A Study of Erdafitinib Compared With Vinflunine or Docetaxel or Pembrolizumab in Participants With Advanced Urothelial Cancer and Selected Fibroblast Growth Factor Receptor (FGFR) Gene Aberrations: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Study of Erdafitinib Compared With Vinflunine or Docetaxel or Pembrolizumab in Participants With Advanced Urothelial Cancer and Selected Fibroblast Growth Factor Receptor (FGFR) Gene Aberrations (NCT03390504), a randomized phase 3 study. The primary outcome is Overall Survival (OS).

2. Clinical Question

3. Trial Design

This was a phase 3, randomized, open-label, parallel-group study with an enrollment of 629. Participants were assigned to treatment groups within two cohorts defined by prior treatment history. Cohort 1 compared erdafitinib with vinflunine or docetaxel among participants previously treated with an anti-PD-(L)1 agent. Cohort 2 compared erdafitinib with pembrolizumab among participants whose prior treatment did not include an anti-PD-(L)1 agent. This cohort structure matters: the treatment comparisons address different clinical histories and should be interpreted separately.

The posted overall-survival groups included 143 participants in the Cohort 1 erdafitinib arm and 135 in its chemotherapy arm. Cohort 2 included 175 participants in the erdafitinib arm and 176 in the pembrolizumab arm. The primary analysis population was the intent-to-treat set: all randomized participants, analyzed according to the treatment to which they were randomized. The registry describes the primary endpoint as measured from randomization until death due to any cause, with a maximum follow-up timeframe of 51.7 months.

The study was not masked. The registry lists the FGFR inhibitor Clinical Trial Assay as a device used to determine molecular eligibility. Participants meeting molecular eligibility criteria could proceed to full study screening. The registry identifies the study as active, not recruiting. Details such as allocation concealment, randomization ratio, stratification factors, and the original sample-size justification are not reported in the registry record. The exact planned analysis procedures beyond those stated for the posted comparisons are also not reported in the registry record.

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.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Intent-to-Treat (ITT) analysis set included all randomized participants. Participants in this population were analyzed according to the treatment to which they were randomized.

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.

6. Primary Endpoint

The registry lists the primary outcome as Overall Survival (OS), measured over From randomization (3 days prior to Cycle 1 Day 1) until death due to any cause (maximum up to 51.7 months).

Overall survival is the time from randomization to death from any cause. It is a time-to-event endpoint: participants who have not died by the end of their observed follow-up do not contribute a death time, but their available follow-up still informs the analysis. The registry specifies the intent-to-treat population, so randomized participants are analyzed in their assigned groups rather than being reassigned according to treatment received.

General statistical explanation: overall survival is commonly used in clinical trials because it is a direct endpoint with an event that can be defined consistently as death from any cause. It does not require deciding whether a particular symptom or imaging change counts as disease progression. However, an overall-survival estimate depends on how much follow-up and how many events are available, and it may take time for enough events to occur to estimate treatment differences precisely.

The registry reports median overall survival in months for each arm, with 95% confidence intervals, as well as hazard ratios for the two within-cohort comparisons. A median describes the time by which half of the relevant group has experienced the event, under the survival analysis. A hazard ratio compares event rates over follow-up under the analysis model; it is not a difference in median survival. The registry timeframe allows follow-up up to 51.7 months, but the posted medians do not describe every participant's individual course.

7. Results Posted in the Registry

• Cohort 1: Arm 1A: Erdafitinib 8 mg/9 mg vs Cohort 1: Arm 1B: Chemotherapy (Vinflunine 320 mg/m^2 or Docetaxel 75 mg/m^2): hazard ratio 0.65 (95% CI 0.48–0.86), reported p = =0.0031 (as reported in the registry).
• Cohort 2: Arm 2A: Erdafitinib 8 mg/ 9 mg vs Cohort 2: Arm 2B: Pembrolizumab 200 mg: hazard ratio 1.16 (95% CI 0.92–1.48), reported p = =0.2121 (as reported in the registry).

8. Secondary Efficacy Endpoints

The registry also posts results for secondary outcomes. For each outcome below, the endpoint type, the test applied to the posted summary statistics, the result and its interpretation are given in turn. Secondary endpoints are usually not adjusted for multiple testing, so they support rather than confirm the primary finding.

