← Clinical Trials
Heart Failure With Preserved Ejection Fraction Phase 3 Randomized NCT01920711

PARAGON-HF: Complete Statistical Analysis of LCZ696 in Heart Failure With Preserved Ejection Fraction

An independent statistical analysis of the randomized phase 3 PARAGON-HF trial evaluating LCZ696 versus valsartan on morbidity and mortality in heart failure patients with preserved ejection fraction.

PARAGON-HF  ·  Phase 3  ·  Completed  ·  Enrollment 4822
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results are taken from the ClinicalTrials.gov trial data posted on ClinicalTrials.gov for PARAGON-HF. The official registry record is the source for the registered endpoint definitions and posted statistical analyses.

Registry note: This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.

1. Trial at a Glance

PARAGON-HF was a randomized, double-blind, parallel phase 3 trial comparing LCZ696 with valsartan in heart failure patients with preserved ejection fraction. The primary endpoint was the cumulative number of cardiovascular death and total first and recurrent heart-failure hospitalizations during total follow-up time of up to 57 months.

4822
Enrollment
Total participants
2
Arms
LCZ696 vs valsartan
0.8698
Primary Rate Ratio
95% CI 0.7526–1.0052
0.6241
Primary Cox P-value
HR 0.9531
FeaturePARAGON-HF
Trial namePARAGON-HF
NCT IDNCT01920711
PhasePhase 3
ConditionHeart Failure With Preserved Ejection Fraction
AllocationRandomized
Design modelParallel
MaskingDouble
Primary purposeTreatment
Enrollment4822
InterventionsLCZ696; Valsartan
Lead sponsorNovartis Pharmaceuticals
Sponsor typeIndustry
Start2014-07-18
Primary completion2019-06-07
StatusCompleted

2. Clinical Question

The trial's central question was whether LCZ696 compared with valsartan could reduce the rate of the composite of cardiovascular death and total first and recurrent heart-failure hospitalizations in patients with heart failure with preserved ejection fraction.

Population

Heart failure patients with preserved ejection fraction. The registry endpoint definition specifies New York Heart Association (NYHA) Class II-IV and left ventricular ejection fraction (LVEF) ≥45%.

Intervention

LCZ696.

Comparator

Valsartan.

Primary question

Does LCZ696 reduce the rate of the primary composite endpoint compared with valsartan during total follow-up time of up to 57 months?

3. Trial Design

01
Randomize 4822 participants
02
LCZ696 Study intervention
03
Valsartan Comparator
04
Follow-up Up to 57 months
05
Analysis Events and clinical measures
Allocation
Randomized allocation was used to compare LCZ696 and valsartan.
Structure
The trial used a parallel design with 2 intervention arms.
Masking
The registry describes the study as double masked.
Primary purpose
Treatment.
ARM A

LCZ696

  • LCZ696
  • Randomized treatment arm
ARM B

Valsartan

  • Valsartan
  • Randomized comparator arm
What the registry data establish: PARAGON-HF was randomized, double masked, and parallel, with LCZ696 and valsartan as the two interventions. The ClinicalTrials.gov record does not provide a treatment crossover analysis, so no crossover effect is inferred here.

4. Primary Endpoint

EndpointRegistry definition / time frameEndpoint analysis
Cumulative Number of Primary Composite Events of Cardiovascular (CV) Death and Total (First and Recurrent) HF Hospitalizations Total follow up time (up to 57 months). The primary objective was to compare LCZ696 to valsartan in reducing the rate of the composite endpoint of CV death and total (first and recurrent) HF hospitalizations in HF patients (NYHA Class II-IV) with preserved ejection fraction (LVEF ≥45%). Three statistical analyses were posted: Proportional Rates Model (LWYY), Joint Frality Model, and Cox's proportional hazard model.

