This page separates reported trial results from statistical interpretation. Numerical results are taken only from the ClinicalTrials.gov data posted on ClinicalTrials.gov for NCT03732677. The registry provides the official trial record.
1. Trial at a Glance
NIAGARA is a randomized, parallel, phase 3 trial in muscle invasive bladder cancer. The trial compares durvalumab plus gemcitabine and cisplatin with gemcitabine and cisplatin, with two registered primary endpoints: pathologic complete response at cystectomy and event-free survival.
| Feature | NIAGARA |
|---|---|
| Trial name | NIAGARA |
| Phase | Phase 3 |
| Condition | Muscle Invasive Bladder Cancer |
| Design | Randomized, parallel, unmasked |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 1063 |
| Primary endpoints | Pathologic complete response (pCR) rates at time of cystectomy; event-free survival (EFS) |
| Primary endpoint types | Binary; Time-to-event |
| Results posted | Yes |
| Statistical analyses posted | 7 |
| Lead sponsor | AstraZeneca |
| Status | ACTIVE_NOT_RECRUITING |
| ClinicalTrials.gov | NCT03732677 |
2. Clinical Question
The central question is whether the durvalumab-containing treatment strategy improves pathologic complete response and event-free survival compared with gemcitabine and cisplatin alone in patients with muscle invasive bladder cancer.
Population
Patients with muscle invasive bladder cancer enrolled in the phase 3 NIAGARA randomized trial.
Intervention
Durvalumab with gemcitabine and cisplatin as neoadjuvant treatment, followed by durvalumab as adjuvant treatment.
Comparator
Gemcitabine and cisplatin.
Primary question
Does the durvalumab-containing strategy improve pCR rates at cystectomy and event-free survival relative to gemcitabine and cisplatin?
3. Trial Design
Durvalumab + Gemcitabine + Cisplatin
- Durvalumab
- Gemcitabine
- Cisplatin
- Neoadjuvant treatment followed by adjuvant durvalumab
Gemcitabine + Cisplatin
- Gemcitabine
- Cisplatin
4. Endpoints
| Endpoint | Registry definition / time frame | Type |
|---|---|---|
| Pathologic Complete Response (pCR) Rates at Time of Cystectomy | pCR rate is defined as the proportion of patients whose pathological staging was T0N0M0 as assessed per central pathology review using specimens obtained via radical cystectomy following the neoadjuvant treatment. The denominator for pCR will be the number of patients in the FAS. Time frame: Up to 6 months. | Binary |
| Event-free Survival (EFS) Per Central Review Defined as Time From Randomization to Event | EFS is defined as the time from randomization to the first recurrence of disease post radical cystectomy, time of first documented progression in patients who were medically precluded for radical cystectomy, or time of expected surgery in patients who refuse to undergo a radical cystectomy or failure to undergo a radical cystectomy in participants with residual disease, or the time of death due to. Time frame: Up to 48 months. | Time-to-event |
| Event-free Survival at 24 Months (EFS24) Per Central Review Defined as Time From Randomization to Event | Time frame: Up to 24 months. | Time-to-event |
| Proportion of Patients Who Undergo Cystectomy | Time frame: Up to 6 months. | Binary |
| Overall Survival | Time frame: Up to 65 months. | Time-to-event |
| Metastasis-free Survival Per Investigator Assessment or Local Biopsy Review. | Time frame: Up to 48 months. | Time-to-event |
| Disease-specific Survival Per Investigator Assessment or Local Biopsy Review. | Time frame: Up to 48 months. | Time-to-event |
5. Analysis Populations and Stratification
The primary analyses were based on the Full Analysis Set (FAS), described in the registry as including all randomized patients. The pCR analysis explicitly identifies the FAS as the denominator population, while the EFS analysis likewise uses the FAS.