Objective Response Rate (ORR) Per RECIST Version 1.1

• Endpoint type: Binary: the percentage of participants with the event (counts rebuilt from the posted percentages and group sizes).
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • Cohort 1: Arm 1B: Chemotherapy (Vinflunine 320 mg/m^2 or Docetaxel 75 mg/m^2): 12.6% (17/135) versus 46.9% (67/143) on control; risk difference -34.3 percentage points (95% CI -44.2 to -24.4), risk ratio 0.27 (0.17–0.43); unadjusted two-proportion z-test p = <0.001.
  • Cohort 2: Arm 2A: Erdafitinib 8 mg/ 9 mg: 40.0% (70/175) versus 46.9% (67/143) on control; risk difference -6.9 percentage points (95% CI -17.8 to 4.1), risk ratio 0.85 (0.66–1.10); unadjusted two-proportion z-test p = 0.220.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the proportion with the event is statistically lower for Cohort 1: Arm 1B: Chemotherapy (Vinflunine 320 mg/m^2 or Docetaxel 75 mg/m^2) than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Cohort 2: Arm 2A: Erdafitinib 8 mg/ 9 mg and the control. The interval gives the range of effects that remain compatible with the data.

Change From Baseline in Physical Functioning Scales of the Functional Assessment of Cancer Therapy - Bladder Cancer (FACT-Bl)

• Endpoint type: Continuous: a mean value per participant, compared as a difference in means.
• Test: Unadjusted two-sample z-test for the difference in means, using the posted standard errors.
• Result:
  • Cohort 1: Arm 1B: Chemotherapy (Vinflunine 320 mg/m^2 or Docetaxel 75 mg/m^2): mean -5.28 versus -3.70 on control; difference -1.58 (95% CI -8.68 to 5.52), standardized effect d = -0.05; z-test p = 0.663.
  • Cohort 2: Arm 2A: Erdafitinib 8 mg/ 9 mg: mean -4.56 versus -3.70 on control; difference -0.86 (95% CI -6.86 to 5.14), standardized effect d = -0.03; z-test p = 0.779.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Cohort 1: Arm 1B: Chemotherapy (Vinflunine 320 mg/m^2 or Docetaxel 75 mg/m^2) and the control. The interval gives the range of effects that remain compatible with the data.
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Cohort 2: Arm 2A: Erdafitinib 8 mg/ 9 mg and the control. The interval gives the range of effects that remain compatible with the data.

Change From Baseline in Utility Scale of the European Quality of Life-5 Dimensions-5 Levels Questionnaire (EQ-5D-5L)

• Endpoint type: Continuous: a mean value per participant, compared as a difference in means.
• Test: Unadjusted two-sample z-test for the difference in means, using the posted standard errors.
• Result:
  • Cohort 1: Arm 1B: Chemotherapy (Vinflunine 320 mg/m^2 or Docetaxel 75 mg/m^2): mean -0.06 versus -0.06 on control; difference 0.00 (95% CI -0.07 to 0.07), standardized effect d = 0.00; z-test p = 1.000.
  • Cohort 2: Arm 2A: Erdafitinib 8 mg/ 9 mg: mean -0.04 versus -0.06 on control; difference 0.02 (95% CI -0.04 to 0.08), standardized effect d = 0.07; z-test p = 0.539.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Cohort 1: Arm 1B: Chemotherapy (Vinflunine 320 mg/m^2 or Docetaxel 75 mg/m^2) and the control. The interval gives the range of effects that remain compatible with the data.
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Cohort 2: Arm 2A: Erdafitinib 8 mg/ 9 mg and the control. The interval gives the range of effects that remain compatible with the data.

The registry names secondary outcomes spanning tumor response and progression, patient-reported functioning and symptoms, quality of life, and duration of response. These outcomes complement overall survival by describing different aspects of disease course and participant experience. They do not measure the same construct as death from any cause, so they can add context but cannot replace the primary endpoint.

Numerical results are posted for objective response rate, change in the FACT-Bl physical-functioning scale, and change in the EQ-5D-5L utility scale. For the response outcome, the arm-level estimates show a different pattern across the two comparisons than a single uniform effect across both cohorts. The physical-functioning and utility outcomes are presented as least-squares means with standard errors. The registry does not provide sponsor-reported analyses for the comparisons attached to these posted secondary summaries. The available recalculations are unadjusted and should be read as descriptive rather than as confirmatory evidence.

