A Study of Tislelizumab (BGB-A317) Versus Chemotherapy as Second Line Treatment in Participants With Advanced Esophageal Squamous Cell Carcinoma: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Study of Tislelizumab (BGB-A317) Versus Chemotherapy as Second Line Treatment in Participants With Advanced Esophageal Squamous Cell Carcinoma (NCT03430843), a randomized phase 3 study. The primary outcome is Overall Survival (OS) in the Intent-to-Treat (ITT) Analysis Set.

2. Clinical Question

The trial addressed a second-line treatment question in advanced esophageal squamous cell carcinoma. Its comparison was between a named drug and a chemotherapy group whose specific regimen was selected by the investigator. That makes the comparator clinically broad: the results describe the chemotherapy strategy as used in this trial, rather than a single chemotherapy drug in isolation.

The primary question is about time until death from any cause, not tumor response alone. The trial also included response, progression-free survival, duration of response, and health-related quality-of-life outcomes among its secondary outcomes. The registry record lists outcomes in the overall intent-to-treat population and in PD-L1-positive analysis sets, but the posted primary result is overall survival in the intent-to-treat set.

The clinical question is comparative, but interpretation depends on more than the point estimates. The study was open-label, and the registry record does not report all details needed to assess the full analysis plan, missing-data handling, or multiplicity procedures. These gaps matter particularly when interpreting secondary findings alongside the primary result.

3. Trial Design

This was a randomized, controlled, open-label, global Phase 3 study with a parallel-group design. The registry record lists an enrollment of 512 and a completed status. The study began on 2018-01-26, and the primary completion date was 2020-12-01. The stated purpose was to evaluate the efficacy and safety of tislelizumab as second-line treatment for advanced unresectable or metastatic esophageal squamous cell carcinoma that had progressed during or after first-line therapy.

Participants were assigned to tislelizumab or investigator-chosen chemotherapy. The tislelizumab arm received the drug on Day 1 every 21 days. The chemotherapy arm could receive paclitaxel on Day 1 every 21 days or on a weekly schedule, docetaxel on Day 1 every 21 days, or irinotecan on Days 1 and 8 every 21 days. As the control treatment allowed several regimens, the comparison is with the investigator-selected chemotherapy approach rather than with one uniform drug regimen.

The trial had no masking: the registry record identifies it as open-label. This means treatment assignment was not concealed from participants and study personnel, although the record excerpt does not report whether any outcome assessments used separate blinded review. Further details about allocation procedures and safeguards against bias are not reported in the registry record.

The primary endpoint was overall survival in the intent-to-treat analysis set, which included all randomized participants. Secondary outcomes included overall survival in a PD-L1-positive set, objective response rate in overall and PD-L1-positive sets, progression-free survival, duration of response, and health-related quality of life. The secondary outcomes were assessed on different populations and timeframes, so they should not be treated as interchangeable measures of the primary question.

The registry record does not report the original sample-size justification, the planned number or timing of interim analyses, or the complete statistical analysis plan. Those details cannot be inferred from the final enrollment or posted results.

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: The ITT analysis set included all randomized participants

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) in the Intent-to-Treat (ITT) Analysis Set, measured over Approximately 2 years and 10 months from date of first randomization.

Overall survival is the length of time from randomization until death from any cause. In this trial, the endpoint was assessed in the intent-to-treat analysis set, which the registry record defines as all randomized participants. Death from any cause is an objective event, and the time-to-event framework uses information about both when deaths occurred and how long participants were followed.

The reported median is the time by which the estimated survival curve reaches the point where half the group is estimated to have experienced the event. It is a descriptive summary of the survival distribution; it does not represent the survival time of every participant. The registry record reports a median for each treatment group with a 95% confidence interval.

A hazard ratio compares the event rates over follow-up, conditional on participants remaining at risk up to a given time. A value below one for tislelizumab relative to chemotherapy indicates a lower estimated hazard in the tislelizumab group under the reported comparison. The hazard ratio is not a difference in median survival, and it does not directly state how long an individual participant will live.

Overall survival is used because it captures the time to death from any cause and is directly relevant to the trial's stated treatment question. Its interpretation still depends on follow-up, censoring, and the analysis methods. Details of censoring procedures and other specific survival-analysis choices are not reported in the registry record.