| Endpoint / analysis | Analysis population | Key adjustment information |
|---|---|---|
| pCR | Full Analysis Set (Intention to Treat): includes all randomized subjects | Logistic regression adjusting for renal function, tumor stage and PDL1 status |
| EFS | Full Analysis Set (FAS): includes all randomized patients | Stratified Cox proportional hazard model adjusting for renal function, tumor stage and PDL1 status |
| EFS24 | Full Analysis Set (FAS), intention to treat, including all randomized patients | Stratified Cox model adjusting for renal function, tumor stage and PDL1 status for HR and CI |
| Cystectomy | Full Analysis Set (FAS), intention to treat, including all randomized patients | Stratified logistic regression adjusting for renal function, tumor stage and PDL1 status |
| Metastasis-free survival | Full Analysis Set (FAS), intention to treat, including all randomized patients | Stratified Cox model adjusting for renal function, tumor stage and PDL1 status |
| Disease-specific survival | Full Analysis Set (FAS), intention to treat, including all randomized patients | Stratified Cox model adjusting for renal function, tumor stage and PDL1 status |
The reported stratification and adjustment variables were renal function (adequate vs borderline), tumor stage (T2N0 vs >T2N0), and PDL1 status (high vs low/negative). These variables therefore enter the statistical interpretation of several reported treatment-effect estimates.
6. Statistical Methodology
Logistic regression for pCR
Pathologic complete response is a binary outcome: each participant either meets the registered pCR definition or does not. The registry reports logistic regression for the primary pCR analysis, with adjustment for renal function, tumor stage and PDL1 status.
The treatment coefficient is transformed into an odds ratio. An odds ratio compares the odds of the binary outcome between treatment groups after the specified covariate adjustment.
Odds ratios
The pCR analysis reports an odds ratio of 1.60. An odds ratio above 1 indicates higher estimated odds of the outcome in the durvalumab-containing group relative to gemcitabine and cisplatin under the fitted model.
Importantly, an odds ratio is not a risk ratio and does not mean that the probability of pCR was 1.60 times as large. The conversion between odds and probability depends on the underlying event probability.
Kaplan-Meier estimation and time-to-event endpoints
EFS, EFS24, overall survival, metastasis-free survival and disease-specific survival are time-to-event outcomes. Kaplan-Meier estimation is the standard framework for describing the probability of remaining event-free or alive over time while accommodating right censoring.
Here, di is the number of events at time ti and ni is the number at risk immediately before that time.
Log-rank testing
The registry reports log-rank testing for the primary EFS endpoint and for overall survival, metastasis-free survival and disease-specific survival. The log-rank test evaluates evidence that the event-time distributions differ between randomized groups.
Cox proportional-hazards modeling
The registry states that the EFS hazard ratio and confidence interval were estimated from a stratified Cox proportional hazard model adjusting for renal function, tumor stage and PDL1 status. The same Cox-model framework is identified in the notes for EFS24, metastasis-free survival and disease-specific survival.
A hazard ratio is a relative time-to-event measure. It is not the same as a relative risk, an absolute risk difference, or the probability that an individual patient will experience the event.
Intention-to-treat analysis
The FAS is explicitly described as including all randomized patients, and several analyses are identified as intention-to-treat. This preserves the treatment comparison created by randomization rather than restricting efficacy analysis to patients who completed treatment or followed a particular treatment pathway.
Stratified analysis
Stratification and covariate adjustment were used for renal function, tumor stage and PDL1 status. Statistically, this allows the treatment effect to be estimated while accounting for these prespecified factors rather than treating the randomized population as completely homogeneous with respect to them.
7. Primary Results
Pathologic Complete Response
Adjusted odds ratio for pCR
95% CI: 1.227–2.084 · P = 0.0005
Two-sided 95% confidence interval; logistic regression in the Full Analysis Set.
| Primary endpoint | Analysis | Effect measure | Estimate | Confidence interval | P-value |
|---|---|---|---|---|---|
| pCR rates at time of cystectomy | Logistic regression | Odds ratio | 1.60 | 95% CI 1.227–2.084 | 0.0005 |
The registry describes this as a strata-adjusted odds ratio from logistic regression adjusting for renal function, tumor stage and PDL1 status. The analysis population is the FAS, which includes all randomized subjects.