The remaining named secondary outcomes include progression-free survival, time until symptom deterioration, PGIS shifts, the EQ-5D-5L visual analog scale, and duration of response. Their numerical results are not reported in the registry record. That absence means their results cannot be used to assess whether they align with, qualify, or differ from the primary overall-survival findings.

General statistical explanation: testing many endpoints or making multiple comparisons increases the chance that at least one result appears statistically notable by chance. A multiplicity procedure can control that risk, but the registry does not report whether an adjustment was used for the secondary outcomes or how the analyses were prioritized. Without those details, unadjusted secondary p-values carry less inferential weight than a prespecified primary analysis. The secondary outcomes can suggest areas for further study, but the posted summaries alone do not establish a comprehensive pattern across all outcomes.

Secondary outcomes registered for this trial, for reference (only those above were recalculated):

• Progression Free Survival (PFS) as Per Response Evaluation Criteria in Solid Tumors (RECIST) Version (v) 1.1
• Objective Response Rate (ORR) Per RECIST Version 1.1
• Change From Baseline in Physical Functioning Scales of the Functional Assessment of Cancer Therapy - Bladder Cancer (FACT-Bl)
• Time Until Symptom Deterioration
• Percentage of Participants With Shift From Baseline in Patient-Global Impression of Severity (PGIS)
• Change From Baseline in Utility Scale of the European Quality of Life-5 Dimensions-5 Levels Questionnaire (EQ-5D-5L)
• Change From Baseline in Visual Analog Scale (VAS) of the European Quality of Life-5 Dimensions-5 Levels Questionnaire (EQ-5D-5L)
• Duration of Response (DOR) as Per RECIST Version 1.1

9. Statistical Interpretation

10. Statistical Methods Explained

Overall survival is analyzed as time-to-event data. The time origin in this study was randomization, and the event was death due to any cause. The registry describes the primary analysis population as intent-to-treat: all randomized participants, kept in the group to which they were assigned. This approach preserves the randomized comparison and includes participants according to assignment even if their treatment course later differs.

What does the median describe?

General statistical explanation: a median survival time is the point at which the estimated survival curve reaches half of its starting level. It summarizes the middle of the estimated time-to-event distribution, rather than an average time for every participant. A median does not express the full shape of the survival experience. Two groups can have similar medians but different patterns at other times, or different medians while their uncertainty remains substantial.

The registry reports medians and 95% confidence intervals separately by arm. The Cohort 1 medians were 12.06 months for erdafitinib and 8.74 months for chemotherapy. The Cohort 2 medians were 10.91 months for erdafitinib and 11.07 months for pembrolizumab. These arm-level summaries are useful descriptive measures, but the reported hazard ratios are the sponsor's inferential comparisons for the primary endpoint. A median comparison is not a substitute for the hazard-ratio analysis.

How should a hazard ratio be read?

General statistical explanation: a hazard is the instantaneous event rate among participants who have not yet experienced the event, conditional on the analysis framework. A hazard ratio compares that rate between groups. A value of 1 corresponds to equal hazards under the comparison; a value below 1 indicates a lower estimated hazard in the numerator group, and a value above 1 indicates a higher estimated hazard in that group. A hazard ratio is not a ratio of median survival times and should not be described as a proportional change in the probability of death over the entire follow-up period.

For Cohort 1, the reported hazard ratio was 0.65, with a 95% confidence interval from 0.48 to 0.86. For Cohort 2, it was 1.16, with a 95% confidence interval from 0.92 to 1.48. The comparisons are ordered as erdafitinib versus the cohort-specific comparator. The first interval does not include 1, whereas the second does. The registry lists a log-rank method for Cohort 1 and a stratified log-rank test for Cohort 2. Details of the stratification variables and any model diagnostics are not reported in the registry record.

What do the tests and p-values add?

General statistical explanation: a log-rank test compares event-time patterns between groups while accounting for the timing of events and censoring. Censoring occurs when a participant's event time is not observed during the available follow-up; the participant's observed time still contributes information up to that point. A p-value measures how inconsistent the observed data are with a specified null hypothesis under the test's assumptions. It is not the probability that the null hypothesis is true, nor does it measure clinical importance.