The endpoint is statistically distinctive because it combines cardiovascular death with total heart-failure hospitalizations, including both first and recurrent hospitalizations. That means the primary rate-based analyses are not simply asking whether a participant experienced one first event. They incorporate the cumulative burden of the composite event over follow-up.

Why recurrent events change the analysis
Total events ≠ number of participants with at least one event

When recurrent hospitalizations contribute to an endpoint, a participant can contribute more than one hospitalization to the event count. Methods designed for recurrent-event or rate-based outcomes therefore address information that a conventional first-event survival analysis does not fully represent.

5. Statistical Methodology

Primary rate analysis: Proportional Rates Model (LWYY)

The first posted primary analysis used a Proportional Rates Model (LWYY) with the treatment arm as a fixed-effect factor and stratification by region, together with a robust variance estimate. The effect measure was a rate ratio.

Rate-ratio interpretation
Rate Ratio = estimated event rate under LCZ696 ÷ estimated event rate under valsartan

A rate ratio below 1 indicates a lower estimated event rate in the LCZ696 group relative to valsartan under the fitted model. It is not a probability ratio and should not be interpreted as the proportion of participants who experienced an event.

Joint frailty model

A second posted primary analysis used a Joint Frality Model. The analysis notes specify total heart-failure hospitalizations, with treatment and region as fixed-effect factors. The reported effect measure was again a rate ratio.

A joint frailty framework is relevant when recurrent hospitalizations and mortality are statistically connected. The model can account for correlation among repeated hospitalization events and the terminal event process rather than treating each observation as independent.

Cox proportional-hazards model

The third posted primary analysis used Cox's proportional hazard model with hazard ratio as the effect measure. The analysis notes identify cardiovascular death as the component analyzed in this Cox model, and the analysis population was the Full Analysis Set.

Conceptual Cox model
h(t | X) = h0(t) exp(βX)

The hazard ratio compares the estimated instantaneous event rates between treatment groups, conditional on the fitted model. The proportional-hazards assumption concerns the relative hazard over time; it is not an assumption that event probabilities remain constant.

Full Analysis Set and intention-to-treat principle

The registry identifies the Full Analysis Set as the primary efficacy population applied in efficacy analyses for all efficacy endpoints. The primary analyses also identify intention-to-treat analysis as an analytical concept.

The statistical importance of this framework is that randomized treatment assignment remains the organizing principle for the efficacy comparison. This protects the treatment contrast created by randomization better than selectively analyzing only participants who remained on treatment.

Stratified analysis

The primary proportional-rates analysis was stratified by region, with region also identified as a fixed-effect factor in the joint frailty analysis. The primary Cox analysis notes cardiovascular death but the ClinicalTrials.gov record does not specify an additional stratification factor for that particular analysis.

6. Primary Results: Cumulative Composite Events

The primary endpoint had three posted statistical analyses. They use different models and therefore answer closely related but not identical statistical questions. They should not be treated as three independent replications of the same estimate.

Proportional Rates Model (LWYY)

Primary composite event rate ratio

0.8698

95% CI: 0.7526–1.0052   ·   P = 0.0587

Full Analysis Set; treatment as fixed-effect factor; stratified by region; robust variance estimate.

Clinical Biostats interpretation

The estimated rate ratio of 0.8698 means that the fitted event rate for the primary composite was estimated at approximately 86.98% of the corresponding rate in the valsartan group under the LWYY model. Equivalently, 0.8698 corresponds to an estimated rate approximately 13.02% lower than the comparator rate.

The rate ratio does not mean that 13.02% fewer participants necessarily experienced an event, nor does it mean that every participant had a 13.02% reduction in risk. Because the endpoint includes recurrent heart-failure hospitalizations, the rate-based interpretation concerns the modeled accumulation of events.

The 95% confidence interval of 0.7526–1.0052 describes uncertainty around the estimated rate ratio under the specified model and sampling framework. Because the interval extends above 1, the data are compatible with a rate ratio slightly above the null value as well as with materially lower rates.