An odds ratio of 1.60 means that the estimated odds of meeting the registered pCR definition were 1.60 times as high in the durvalumab + gemcitabine + cisplatin group as in the gemcitabine + cisplatin group, after the reported adjustment for renal function, tumor stage and PDL1 status.
This does not mean that 60% more patients achieved pCR, nor does it mean that the probability of pCR was exactly 60% higher. Odds and probabilities are different quantities.
The two-sided 95% CI of 1.227–2.084 describes uncertainty around the estimated odds ratio under the fitted logistic model and statistical framework. It does not describe the range of treatment effects experienced by individual patients.
The p-value of 0.0005 addresses the compatibility of the observed data with the statistical null hypothesis under the specified analysis. It does not measure the size or clinical importance of the treatment effect.
Because this is a covariate-adjusted odds ratio, its interpretation depends on the specified model and adjustment variables. It should not be substituted for an absolute pCR difference that is not reported in the ClinicalTrials.gov record.
Event-free Survival
Hazard ratio for EFS
Registry CI fields: 95.877% CI 0.554–0824 · P < 0.0001
Two-sided confidence interval; log-rank analysis with HR and CI estimated from a stratified Cox proportional hazard model.
| Primary endpoint | Analysis | Effect measure | Estimate | Confidence interval | P-value |
|---|---|---|---|---|---|
| EFS per central review | Log-rank test; stratified Cox model for HR and CI | Hazard ratio | 0.68 | 95.877% CI 0.554–0824 | <0.0001 |
The EFS analysis used the FAS and compared durvalumab + gemcitabine + cisplatin with gemcitabine + cisplatin. The registry states that the HR and CI were estimated from a stratified Cox proportional hazard model adjusting for renal function, tumor stage and PDL1 status.
An HR of 0.68 means that, under the fitted Cox model, the estimated instantaneous rate of an EFS event in the durvalumab-containing group was approximately 68% of that in the comparator group. Equivalently, 0.68 corresponds to an estimated 32% lower hazard because 1 − 0.68 = 0.32.
This does not mean that 32% of patients avoided an event, that individual patients experienced a 32% reduction in risk, or that the absolute probability of an event was reduced by 32 percentage points.
The confidence interval quantifies uncertainty around the model-based HR. The registry fields and analysis notes contain two different CI representations, so the discrepancy itself is important to preserve when interpreting the posted result.
The p-value of <0.0001 indicates strong statistical evidence against the relevant null hypothesis under the reported test framework. It is not a measure of the magnitude of the hazard ratio and does not by itself quantify clinical importance.
Because the HR comes from a Cox proportional-hazards model, its interpretation relies on the model's assumptions. In particular, a single HR is most straightforward when the relative hazards are reasonably described by the proportional-hazards structure over follow-up.
8. Secondary Endpoint Results
Event-free Survival at 24 Months
Reported hazard ratio
95% CI: 0.62–0.90 · P = 0.0021
The registry states that the p-value is based on a chi-squared test with one degree of freedom. The HR and CI were estimated from a stratified Cox proportional hazard model.
The EFS24 analysis was conducted in the FAS on an intention-to-treat basis. The reported Cox model adjusted for renal function, tumor stage and PDL1 status.
An HR of 0.75 corresponds to an estimated 25% lower instantaneous event rate under the reported hazard-ratio framework. It does not mean that the 24-month event probability was reduced by 25 percentage points.
The 95% CI of 0.62–0.90 describes uncertainty around the estimated HR. The reported p-value of 0.0021 concerns the statistical test and does not itself quantify treatment-effect size.
There is also an important methodological distinction in the registry record: the p-value is identified as coming from a chi-squared test with one degree of freedom, whereas the HR and CI are identified as coming from a stratified Cox model. These statistics therefore should not be treated as if they were all generated by a single simple calculation.