The posted p-values are =0.003 for Cohort 1 and =0.212 for Cohort 2. The registry identifies the test methods and the hazard-ratio estimates, but the full analysis plan, details of censoring rules, handling of missing data, assumptions checks, and any interim analyses are not reported in the registry record. Therefore, the summaries support interpretation of the listed analyses, not reconstruction of every analytical decision.

Why can secondary recalculations differ from trial analyses?

General statistical explanation: a simple recalculation from arm-level summaries may use an unadjusted comparison and omit stratification, covariates, repeated-measure structure, or other features of the trial's analysis. Such a calculation can be useful for orientation, but it does not reproduce an analysis that uses participant-level data or a prespecified model. The primary overall-survival hazard ratios presented for this study are the sponsor-reported estimates; no unadjusted hazard-ratio recalculation is used for those comparisons. For secondary outcomes, recalculated comparisons are identified as unadjusted and should not be confused with a sponsor-reported confirmatory analysis.

11. Confidence Intervals and Precision

The 95% confidence intervals communicate uncertainty around the estimates under the analysis method. General statistical explanation: if the same design and analysis were repeated many times under the model assumptions, confidence intervals constructed in this way would cover the underlying quantity at the stated rate over repeated studies. A confidence interval is not a range that has a 95% probability of containing the fixed true value after the data have been observed.

For Cohort 1, the median overall-survival interval was 10.25 to 16.36 months in the erdafitinib group and 6.67 to 10.71 months in the chemotherapy group. The hazard-ratio interval was 0.48 to 0.86. Its limits are both below 1, which is consistent with the reported test result of p =0.003. For Cohort 2, the median intervals were 9.23 to 12.58 months for erdafitinib and 9.72 to 13.63 months for pembrolizumab. The hazard-ratio interval was 0.92 to 1.48 and includes 1; the reported p-value was =0.212. These interval patterns describe statistical precision, not certainty about individual outcomes.

The width of an interval reflects the amount of information and the variability represented under the analysis. It does not address every source of uncertainty, including whether a trial population applies to other settings, whether all analysis assumptions are adequate, or how multiple endpoints were handled. The registry does not report a separate interval for each recalculated secondary comparison, and such unadjusted calculations should not be read as having the same analytical status as the sponsor-reported primary results.

12. Planning a Similar Trial: Sample Size

This is a planning illustration, not a re-evaluation of the original trial. It asks: if a new study wanted to detect a difference like the one observed, or a smaller one, how large would it need to be? It is deliberately not “observed power” of the finished trial, which is not a meaningful quantity.

The planning illustrations translate observed effects into hypothetical sample-size requirements under stated assumptions; they are not the study's original design calculations. For the time-to-event comparisons, the figures are total numbers of events, not participants per arm. For other comparisons, the figures are participants per arm. The planning assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. Each scenario targets the stated fraction of the observed effect. The original trial's sample-size assumptions are not reported in the registry record.

The 80% power column means the hypothetical design would have an 80% probability of detecting the specified effect at the stated significance level if the assumptions and effect were correct. The 90% power column makes the corresponding probability 90%. These describe design targets, not a guarantee that a study will detect an effect or that its result will be statistically significant. The tables give total events for the overall-survival comparisons using the Schoenfeld approach. For example, the Cohort 1 planning illustration lists 170 total events for 80% power and 227 for 90% power when planning for 100% of the observed effect. The Cohort 2 illustration lists 1,426 and 1,908 total events for those same targets and effect fraction.

Planning for a smaller effect requires more information to detect it reliably under the same assumptions. Accordingly, figures for 75% and 50% of the observed effect are larger than those for the full observed effect in the posted illustrations. This is not observed power: it does not describe the probability that the completed trial's result was correct, nor does it measure the certainty of the observed estimate. It is a forward-looking planning calculation based on the stated assumptions and an effect anchored to the observed result.

13. How the Sample Size Is Derived

Power is the probability that a study of a given size rejects the null hypothesis when a specific true effect exists. It depends on four things: the effect to detect, the variability of the outcome, the significance level (here a two-sided α of 0.05, so z_1−α/2 = 1.960) and the desired power (z_1−β = 0.842 for 80% and 1.282 for 90%).

Formula

d is the total number of events needed (Schoenfeld's formula) with 1:1 allocation and proportional hazards.

