate primary estimates for those components. It also does not provide participant-level results or enough detail to translate the reported rate into an individual’s probability of an event over a particular duration.

The result applies to the studied population and trial context: patients with chronic kidney disease at risk of progression, with or without diagnosed diabetes mellitus, assigned to empagliflozin or matching placebo on top of standard of care. The registry record does not provide detailed baseline characteristics that would allow assessment of how the enrolled population compares with other groups. Nor does the primary result establish what happens beyond the recorded interventional follow-up. The record describes observational follow-up of a subset, but that is distinct from the randomized interventional comparison.

Statistical significance is not the same as proof that all assumptions were met or that the result is free of bias. The registry does not report several analysis details, including censoring and missing-data methods. Secondary results address other outcomes and should not be used to expand the primary conclusion beyond its endpoint. The findings should be understood as evidence from the reported randomized comparison, with the endpoint definition, analysis method, follow-up, and registry reporting limitations kept in view.

14. Why This Trial Matters Statistically

EMPA-KIDNEY illustrates why clinical trial results should be read through the endpoint definition and the analysis scale, not only through a p-value. The primary outcome is a composite of kidney disease progression or adjudicated cardiovascular death, and the registry reports both group incidence rates and a Cox hazard ratio. The rates relate first events to time at risk; the hazard ratio compares event hazards over time. Each offers useful information, but they are not interchangeable summaries.

The trial also shows why precision and statistical evidence deserve separate attention. The group rates have confidence intervals, and the sponsor’s comparison has its own confidence interval at a different stated confidence level. Those intervals communicate uncertainty in their respective estimates. The small reported p-value adds evidence against the null hypothesis under the sponsor’s analysis, but does not measure clinical importance or resolve questions about the endpoint’s component contributions.

Registry reporting also makes visible the difference between a prespecified primary comparison and a larger family of secondary outcomes. Secondary endpoints cover diverse outcomes and include sub-studies, while the record does not report a multiplicity strategy. This reinforces the need to avoid treating every secondary p-value as equally confirmatory. Simple re-calculations can be useful checks, but they do not replace model-based sponsor analyses or recover unreported design and analysis details.

Finally, the record illustrates that a completed trial’s public results can still leave important interpretive details unreported. Population definitions, follow-up, event definitions, and the distinction between randomized and observational evidence all shape what can be concluded. Careful statistical reading keeps conclusions attached to the comparison actually made and recognizes where the registry record does not permit stronger claims.

15. 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 limitations specific to the registry record constrain interpretation. The primary endpoint is a composite, but the record does not provide separate primary results for kidney disease progression and cardiovascular death. As a result, the contribution of each component to the reported comparison cannot be assessed from these results alone. The endpoint definition is detailed, but the record does not report the adjudication procedures or the number of events for each component.

The primary incidence rates and the sponsor’s hazard ratio are distinct summaries. The registry reports the Cox method but not the model covariates, proportional-hazards assessment, or censoring rules. The method used to calculate the incidence-rate confidence intervals is also not reported. These omissions limit independent evaluation and reproduction of the analysis.

Detailed secondary findings are available for sub-studies, but the registry record does not report enough information to assess selection into those sub-studies or the representativeness of their analysis populations. Their estimates should not be generalized automatically to the full randomised set. The record also lists additional secondary outcomes without detailed results in the information, limiting a comprehensive account of consistency across all secondary endpoints.

For safety, the posted counts are descriptive. The registry record does not provide exposure-adjusted comparisons, complete event definitions, or the detailed assessment needed to attribute events to treatment. The record also does not report the multiplicity strategy, missing-data procedures, interim analyses, or a full statistical analysis plan. These gaps do not negate the reported findings, but they restrict assessment of robustness and the strength of conclusions beyond the sponsor’s primary analysis.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: EMPA-KIDNEY (The Study of Heart and Kidney Protection With Empagliflozin) — NCT03594110.
• Registry record last updated 2025-07-20; status last verified 2025-06.
• Results first posted 2023-07-27.
• 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.

19. Record Summary

EMPA-KIDNEY was a phase 3, randomized, double-blind, parallel-group, placebo-controlled trial sponsored by Boehringer Ingelheim. It compared empagliflozin 10 mg once daily with matching placebo on top of standard of care in patients with chronic kidney disease at risk of progression, with or without diagnosed diabetes mellitus. The registry reports enrollment of 6609 and a completed status.

The primary endpoint was time to first kidney disease progression or adjudicated cardiovascular death, reported as an incidence rate per 100 patient-years at risk. The sponsor’s Cox analysis reported a hazard ratio of 0.72, with a 99.83% confidence interval from 0.59 to 0.89 and p < 0.001. The registry also reports secondary outcomes, sub-study findings, and descriptive adverse-event counts.

Interpretation is limited by unreported details on multiplicity, missing data, censoring, model specification, and component-specific primary results. Secondary outcomes and safety counts require separate, cautious interpretation.