Proportion of Patients Who Undergo Cystectomy
Adjusted odds ratio for undergoing cystectomy
95% CI: 1.047–2.095 · P = 0.0265
Two-sided confidence interval; stratified logistic regression.
The analysis used the FAS and was described as intention-to-treat. The reported logistic regression was adjusted for renal function, tumor stage and PDL1 status.
An OR of 1.48 means that the estimated odds of undergoing cystectomy were 1.48 times those in the comparator group under the specified adjusted logistic model.
It does not mean that 48% more participants underwent cystectomy, because an odds ratio is not a percentage-point difference or a risk ratio.
The 95% CI of 1.047–2.095 describes uncertainty around the adjusted odds ratio. The p-value of 0.0265 measures evidence against the statistical null hypothesis under the reported analysis, not the size of the effect.
Overall Survival
Hazard ratio for overall survival
Registry CI fields: 98.457% CI 0.563–0.985 · P = 0.0106
The analysis used a log-rank test; the registry notes stratification by renal function, tumor stage and PDL1 status.
The registry-reported analysis notes additionally state 95% CI: 0.594 to 0.934. As with EFS, both numerical representations are retained because the ClinicalTrials.gov record contains both.
An HR of 0.75 indicates an estimated instantaneous death rate approximately 75% of that in the comparator group under the reported time-to-event framework. Expressed as a relative complement, this corresponds to a 25% lower estimated hazard.
The HR does not give the absolute probability of death by a particular time and does not mean that every patient experienced the same proportional reduction.
The registry fields give a 98.457% CI of 0.563–0.985, while the analysis notes give a 95% CI of 0.594–0.934. These should not be silently combined or substituted for one another.
The p-value of 0.0106 is evidence from the reported statistical test; it does not quantify effect size, patient-level benefit, or the clinical importance of the estimate.
Metastasis-free Survival
Hazard ratio
95% CI: 0.541–0.826 · P = 0.0002
Log-rank test with HR and CI estimated from a stratified Cox proportional hazard model.
The registry describes this analysis as based on the FAS, intention to treat, with adjustment for renal function, tumor stage and PDL1 status.
Disease-specific Survival
Hazard ratio
95% CI: 0.516–0.907 · P = 0.0081
Log-rank test with HR and CI estimated from a stratified Cox proportional hazard model.
This analysis likewise used the FAS and the reported stratification factors of renal function, tumor stage and PDL1 status.
| Secondary endpoint | Effect measure | Estimate | Confidence interval | P-value |
|---|---|---|---|---|
| EFS at 24 months | Hazard ratio | 0.75 | 95% CI 0.62–0.90 | 0.0021 |
| Proportion undergoing cystectomy | Odds ratio | 1.48 | 95% CI 1.047–2.095 | 0.0265 |
| Overall survival | Hazard ratio | 0.75 | 98.457% CI 0.563–0.985; analysis notes also state 95% CI 0.594–0.934 | 0.0106 |
| Metastasis-free survival | Hazard ratio | 0.67 | 95% CI 0.541–0.826 | 0.0002 |
| Disease-specific survival | Hazard ratio | 0.69 | 95% CI 0.516–0.907 | 0.0081 |
The secondary results show a consistent pattern of estimates below 1 for the reported time-to-event hazard ratios and estimates above 1 for the two reported odds ratios. That pattern is descriptive of the posted analyses; it should not be converted into a single composite treatment effect.
9. Safety
The ClinicalTrials.gov record reports serious adverse events by randomized arm as affected participants divided by participants at risk.
| Safety measure | Durvalumab + Gemcitabine + Cisplatin | Gemcitabine + Cisplatin |
|---|---|---|
| Serious adverse events | 326 / 530 | 287 / 526 |
These figures describe the number affected and the number at risk in each arm. The ClinicalTrials.gov record does not provide additional safety endpoint definitions, event-specific categories, severity distributions, or formal between-group safety tests, so no such analyses are added here.