Worked calculation for this trial

1. Hazard ratio HR = 0.650, so ln(HR) = -0.4308.
2. d = 4 × (1.96 + 0.842)² / [ln(0.650)]² = 169.2, rounded up to 170 events in total for 80% power.

The smaller-effect rows in the table above repeat the same calculation with the effect multiplied by 0.75 and 0.5 (for ratios, applied to the logarithm of the ratio).

R code

alpha <- 0.05
z_a   <- qnorm(1 - alpha / 2)          # two-sided
hr <- 0.65                              # observed hazard ratio
events_needed <- function(power, frac = 1) ceiling(round(4 * (z_a + qnorm(power))^2 / log(hr^frac)^2, 6))   # Schoenfeld, 1:1 allocation
for (f in c(1, 0.75, 0.5))
  cat(sprintf("%3.0f%% of observed effect: events = %s (80%% power), %s (90%% power)\n", 100 * f,
              format(events_needed(0.80, f), big.mark = ","), format(events_needed(0.90, f), big.mark = ",")))

# Simulation: given d total events under proportional hazards, events in the treatment arm are Binomial(d, hr / (1 + hr)).
# The score (log-rank type) statistic is (D1 - d / 2) / sqrt(d / 4).
set.seed(3390504)
d    <- events_needed(0.80)
reps <- 20000
D1 <- rbinom(reps, d, hr / (1 + hr))
z  <- (D1 - d / 2) / sqrt(d / 4)
cat(sprintf("Simulated power with %s events: %.3f (%d replicates)\n", format(d, big.mark = ","), mean(abs(z) > z_a), reps))

Output from running the code

100% of observed effect: events = 170 (80% power), 227 (90% power)
 75% of observed effect: events = 301 (80% power), 403 (90% power)
 50% of observed effect: events = 677 (80% power), 906 (90% power)
Simulated power with 170 events: 0.809 (20000 replicates)

The simulation uses the fact that, with d events in total and proportional hazards, the number of events in the treatment arm is binomial. It is an approximation that conditions on the total number of events, and it should land close to the target power.

14. Group-Sequential Design Analysis

Many time-to-event trials, especially in oncology, stop or report early when interim data are overwhelming, so a single fixed-sample number understates the real design problem. The calculation below is an illustrative group-sequential design built from this trial's posted control median and hazard ratio. The registry record does not state the planned number of looks or the spending function, so the design is a planning illustration, not a reconstruction of the protocol. Assumed: one-sided α = 0.025, 90% power, three equally spaced looks on information, Lan-DeMets O'Brien-Fleming spending, 24 months of accrual, 60 months in total and a 1% monthly dropout hazard.

Under these assumptions the design needs about 274 participants in total, with the boundaries below.

Fixed design versus group-sequential alternatives

Group-sequential designs cost a small increase in maximum sample size and events compared with the fixed design, in exchange for the chance to stop early for overwhelming efficacy. O'Brien-Fleming spending keeps early boundaries very conservative and the final boundary close to the fixed-design value, whereas Pocock spending spends more alpha early.

Sensitivity to a smaller true effect

Planning for exactly the observed effect is optimistic, because observed effects in published trials tend to be somewhat larger than the true effect. The table repeats the main design (three looks, O'Brien-Fleming spending, 90% power) for smaller true hazard ratios.

For context, this trial enrolled 629 participants.

Cross-check, upper boundaries: gsDesign 3.71, 2.511, 1.993 | rpact 3.71, 2.511, 1.993. Agreement between the two packages confirms the boundary calculation.

R code and output

suppressWarnings(suppressMessages({ library(gsDesign); library(rpact) }))
med_control <- 8.74          # control-arm median, months (converted from the posted unit)
hr          <- 0.65          # hazard ratio used for planning
lambdaC     <- log(2) / med_control
eta         <- 0.01               # dropout hazard per month (assumption)
acc_months  <- 24                 # accrual duration (assumption)
study_months<- 60                 # total study duration (assumption)

design <- function(k, sfu, beta, h = hr) gsSurv(k = k, test.type = 1, alpha = 0.025, beta = beta, sfu = sfu,
  lambdaC = lambdaC, hr = h, hr0 = 1, eta = eta, gamma = 1, R = acc_months, T = study_months, minfup = study_months - acc_months)