The P-value of 0.0587 addresses the compatibility of the observed result with the null hypothesis under the prespecified statistical framework; it does not measure the size or clinical importance of the estimated effect. The confidence interval is more informative for understanding the precision and range of compatible rate ratios.

Joint Frailty Model

Primary composite event rate ratio

0.8511

95% CI: 0.7216–1.0039   ·   P = 0.0556

Full Analysis Set; total hospitalizations for heart failure; treatment and region as fixed-effect factors.

Clinical Biostats interpretation

The estimated rate ratio of 0.8511 indicates an estimated event rate approximately 85.11% of that in the valsartan group under the joint frailty model. In relative terms, this corresponds directly to an estimated rate approximately 14.89% lower than the comparator rate.

That estimate should not be interpreted as a 14.89% reduction in the probability of death or hospitalization for every patient. The joint frailty model is designed for an event process in which repeated heart-failure hospitalizations and the terminal event process are statistically related.

The 95% CI of 0.7216–1.0039 indicates appreciable uncertainty around the point estimate. The interval includes values just above the null value of 1, so the interval does not restrict the compatible effect entirely to rate reductions.

The P-value of 0.0556 is a hypothesis-testing quantity, not an effect-size measure. Its interpretation also depends on the prespecified superiority framework and the particular model used. It should not be read as a 94.44% probability that one treatment is better than the other.

Cox's proportional hazard model

Cardiovascular death hazard ratio

0.9531

95% CI: 0.7863–1.1551   ·   P = 0.6241

Full Analysis Set; cardiovascular death analyzed with Cox's proportional hazard model.

Clinical Biostats interpretation

The hazard ratio of 0.9531 corresponds to an estimated instantaneous cardiovascular-death hazard about 95.31% of the comparator hazard under the Cox model. A hazard ratio below 1 is consistent with a lower estimated instantaneous event rate, but the estimate is close to 1.

The hazard ratio does not mean that cardiovascular mortality was 4.69% lower in absolute terms, and it does not mean that an individual patient's probability of cardiovascular death was reduced by exactly 4.69%. Hazard is a model-based instantaneous rate, not an absolute probability.

The 95% CI of 0.7863–1.1551 is relatively broad around the point estimate and crosses 1. The interval therefore includes both lower and higher cardiovascular-death hazards relative to valsartan.

The P-value of 0.6241 does not quantify the magnitude of the observed treatment effect. It addresses the statistical evidence against the relevant null hypothesis under the Cox model. Interpretation also depends on the proportional-hazards assumption and on censoring and analysis-population conventions.

7. Comparing the Three Primary Analyses

Primary analysisEffect measureEstimate95% CIP-value
Proportional Rates Model (LWYY) Rate ratio 0.8698 0.7526–1.0052 0.0587
Joint Frality Model Rate ratio 0.8511 0.7216–1.0039 0.0556
Cox's proportional hazard model Hazard ratio 0.9531 0.7863–1.1551 0.6241

The first two analyses focus on the rate of the composite event process, while the Cox analysis is reported for cardiovascular death. Consequently, their effect measures should not be compared as if they were interchangeable versions of the same parameter.

Rate ratio

Describes the relative rate of accumulated events under a rate-based model. It is particularly relevant when recurrent hospitalizations contribute to the endpoint.

Hazard ratio

Describes a relative instantaneous event rate under the Cox model. Here, the posted analysis notes identify cardiovascular death as the analyzed component.

8. Secondary Endpoint Results

The registry contains six distinct secondary endpoint analyses in the ClinicalTrials.gov record. They illustrate several different statistical structures: longitudinal continuous outcomes, ordinal functional-class outcomes, and time-to-event renal and mortality endpoints.

Change in the Clinical Summary Score From Baseline to Month 8 by KCCQ

Least Squares Mean Difference

1.0264

95% CI: -0.0047–2.0576   ·   P = 0.0510

Baseline to Month 8; Full Analysis Set; Mixed Models Analysis.