10. Statistical Methods Explained
Why was logistic regression used for pCR?
pCR is defined as a binary outcome: a participant either meets the T0N0M0 definition or does not. Logistic regression is designed for binary outcomes and allows the treatment comparison to be adjusted for renal function, tumor stage and PDL1 status. The resulting odds ratio summarizes the adjusted relative odds of pCR.
What does an odds ratio of 1.60 mean?
An OR of 1.60 means the estimated odds of pCR were 1.60 times as high in the durvalumab-containing group under the fitted adjusted model. It is not equivalent to saying that the pCR probability increased by 60%, because odds and probability are mathematically different.
Why was a hazard ratio used for EFS?
EFS is measured as time from randomization until a defined event. Some participants may remain event-free when observation ends, producing right-censored data. A Cox proportional-hazards model is a standard way to estimate a relative event-rate measure while incorporating differing follow-up times.
What does an EFS HR of 0.68 mean?
Under the reported Cox model, the estimated instantaneous event rate in the durvalumab-containing group was approximately 68% of that in the comparator group. The complementary calculation, 1 − 0.68, gives 0.32, or a 32% lower estimated hazard. That is not a 32-percentage-point reduction in event probability.
Why were renal function, tumor stage and PDL1 status included?
The registry explicitly identifies these variables in the adjusted and stratified analyses. Accounting for them allows the reported treatment-effect estimates to reflect the specified statistical comparison while controlling for these trial stratification factors.
Why are the confidence intervals important?
A point estimate such as an HR of 0.68 or an OR of 1.60 is only one estimate from the observed data. The confidence interval communicates statistical uncertainty around that estimate. It does not describe the range of effects that individual patients experience.
Why should the p-value not be treated as the effect size?
A p-value addresses evidence against a null hypothesis under a specified statistical model and test. It depends on both the magnitude of the observed difference and the information available in the analysis. The effect estimate and confidence interval are therefore needed to understand the direction, magnitude and precision of the treatment comparison.
11. Interpreting the Pattern of Results
Binary endpoint
The primary pCR analysis uses an adjusted odds ratio of 1.60, linking the treatment comparison to a binary pathologic response outcome.
Time-to-event endpoint
The primary EFS analysis uses a hazard ratio of 0.68, reflecting a model-based comparison of event rates over time.
Adjustment
Renal function, tumor stage and PDL1 status appear in the reported adjustment or stratification framework.
Multiple outcomes
The registry reports several secondary efficacy analyses in addition to the two primary endpoints. Their statistical interpretation should remain endpoint-specific.
The different effect measures should not be placed on a common numerical scale. An odds ratio describes relative odds for a binary outcome, while a hazard ratio describes a relative event rate within a time-to-event framework. A value of 1.60 for pCR and a value of 0.68 for EFS therefore answer different statistical questions.
Likewise, the reported p-values should not be interpreted as a ranking of endpoints. The primary pCR analysis has a p-value of 0.0005, while the primary EFS analysis reports P < 0.0001. These are results from different endpoint types and statistical procedures.
12. Confidence Intervals and Precision
| Analysis | Estimate | Reported interval | What the interval contributes |
|---|---|---|---|
| Primary pCR | OR 1.60 | 95% CI 1.227–2.084 | Quantifies uncertainty around the adjusted odds ratio. |
| Primary EFS | HR 0.68 | 95.877% CI 0.554–0824; analysis notes also state 95% CI 0.558–0.817 | Provides uncertainty around the model-based EFS hazard ratio, with an internal registry discrepancy that should be preserved. |
| EFS24 | HR 0.75 | 95% CI 0.62–0.90 | Shows uncertainty around the time-to-event estimate. |
| Cystectomy | OR 1.48 | 95% CI 1.047–2.095 | Shows uncertainty around the adjusted odds ratio. |
| Overall survival | HR 0.75 | 98.457% CI 0.563–0.985; analysis notes also state 95% CI 0.594–0.934 | Shows uncertainty while highlighting another CI-level/value discrepancy in the ClinicalTrials.gov record. |
| Metastasis-free survival | HR 0.67 | 95% CI 0.541–0.826 | Shows uncertainty around the estimated hazard ratio. |
| Disease-specific survival | HR 0.69 | 95% CI 0.516–0.907 | Shows uncertainty around the estimated hazard ratio. |
Confidence intervals are particularly useful because they prevent a point estimate from being interpreted as exact. For example, the EFS estimate is 0.68, but the interval reported in the registry fields spans a range of values around that point estimate. The additional CI reported in the analysis notes reinforces the need to distinguish the posted statistical fields from a single reconstructed value.