# Main design: 3 looks, Lan-DeMets O'Brien-Fleming spending, one-sided alpha 0.025, 90% power
x <- design(3, sfLDOF, 0.10)
cat(sprintf("N_TOTAL|%s\n", format(ceiling(sum(x$gamma * x$R)), big.mark = ",")))
for (i in 1:3) cat(sprintf("GS|%d|%.1f|%.0f|%.3f|%.5f|%.5f\n", i, x$T[i], x$n.I[i], x$upper$bound[i], pnorm(-x$upper$bound[i]), cumsum(x$upper$spend)[i]))
cat("\n"); print(gsBoundSummary(x))

# Comparison of designs
cmp <- list(list("Fixed design, one analysis", 1, sfLDOF), list("Two looks, O'Brien-Fleming", 2, sfLDOF),
            list("Three looks, O'Brien-Fleming", 3, sfLDOF), list("Three looks, Pocock", 3, sfLDPocock))
cat("\n")
for (b in c(0.20, 0.10)) for (d in cmp) {
  y <- design(d[[2]], d[[3]], b)
  cat(sprintf("DES|%s|%d|%.0f|%s|%.1f\n", d[[1]], round(100 * (1 - b)), tail(y$n.I, 1), format(ceiling(sum(y$gamma * y$R)), big.mark = ","), tail(y$T, 1)))
}

# Sensitivity: what if the true effect is smaller than the observed one? (3 looks, O'Brien-Fleming, 90% power)
cat("\n")
for (fr in c(1, 0.75, 0.5)) {
  y <- design(3, sfLDOF, 0.10, exp(log(hr) * fr))
  cat(sprintf("SENS|%.2f|%.3f|%.0f|%s\n", fr, exp(log(hr) * fr), tail(y$n.I, 1), format(ceiling(sum(y$gamma * y$R)), big.mark = ",")))
}

# Independent cross-check of the boundaries with rpact
r <- getDesignGroupSequential(kMax = 3, alpha = 0.025, beta = 0.10, sided = 1, typeOfDesign = "asOF", informationRates = x$timing)
cat(sprintf("\nCross-check, upper boundaries: gsDesign %s | rpact %s\n", paste(round(x$upper$bound, 3), collapse = ", "), paste(round(r$criticalValues, 3), collapse = ", ")))

Method: LachinFoulkes 
    Analysis               Value Efficacy
   IA 1: 33%                   Z   3.7103
      N: 200         p (1-sided)   0.0001
  Events: 76        ~HR at bound   0.4268
   Month: 17    P(Cross) if HR=1   0.0001
             P(Cross) if HR=0.65   0.0338
   IA 2: 67%                   Z   2.5114
      N: 274         p (1-sided)   0.0060
 Events: 152        ~HR at bound   0.6653
   Month: 28    P(Cross) if HR=1   0.0060
             P(Cross) if HR=0.65   0.5603
       Final                   Z   1.9930
      N: 274         p (1-sided)   0.0231
 Events: 228        ~HR at bound   0.7679
   Month: 60    P(Cross) if HR=1   0.0250
             P(Cross) if HR=0.65   0.9000



Cross-check, upper boundaries: gsDesign 3.71, 2.511, 1.993 | rpact 3.71, 2.511, 1.993

15. Analyses Reported by the Sponsor

The primary endpoint comparisons are reported by the sponsor as hazard ratios from the named time-to-event tests. For Cohort 1, the reported hazard ratio was 0.65, with a 95% confidence interval from 0.48 to 0.86 and p =0.003. For Cohort 2, it was 1.16, with a 95% confidence interval from 0.92 to 1.48 and p =0.212. These are the relevant primary analyses in the registry record, rather than recalculated hazard ratios based on the arm-level medians.

Simple unadjusted recalculations are available for some posted secondary outcomes, including objective response rate and selected patient-reported measures. They are derived from the arm-level summaries and are not a substitute for a sponsor's participant-level analysis. The registry does not list sponsor-reported analyses for those secondary comparisons. The recalculated p-values therefore provide a descriptive check, not the trial's prespecified inferential conclusions. The record marks no discrepancy between the posted primary analyses and any recalculation, and no recalculation is used in place of the sponsor-reported overall-survival estimates.