The registry reports a mean difference of 1.0264 points, specifically a least-squares mean difference from a mixed-model analysis. The confidence interval ranges from -0.0047 to 2.0576, crossing zero. The P-value is 0.0510.

How to read this analysis

A least-squares mean difference compares model-adjusted group means rather than simply subtracting two raw arithmetic means. Here, the estimated difference is 1.0264 points from baseline to Month 8.

The confidence interval is important because it includes zero. Thus, the posted estimate is compatible with a small difference in either direction as well as with a larger positive difference within the interval. The P-value of 0.0510 should not be treated as a measure of effect magnitude.

Change From Baseline to Month 8 in NYHA Functional Class

Odds ratio

1.4475

95% CI: 1.1294–1.8552   ·   P = 0.0035

Baseline to Month 8; repeated measures cumulative odds model.

The registry reports an odds ratio of 1.4475 from a repeated-measures cumulative odds model. The analysis uses the change from baseline to any scheduled time points up to Month 8.

How to read this odds ratio

An odds ratio of 1.4475 means that the modeled odds of the relevant NYHA functional-class outcome were estimated to be 1.4475 times those in the valsartan group under the cumulative-odds model.

An odds ratio is not a risk ratio or probability ratio. If the outcome probability is not small, the numerical odds ratio can differ substantially from the corresponding probability ratio.

The 95% CI of 1.1294–1.8552 lies above 1, indicating that the interval is entirely on the same side of the null odds ratio. The P-value of 0.0035 is evidence against the null hypothesis under the posted model, but it does not establish the magnitude of benefit for an individual patient.

Participants With First Occurrence of a Composite Renal Endpoint

The registry reports three separate component analyses for the composite renal endpoint. Each uses a Cox proportional-hazards model in the Full Analysis Set, with treatment as a fixed factor and stratification by region.

Renal analysisEffect measureEstimate95% CIP-value
Composite renal endpoint Hazard ratio 0.5041 0.3312–0.7673 0.0014
Renal Death Hazard ratio 0.9295 0.0581–14.861 0.9588
Reaching ESRD Hazard ratio 0.5774 0.2272–1.4672 0.2484
≥50% decline in eGFR from baseline Hazard ratio 0.4407 0.2798–0.6942 0.0004

Composite renal endpoint

Hazard ratio

0.5041

95% CI: 0.3312–0.7673   ·   P = 0.0014

Randomization to total follow-up time (up to 57 months).

The hazard ratio of 0.5041 indicates an estimated instantaneous hazard approximately 50.41% of the valsartan-group hazard under the posted Cox model. The 95% CI of 0.3312–0.7673 quantifies uncertainty around that estimate and remains below 1.

Clinical Biostats interpretation

The estimate describes a relative time-to-event effect, not an absolute reduction in the number or percentage of participants experiencing a renal event. The confidence interval provides the principal measure of precision around the hazard-ratio estimate.

The P-value of 0.0014 provides evidence against the null hypothesis under the posted Cox analysis. It does not say that there is a 99.86% probability that LCZ696 is beneficial, nor does it quantify clinical importance.

Because this is a time-to-event analysis, censoring and the proportional-hazards assumption remain relevant to interpretation.

Renal Death

Hazard ratio

0.9295

95% CI: 0.0581–14.861   ·   P = 0.9588

Randomization to total follow-up time (up to 57 months).

The estimated hazard ratio for renal death was 0.9295. The confidence interval, 0.0581–14.861, is extremely wide and spans the null value by a large amount. The P-value was 0.9588.

Clinical Biostats interpretation

The point estimate alone is not an adequate summary here. The very wide confidence interval indicates substantial statistical uncertainty around the renal-death hazard ratio. Values representing markedly lower or markedly higher hazards are compatible with the interval.