13. Primary Endpoint Interpretation
The adjusted OR of 1.60 indicates higher estimated odds of the registered pathologic complete response in the durvalumab-containing group. The estimate comes from logistic regression and is adjusted for renal function, tumor stage and PDL1 status.
The HR of 0.68 indicates a lower estimated instantaneous event rate in the durvalumab-containing group under the reported stratified Cox model. The interpretation concerns time to the registered EFS event, not a fixed-time risk difference.
Together, these endpoints illustrate two complementary approaches to clinical-trial analysis. pCR captures a binary pathologic outcome assessed at cystectomy, while EFS incorporates the timing of clinically relevant events after randomization. The statistical methods appropriately differ because the outcome structures differ.
14. Limitations
- Registry-level data: This analysis is limited to the numerical and methodological information contained in the ClinicalTrials.gov extract. It does not add unreported baseline characteristics, subgroup estimates, median event times, or additional safety details.
- Confidence-interval discrepancies: The registry-reported EFS and overall-survival analyses contain CI fields and analysis-note values that differ in level and/or numerical limits. Both versions are reported rather than silently reconciled.
- Hazard-ratio interpretation: Cox-model HRs are model-based and depend on the proportional-hazards framework. A single HR should not automatically be interpreted as a constant individual-level risk reduction.
- Odds-ratio interpretation: Odds ratios should not be read as risk ratios or absolute probability differences.
- Multiplicity: The ClinicalTrials.gov record identifies two primary endpoints and several secondary analyses, but do not provide a multiplicity-control strategy. The posted p-values therefore should not be assigned an unstated multiplicity procedure.
- Missing data: The ClinicalTrials.gov record does not specify missing-data handling or imputation methods. No such method is inferred.
- Interim analysis: The ClinicalTrials.gov record does not identify an interim-analysis strategy or alpha-spending plan. None is inferred.
- Crossover: The ClinicalTrials.gov record does not report crossover. No crossover adjustment or interpretation is added.
- Bayesian methods: No Bayesian analysis is identified in the ClinicalTrials.gov record.
15. Why This Trial Matters Statistically
NIAGARA is a useful teaching example because its posted analyses combine binary and time-to-event endpoints within a randomized phase 3 framework. The primary pCR endpoint illustrates adjusted logistic regression and odds ratios, while EFS illustrates stratified survival analysis, log-rank testing and Cox-model hazard ratios.