The registry does not report enough detail to reproduce the primary analyses independently, including the full analysis plan, test assumptions, stratification details, and handling of every data issue. Interpretation should therefore preserve the distinction between sponsor-reported primary results and exploratory, unadjusted calculations from summary statistics.

16. Multiplicity, Missing Data and Analysis Populations

The primary endpoint is overall survival, and the registry reports two cohort-specific treatment comparisons. It does not report a multiplicity strategy for controlling error across those comparisons and the multiple secondary outcomes. The presence of two cohorts and numerous secondary measures makes the interpretation of any individual p-value dependent on whether analyses were prespecified, prioritized, or adjusted. Those details are not reported in the registry record, so the posted p-values should be understood as applying to the reported tests rather than as evidence that all possible comparisons were controlled together.

The intent-to-treat population is explicitly described for the primary endpoint: all randomized participants are analyzed according to randomized assignment. The registry lists the arm-level sample sizes for the primary and posted secondary outcomes, but the detailed analysis populations for each secondary endpoint are not reported in the registry record. The record also does not specify how missing measurements were handled for the patient-reported outcomes, whether imputation was used, or how participants with incomplete follow-up contributed to each secondary analysis.

For overall survival, participants may have follow-up that ends before death is observed; general statistical methods account for such censoring, but the registry does not provide the specific censoring rules or sensitivity analyses. The handling of withdrawals, treatment discontinuation, and other missing data is not reported in the registry record. The full statistical analysis plan, interim-analysis procedures, stopping rules, and any adjustments for repeated looks at the data are also not reported in the registry record.

These omissions do not invalidate the posted results, but they limit independent assessment of the analysis choices and the strength of conclusions across endpoints. The sponsor-reported primary results can be interpreted as the analyses identified in the registry, while secondary recalculations from summary data remain unadjusted. A reader should avoid treating nominal significance on a secondary measure as if it were an independently controlled confirmatory finding.

17. 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: febrile neutropenia (6.8%), urinary tract infection (5.2%), haematuria (3.5%), intestinal obstruction (3.5%), acute kidney injury (3.5%).

The registry's safety summary lists participants at risk and participants affected by serious adverse events and by other adverse events. In Cohort 1, the serious-event summary lists 63 affected among 142 at risk for erdafitinib and 50 among 117 at risk for chemotherapy. In Cohort 2, it lists 69 affected among 173 at risk for erdafitinib and 81 among 173 at risk for pembrolizumab. The other-adverse-event figures are 141 among 142, 104 among 117, 173 among 173, and 157 among 173, respectively. These are counts in the registry summary; they should not be treated as adjusted comparisons of risk.

The listed serious adverse-event terms include febrile neutropenia at 6.8%, urinary tract infection at 5.2%, haematuria at 3.5%, intestinal obstruction at 3.5%, acute kidney injury at 3.5%, and febrile bone marrow aplasia at 3.4%. The registry also lists death counts in the safety summary: 91, 90, 141, and 131 across the four groups in the order above. These counts are not a substitute for the time-to-event overall-survival analysis.

Safety counts alone do not show causation, severity for each participant, duration, or comparative risk adjusted for follow-up. Detailed definitions, exposure-adjusted rates, and further safety analyses are not reported in the registry record.

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

The results describe two randomized comparisons in populations defined by prior treatment history. In Cohort 1, the sponsor-reported overall-survival hazard ratio was below 1 and its 95% confidence interval did not include 1. In Cohort 2, the estimate was above 1, but its confidence interval included 1. These findings indicate different degrees of statistical evidence in the two posted comparisons; they should not be collapsed into a single answer about erdafitinib across all participants.

The Cohort 1 result applies to the enrolled comparison between erdafitinib and vinflunine or docetaxel among participants with prior anti-PD-(L)1 treatment. The Cohort 2 result applies to erdafitinib versus pembrolizumab among participants whose prior treatment did not include an anti-PD-(L)1 agent. Because both the treatment history and comparator differ, cross-cohort contrasts cannot establish that the effect changes by prior treatment. The record does not report a formal interaction analysis to test such a difference.

A hazard ratio summarizes relative event rates under an analysis model; it is not an individual prediction, an absolute probability, or a direct measure of how many months a particular participant may live. Median survival is a group summary, and its confidence interval describes uncertainty around that summary under the analysis. Neither measure captures all clinical considerations or substitutes for the full set of benefits, risks, and patient circumstances.