The P-value of 0.9588 does not imply that the treatment effects are exactly equal. Rather, under the posted model and null hypothesis, the observed result does not provide strong statistical evidence against the null.

Reaching ESRD

Hazard ratio

0.5774

95% CI: 0.2272–1.4672   ·   P = 0.2484

Randomization to total follow-up time (up to 57 months).

Clinical Biostats interpretation

The estimated hazard ratio of 0.5774 is below 1, corresponding to an estimated instantaneous hazard approximately 57.74% of the comparator hazard under the fitted model. However, the 95% CI of 0.2272–1.4672 crosses 1 and is therefore compatible with both lower and higher hazards.

The P-value of 0.2484 does not provide strong evidence against the null hypothesis under the posted analysis. It should not be interpreted as a probability that the treatment has no effect.

≥50% decline in eGFR from baseline

Hazard ratio

0.4407

95% CI: 0.2798–0.6942   ·   P = 0.0004

Randomization to total follow-up time (up to 57 months).

Clinical Biostats interpretation

The hazard ratio of 0.4407 corresponds to an estimated instantaneous hazard approximately 44.07% of the valsartan-group hazard under the posted Cox model. The 95% CI of 0.2798–0.6942 remains below 1.

The P-value of 0.0004 provides statistical evidence against the null hypothesis within this specific analysis. It does not convert the hazard ratio into an absolute probability or establish that every participant experiences the same relative reduction.

All-cause Mortality

Hazard ratio

0.9696

95% CI: 0.8352–1.1255   ·   P = 0.6846

Randomization to total follow-up time (up to 57 months).

Clinical Biostats interpretation

The all-cause mortality hazard ratio of 0.9696 is close to 1. Under the Cox model, this corresponds to an estimated instantaneous mortality hazard approximately 96.96% of the comparator hazard.

The 95% CI of 0.8352–1.1255 crosses 1, indicating that the posted estimate is compatible with both lower and higher hazards. The P-value of 0.6846 does not provide strong statistical evidence against the null hypothesis under the posted analysis.

As with the other Cox results, the hazard ratio is a relative model-based measure. It is not an absolute mortality difference and does not describe the proportion of participants who benefited.

9. Secondary Results in Context

EndpointMethodEffect95% CIP-value
KCCQ Clinical Summary Score, baseline to Month 8 Mixed Models Analysis Mean difference 1.0264 -0.0047–2.0576 0.0510
NYHA Functional Class, baseline to Month 8 Repeated measures cumulative odds model OR 1.4475 1.1294–1.8552 0.0035
Composite renal endpoint Cox proportional-hazards model HR 0.5041 0.3312–0.7673 0.0014
Renal Death Cox proportional-hazards model HR 0.9295 0.0581–14.861 0.9588
Reaching ESRD Cox proportional-hazards model HR 0.5774 0.2272–1.4672 0.2484
≥50% decline in eGFR from baseline Cox proportional-hazards model HR 0.4407 0.2798–0.6942 0.0004
All-cause mortality Cox proportional-hazards model HR 0.9696 0.8352–1.1255 0.6846

The secondary results demonstrate why a trial should not be summarized by a single P-value. Different endpoints measure different clinical processes and require different statistical models. A renal composite, renal death, a longitudinal questionnaire score, a functional-class outcome, and all-cause mortality cannot be treated as interchangeable endpoints.

10. Safety Results

The ClinicalTrials.gov record reports serious adverse events by randomized arm using affected participants over participants at risk.

Safety measureLCZ696ValsartanAll Patients
Serious adverse events 1424/2419 1416/2402 2840/4821
Serious adverse events: affected / at risk
LCZ696
1424/2419
Valsartan
1416/2402

The registry reports the counts as affected/at risk. The denominators differ from the overall enrollment of 4822, which is important when interpreting the safety population. The ClinicalTrials.gov record does not provide a formal statistical comparison for serious adverse events, so no additional hypothesis test or relative safety measure is inferred.