| Concept | How it appears in NIAGARA |
|---|---|
| Randomization | 1063 participants enrolled in a randomized, parallel phase 3 trial with 2 arms. |
| Intention-to-treat analysis | The FAS includes all randomized patients and is used for primary efficacy analyses. |
| Binary endpoint analysis | pCR and cystectomy use logistic regression. |
| Odds ratio | pCR and cystectomy treatment effects are reported as odds ratios. |
| Time-to-event analysis | EFS, EFS24, overall survival, metastasis-free survival and disease-specific survival use time-to-event methods. |
| Log-rank test | Reported for EFS, overall survival, metastasis-free survival and disease-specific survival. |
| Cox model | Used to estimate hazard ratios and confidence intervals for the reported survival analyses. |
| Covariate adjustment | Renal function, tumor stage and PDL1 status are included in reported adjusted analyses. |
| Stratified analysis | Renal function, tumor stage and PDL1 status form the reported stratification/adjustment framework. |
| Confidence intervals | Reported for all seven posted statistical analyses. |
| P-values | Reported for all seven posted statistical analyses. |
| Safety analysis | Serious adverse events are reported as affected participants divided by participants at risk by arm. |
16. Statistical Methods and Endpoint Map
| Endpoint | Outcome structure | Reported method | Effect measure |
|---|---|---|---|
| pCR | Binary | Logistic regression | Odds ratio |
| EFS | Time-to-event | Log-rank; stratified Cox model for HR/CI | Hazard ratio |
| EFS24 | Time-to-event | Chi-squared for p-value; stratified Cox model for HR/CI | Hazard ratio |
| Cystectomy | Binary | Logistic regression | Odds ratio |
| Overall survival | Time-to-event | Log-rank | Hazard ratio |
| Metastasis-free survival | Time-to-event | Log-rank | Hazard ratio |
| Disease-specific survival | Time-to-event | Log-rank | Hazard ratio |
This mapping makes an important statistical point: the same randomized treatment comparison can generate several different effect measures because each endpoint has a different data structure. Choosing the analysis method is therefore driven by the endpoint definition rather than by the treatment itself.
17. A Closer Look at the Primary pCR Analysis
The pCR endpoint is defined using a specific pathologic criterion, T0N0M0, assessed through central pathology review of radical cystectomy specimens after neoadjuvant treatment. The registry specifies that the denominator is the number of patients in the FAS.
The logistic-regression approach then estimates the treatment association while adjusting for renal function, tumor stage and PDL1 status. This is different from simply comparing two raw proportions with an unadjusted chi-squared test.
What the model adds
It permits the reported odds ratio to account for the specified renal function, tumor stage and PDL1 variables.
What the model does not add
It does not create an absolute pCR difference that is absent from the ClinicalTrials.gov record.
The estimate of 1.60 should therefore be read specifically as a model-based adjusted odds ratio, not as a generic “60% improvement” statistic.
18. A Closer Look at the Primary EFS Analysis
EFS is fundamentally different from pCR because the time until the first qualifying event matters. A participant who remains event-free for a longer period contributes different information from a participant who experiences an event earlier, while a participant without an observed event by the end of follow-up may be censored.
The registry reports a log-rank test and identifies a stratified Cox proportional hazard model for estimation of the EFS HR and CI. The model adjusts for renal function, tumor stage and PDL1 status.
The complementary value of 0.32 corresponds to a 32% lower estimated hazard. This arithmetic is a translation of the reported HR, not a separate reported endpoint.
The distinction between hazard and probability is critical. An HR does not tell us directly what fraction of participants were event-free at 48 months, nor does it provide a median EFS unless those quantities are separately reported.
19. Secondary Survival Endpoints
The secondary time-to-event analyses provide a broader statistical picture without changing the definition of the primary endpoints. Metastasis-free survival has a reported HR of 0.67, while disease-specific survival has a reported HR of 0.69. Overall survival has a reported HR of 0.75.
Metastasis-free survival
HR 0.67; 95% CI 0.541–0.826; P = 0.0002.
Disease-specific survival
HR 0.69; 95% CI 0.516–0.907; P = 0.0081.
Overall survival
HR 0.75; registry fields give 98.457% CI 0.563–0.985; P = 0.0106.
EFS24
HR 0.75; 95% CI 0.62–0.90; P = 0.0021.
These estimates are related because they are all time-to-event outcomes, but they are not interchangeable. Each endpoint has its own event definition and time frame, so each hazard ratio answers a different question.