The secondary outcomes offer additional perspectives, but the numerical results are incomplete and the available secondary recalculations are unadjusted. The registry does not report enough information about multiplicity, missing data, or the full statistical plan to support broader claims from those measures. No treatment recommendation follows from these statistical summaries alone.

19. Why This Trial Matters Statistically

This study illustrates why trial results need to be read within their comparison structure. Although four treatment arms appear in the registry, they form two separate randomized comparisons based on different prior-treatment histories. Treating all arms as one comparison would ignore the design and could blur distinct clinical questions. The Cohort 1 and Cohort 2 hazard ratios also have different interval patterns, which reinforces the need to report each comparison on its own terms.

The registry makes it possible to distinguish the sponsor-reported primary analyses from calculations based only on arm-level summaries. The primary endpoint uses time-to-event methods and reports hazard ratios with confidence intervals. Some secondary outcomes permit simple unadjusted recalculation, but such calculations cannot reproduce analyses that may use participant-level information or prespecified modeling. This distinction is central to transparent statistical interpretation.

The record also demonstrates that significance and precision are related but not interchangeable. A confidence interval can show the range of estimates compatible with the analysis while leaving uncertainty about the magnitude and generalizability of the result. Planning illustrations further show how the amount of information needed depends on the effect targeted, but those figures are not a measure of the completed study's observed power or a record of its original design rationale.

Finally, the registry identifies important limits on interpretation: it reports a primary endpoint and some secondary outcomes but does not give the full multiplicity plan, missing-data procedures, or detailed analysis plan. Recognizing what is reported and what remains unspecified is part of sound biostatistical reading, particularly in a multi-cohort, multi-endpoint trial.

20. 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.

The main limitation is that the registry record provides headline results but not the full protocol or statistical analysis plan. It does not report the original sample-size justification, randomization ratio, allocation-concealment details, stratification variables, interim analyses, or stopping rules. The reported methods identify log-rank and stratified log-rank procedures, but the details needed to reproduce or fully audit those analyses are not reported in the registry record.

The study has two distinct cohorts and different comparators. Cohort 1 compares erdafitinib with a chemotherapy arm that includes vinflunine or docetaxel, while Cohort 2 compares erdafitinib with pembrolizumab. Differences between the cohort results cannot be attributed to treatment history or comparator without a formal analysis of interaction, which is not reported in the registry record.

Numerical secondary results are available for only some of the listed outcomes in the record. Several named outcomes have no numerical result reported in the registry record. The available simple secondary comparisons are unadjusted, and a multiplicity procedure is not reported. Secondary findings therefore should not be treated as equivalent to the primary sponsor analyses.

Missing-data rules and analysis populations for secondary measures are not reported in the registry record. This is particularly relevant to patient-reported outcomes, which can be affected by incomplete assessments over follow-up. The safety summary gives affected-participant counts and selected serious-event terms, but does not provide enough detail to assess exposure-adjusted rates, timing, duration, or attribution. Finally, the posted estimates reflect the enrolled population and follow-up represented in the registry; the record alone does not establish how well they generalize beyond that population.

21. Related Tutorials

22. Related Statistical Calculators

23. Sources

• ClinicalTrials.gov: A Study of Erdafitinib Compared With Vinflunine or Docetaxel or Pembrolizumab in Participants With Advanced Urothelial Cancer and Selected Fibroblast Growth Factor Receptor (FGFR) Gene Aberrations — NCT03390504.
• Registry record last updated 2026-11-03; status last verified 2026-11.
• Results first posted 2025-05-14.
• 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.

24. Record Summary

NCT03390504 is a randomized, open-label, parallel-group phase 3 study of erdafitinib in advanced urothelial cancer with selected FGFR gene aberrations. The two cohorts compare erdafitinib with chemotherapy or pembrolizumab according to participants' prior treatment history. The primary endpoint is overall survival, analyzed in the intent-to-treat population.

The sponsor-reported hazard ratio was 0.65 (95% CI 0.48 to 0.86; p =0.003) for erdafitinib versus chemotherapy in Cohort 1, and 1.16 (95% CI 0.92 to 1.48; p =0.212) for erdafitinib versus pembrolizumab in Cohort 2. Selected secondary and safety summaries are also posted, but several analysis details are not reported in the registry record.