11. Statistical Methods Explained

Why was a rate-based model used for the primary endpoint?

The primary endpoint counts cardiovascular death and total first and recurrent heart-failure hospitalizations. Because hospitalizations can recur, a rate-based analysis can incorporate the cumulative event experience rather than reducing every participant to only the first event. The registry specifically reports a Proportional Rates Model (LWYY) and a Joint Frality Model.

What does a rate ratio of 0.8698 mean?

A rate ratio of 0.8698 means that the modeled rate of the primary composite event under the LWYY analysis was estimated to be 0.8698 times the corresponding valsartan rate. The direct relative interpretation is approximately a 13.02% lower estimated rate. It is not a 13.02% reduction in the probability that an individual patient experiences the endpoint.

Why is the primary endpoint analyzed in more than one way?

The posted analyses use different statistical models for related aspects of the clinical outcome. The LWYY model addresses the primary composite event rate, the joint frailty model addresses the recurrent hospitalization and terminal-event structure, and the Cox model is reported for cardiovascular death. Their estimates are therefore not interchangeable.

What does a hazard ratio of 0.5041 mean?

The renal composite hazard ratio of 0.5041 indicates an estimated instantaneous event hazard approximately 50.41% of the comparator hazard under the fitted Cox model. It does not mean that exactly 50.41% as many participants experienced the event, nor does it represent an absolute risk reduction.

Why does the confidence interval matter more than the point estimate alone?

A point estimate is only one summary of the observed data. The confidence interval describes statistical uncertainty around that estimate under the model. For example, the primary LWYY estimate is 0.8698, but its 95% CI extends from 0.7526 to 1.0052. Reading the estimate without its interval would hide important uncertainty.

What does the P-value actually tell us?

A P-value measures how compatible the observed result is with a specified null hypothesis under the statistical model and testing framework. It is not the probability that the null hypothesis is true, the probability that the treatment works, or a measure of clinical effect size. Effect estimates and confidence intervals should therefore accompany P-values.

Why does the Cox model require attention to proportional hazards?

The Cox hazard ratio summarizes the relative hazard under a proportional-hazards model. If the relative hazards change substantially over time, one constant hazard ratio may not fully describe the treatment difference. The ClinicalTrials.gov record identifies the Cox method but do not provide a separate assessment of the proportional-hazards assumption.

12. Intention-to-Treat Analysis and Censoring

The primary efficacy analyses used the Full Analysis Set, described in the registry as the primary efficacy population applied to efficacy analyses for all efficacy endpoints. The primary analyses also identify intention-to-treat analysis as an analytical concept.

This distinction matters because randomized treatment assignment defines the principal comparison. A treatment-effect estimate based on the randomized groups preserves the causal structure created by randomization more directly than an analysis that excludes participants based on post-randomization behavior.

Time-to-event analyses additionally depend on censoring. Participants who have not experienced the event by the time their available follow-up ends do not simply disappear from the analysis; their observed follow-up contributes information up to the censoring point under the assumptions of the analysis.

Registry limitation: The ClinicalTrials.gov record identifies the analysis populations and statistical methods but do not provide detailed missing-data or censoring rules. Those rules should therefore not be reconstructed from assumptions.

13. Longitudinal Analysis of KCCQ

The KCCQ Clinical Summary Score endpoint was measured from baseline to Month 8 and analyzed using Mixed Models Analysis. The statistical method is a mixed-effects model, with the effect measure reported as a least-squares mean difference.

Why mixed models are useful for longitudinal data
Outcomeij = fixed effects + participant-specific variation + residual variation

Repeated measurements from the same participant are correlated. A mixed-effects framework can represent that within-participant structure rather than treating every repeated observation as statistically independent.

The reported estimate of 1.0264 points is a model-based between-group difference at the specified longitudinal analysis. It should not be confused with a simple unadjusted difference calculated only among participants with complete Month 8 observations.