20. Important Statistical Distinctions
| Distinction | Why it matters |
|---|---|
| Odds ratio vs hazard ratio | An OR compares odds of a binary outcome; an HR compares instantaneous event rates over time. |
| p-value vs effect estimate | The p-value addresses evidence against a null hypothesis; the OR or HR describes the estimated treatment effect. |
| Confidence interval vs individual variability | A confidence interval describes statistical uncertainty around an estimate, not the range of outcomes for individual patients. |
| Absolute risk vs relative effect | Neither the reported ORs nor HRs directly provides an absolute percentage-point difference. |
| Primary vs secondary endpoints | Primary endpoints define the central confirmatory questions; secondary endpoints address additional questions and should be interpreted separately. |
| Randomization vs adjustment | Randomization creates the treatment comparison; adjustment accounts for specified variables in the statistical model. |
| Log-rank vs Cox model | The log-rank test evaluates evidence of differences between time-to-event distributions, while the Cox model estimates a hazard ratio and its uncertainty. |
21. Trial Timeline
Trial start
The NIAGARA phase 3 trial began on 2018-11-16 according to the ClinicalTrials.gov record.
Primary completion
The ClinicalTrials.gov record gives 2024-04-29 as the primary completion date.
Active, not recruiting
the ClinicalTrials.gov record lists the status as ACTIVE_NOT_RECRUITING.
22. What the Reported Results Do Not Establish
- The pCR odds ratio does not provide the absolute pCR rate in either treatment group.
- The EFS hazard ratio does not provide a median EFS or an absolute event-free probability at a specific time.
- The reported secondary HRs do not establish that all patients experienced the same proportional reduction in event hazard.
- The posted p-values do not establish an overall multiplicity-adjusted significance framework because no such procedure is reported in the ClinicalTrials.gov record.
- The reported serious adverse-event counts do not by themselves characterize the full safety profile.
- The ClinicalTrials.gov record does not support conclusions about treatment effects in individual demographic or clinical subgroups.
- The ClinicalTrials.gov record does not identify crossover, missing-data imputation, interim-analysis boundaries, Bayesian methods, or a non-inferiority margin.
23. Why This Trial Is a Useful Statistical Case Study
NIAGARA illustrates how a single randomized phase 3 trial can require several statistical tools because its endpoints have different structures. The pCR endpoint is binary and is analyzed with logistic regression. EFS and the other survival outcomes are time-to-event endpoints and are analyzed with log-rank and Cox-model methods. The analysis also demonstrates why stratification and covariate adjustment matter when treatment effects are reported as model-based estimates.
For statistical education, the trial is especially useful because it allows the reader to compare odds ratios and hazard ratios without treating them as interchangeable. It also shows why confidence intervals should accompany point estimates and why a p-value should never be used as a substitute for an effect estimate.
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Statistical Calculators
26. Sources
- ClinicalTrials.gov: NIAGARA (NCT03732677).
- Linked publication: PubMed record for PMID 39282910.
Continue through the Clinical Biostats statistical library
Use the related tutorials and calculators to explore the statistical methods represented in NIAGARA, including logistic regression, odds ratios, survival analysis, hazard ratios, confidence intervals and hypothesis testing.
27. Record Summary
NIAGARA provides a clear example of endpoint-specific statistical analysis within a randomized phase 3 clinical trial. The two primary endpoints use different statistical frameworks: pCR is a binary outcome analyzed with adjusted logistic regression and reported as an odds ratio of 1.60, while EFS is a time-to-event endpoint analyzed with a log-rank framework and a stratified Cox model, producing a reported hazard ratio of 0.68. Secondary analyses extend the time-to-event framework to EFS24, overall survival, metastasis-free survival and disease-specific survival, while cystectomy is analyzed using logistic regression.
The most useful statistical reading of the record is therefore not to reduce the trial to a single number. Instead, the analysis should keep the endpoint definition, analysis population, effect measure, confidence interval, p-value, and model assumptions connected to one another. The ClinicalTrials.gov record also contain discrepancies in certain confidence-interval fields, which illustrates why careful statistical reporting requires preserving the source information rather than silently reconstructing or rounding it.