14. Ordinal Analysis of NYHA Functional Class

The NYHA endpoint was analyzed using a repeated measures cumulative odds model, a mixed model for repeated measures. The analysis evaluates change from baseline to any scheduled time points up to Month 8.

Because functional class is ordered rather than a naturally continuous measurement, an odds-ratio interpretation differs from the mean-difference interpretation used for the KCCQ score. The cumulative-odds framework models the ordering of the outcome categories rather than assuming that the numerical distance between categories is equivalent.

Odds ratio

The posted estimate was 1.4475, with a 95% CI of 1.1294–1.8552.

Hypothesis test

The posted P-value was 0.0035. This is evidence against the relevant null hypothesis under the specified model, not a measure of effect size.

15. Renal Endpoint Structure

The renal endpoint results illustrate an important statistical distinction between a composite endpoint and its individual components. The ClinicalTrials.gov record reports a composite renal endpoint along with separate analyses for renal death, reaching ESRD, and a ≥50% decline in eGFR from baseline.

Renal component analysisHR95% CIP-value
Composite renal endpoint0.50410.3312–0.76730.0014
Renal Death0.92950.0581–14.8610.9588
Reaching ESRD0.57740.2272–1.46720.2484
≥50% decline in eGFR from baseline0.44070.2798–0.69420.0004

The individual components have different event frequencies and therefore different statistical precision. The extremely wide renal-death confidence interval illustrates how a point estimate can be unstable when relatively few events contribute information. A composite can have greater statistical information than one of its individual components, but a composite result should not automatically be attributed to every component.

16. Multiplicity and Multiple Endpoint Interpretation

The ClinicalTrials.gov record identifies one registered primary endpoint and multiple posted statistical analyses, together with several secondary endpoints. The primary hypothesis type is superiority.

Analysis familyRole in the ClinicalTrials.gov recordKey issue
Primary composite endpointPrimaryMultiple statistical models are posted for related aspects of the endpoint.
KCCQ Clinical Summary ScoreSecondaryLongitudinal continuous outcome.
NYHA Functional ClassSecondaryRepeated-measures cumulative odds analysis.
Renal endpointsSecondaryComposite and component time-to-event analyses.
All-cause mortalitySecondaryTime-to-event analysis using Cox regression.

Multiplicity is important because a trial can generate multiple statistical tests. The more hypotheses tested, the more carefully nominal P-values need to be interpreted in the context of the prespecified testing strategy. The ClinicalTrials.gov record does not provide a multiplicity-adjustment scheme, alpha allocation, or hierarchical testing procedure, so no such procedure is inferred here.

Important distinction: a statistically small P-value for one secondary endpoint should not automatically be treated as proof that every other endpoint is positive. Each endpoint has its own estimand, model, confidence interval, and hypothesis test, and the overall inferential framework depends on the prespecified multiplicity strategy.

17. Why a P-value Does Not Measure Effect Size

PARAGON-HF provides several useful examples of why statistical significance and effect magnitude are separate concepts.

ResultEstimate95% CIP-value
Primary LWYY rate ratio0.86980.7526–1.00520.0587
NYHA odds ratio1.44751.1294–1.85520.0035
Renal composite HR0.50410.3312–0.76730.0014
All-cause mortality HR0.96960.8352–1.12550.6846

These examples span different estimands and different scales. A P-value cannot substitute for the estimate and its confidence interval. The estimate describes the observed direction and magnitude on the chosen statistical scale, while the confidence interval describes uncertainty around it.

18. What the Primary Rate Ratio Does — and Does Not — Mean

Statistical interpretation

The primary LWYY rate ratio of 0.8698 means that the estimated rate of the primary composite event under the fitted model was 0.8698 times the corresponding rate under valsartan.

It does not mean that 86.98% of patients avoided an event, that 13.02% of patients were protected, or that every patient experienced a 13.02% reduction in risk.

Why the confidence interval matters

The 95% CI of 0.7526–1.0052 shows the uncertainty around the rate-ratio estimate. The interval includes values below 1 and values slightly above 1, so the estimate should be interpreted together with this uncertainty rather than as an isolated point estimate.

Why the model matters

The primary endpoint includes recurrent heart-failure hospitalizations. Consequently, a rate ratio from the LWYY analysis and a hazard ratio from a Cox analysis represent different statistical quantities. A direct numerical comparison between 0.8698 and 0.9531 would not be a comparison of like with like.

19. Important Limitations and Interpretation Issues

20. Why This Trial Matters Statistically

PARAGON-HF is a useful teaching case because its posted analyses span several major areas of clinical-trial statistics within one randomized comparison. The primary endpoint combines a terminal event with recurrent hospitalizations, while the secondary endpoints extend into longitudinal, ordinal, renal, and mortality outcomes.

ConceptHow it appears in PARAGON-HF
RandomizationRandomized comparison of LCZ696 and valsartan.
Double maskingThe registry describes the trial as double masked.
Intention-to-treat analysisIdentified among the analytical concepts for primary and secondary efficacy analyses.
Full Analysis SetPrimary efficacy population for efficacy analyses.
Rate ratioUsed for the primary composite event analysis.
Recurrent eventsTotal first and recurrent heart-failure hospitalizations are part of the primary endpoint.
Joint frailty modelPosted as a primary analysis involving total heart-failure hospitalizations.
Cox proportional-hazards modelUsed for cardiovascular death and multiple secondary time-to-event endpoints.
Hazard ratioReported for cardiovascular death, renal endpoints, and all-cause mortality.
Mixed-effects modelUsed for the KCCQ Clinical Summary Score analysis.
Repeated-measures cumulative oddsUsed for the NYHA Functional Class analysis.
Stratified analysisRegion is incorporated into the primary rate-based analyses.
Confidence intervalsPosted for all registry-reported statistical effect estimates.
P-valuesPosted for all registry-reported statistical analyses.

21. Related Tutorials

Learn more about the methods used in this trial:

22. Related Calculators

23. Sources

Continue through the Clinical Biostats statistical pathway

Connect the trial's endpoints and methods to deeper statistical tutorials and practical analysis tools.

24. Record Summary

PARAGON-HF provides a particularly useful statistical case study because its primary endpoint combines cardiovascular death with total first and recurrent heart-failure hospitalizations. The registry consequently reports both rate-based analyses and a Cox analysis, while its secondary endpoints extend to longitudinal KCCQ measurement, repeated-measures NYHA functional class, renal time-to-event outcomes, and all-cause mortality.

The primary LWYY analysis produced a rate ratio of 0.8698 with a 95% CI of 0.7526–1.0052 and P = 0.0587. The joint frailty analysis produced a rate ratio of 0.8511 with a 95% CI of 0.7216–1.0039 and P = 0.0556. The posted Cox analysis for cardiovascular death produced an HR of 0.9531 with a 95% CI of 0.7863–1.1551 and P = 0.6241.

The secondary analyses illustrate why interpretation should remain endpoint-specific. The KCCQ analysis reported a least-squares mean difference of 1.0264, the NYHA analysis reported an odds ratio of 1.4475, renal analyses reported hazard ratios ranging from 0.4407 to 0.9295, and all-cause mortality had an HR of 0.9696. These estimates operate on different statistical scales and should not be collapsed into a single summary measure.

Clinical Biostats methodology: A rigorous trial-results page should distinguish the endpoint being analyzed, the estimand, the statistical model, the effect measure, the confidence interval, and the hypothesis test. PARAGON-HF demonstrates why that distinction is especially important when a primary endpoint includes recurrent events and when multiple secondary outcomes use different analytical frameworks.