7. Results Posted in the Registry

• Tislelizumab vs Investigator Chosen Chemotherapy: hazard ratio 0.70 (95% CI 0.57–0.85), reported p < 0.001 (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) in the ITT Analysis Set

• 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:
  • Investigator Chosen Chemotherapy: 9.8% (25/256) versus 20.3% (52/256) on control; risk difference -10.5 percentage points (95% CI -16.7 to -4.4), risk ratio 0.48 (0.31–0.75); unadjusted two-proportion z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the proportion with the event is statistically lower for Investigator Chosen Chemotherapy than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

Overall Response Rate (ORR) in the PD-L1 Positive Analysis Sets

• 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:
  • Investigator Chosen Chemotherapy: 11.3% (7/62) versus 26.2% (21/80) on control; risk difference -15.0 percentage points (95% CI -27.4 to -2.5), risk ratio 0.43 (0.20–0.95); unadjusted two-proportion z-test p = 0.026.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the proportion with the event is statistically lower for Investigator Chosen Chemotherapy than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

The secondary outcomes broaden the trial's description beyond overall survival. They include response, progression-free survival, duration of response, and health-related quality of life, with some outcomes assessed in the intent-to-treat set and others also assessed in PD-L1-positive sets. Together, these outcomes address different aspects of disease course and patient-reported status, but they do not measure the same thing as overall survival.

The posted response estimates are directionally consistent across the overall and PD-L1-positive populations, with higher response percentages in the tislelizumab group. These are secondary results and should be interpreted in the context of the primary survival finding, not as separate substitutes for it. The registry record does not list sponsor-reported analyses for the response outcomes; the available contrast statistics are unadjusted recalculations.

The registry record lists 14 secondary outcomes. Testing multiple outcomes and analysis populations creates multiple opportunities for apparently notable results to arise by chance. Unless a multiplicity strategy is specified and applied, unadjusted p-values do not control the overall false-positive risk across the family of tests. The multiplicity plan is not reported in the registry record, so the secondary p-values should carry less inferential weight than the reported primary analysis.

Some secondary endpoints may also depend on definitions and assessment procedures that are not detailed in the registry record. For example, response and progression outcomes can be influenced by assessment timing and rules, while quality-of-life outcomes may be affected by missing questionnaires. The posted information does not provide enough detail to evaluate those issues fully. The secondary results are therefore best treated as context for the primary outcome rather than a comprehensive or confirmatory set of independent claims.

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

• Overall Survival (OS) in the PDL-1 Positive Analysis Set
• Objective Response Rate (ORR) in the ITT Analysis Set
• Overall Response Rate (ORR) in the PD-L1 Positive Analysis Sets
• Progression-free Survival (PFS) in the ITT Analysis Set
• Progression-free Survival (PFS) in the PDL-1 Positive Analysis Set
• Duration of Response (DOR) in the ITT Analysis Set
• Duration of Response (DOR) in the PDL-1 Positive Analysis Set.
• Health-Related Quality of Life (HRQoL) as Assessed by European Organization for Research and Treatment of Cancer Quality of Life Questionnaire-Core 30 (EORTC QLQ-C-30) in the ITT Analysis Set

9. Statistical Interpretation

10. Statistical Methods Explained

The primary endpoint is time-to-event data, because it records the duration from randomization until death. Unlike a simple count of events by a fixed date, survival analysis uses the timing of events and can retain participants' follow-up information even when their event time is not observed during the study. Participants without a recorded death by the end of their observed follow-up are generally censored, but the specific censoring rules are not reported in the registry record.

What does the median survival describe?

The median is the time at which the estimated survival probability reaches the halfway point. It is a compact summary of the survival curve, not an average and not an estimate that every participant will experience. The posted medians are 8.6 months for tislelizumab and 6.3 months for investigator-chosen chemotherapy. Each is accompanied by a 95% confidence interval, which communicates uncertainty about the median estimate.

What does the hazard ratio describe?

The registry record reports a hazard ratio for tislelizumab versus investigator-chosen chemotherapy. In general, a hazard is the instantaneous event rate among participants who have not yet experienced the event and remain under observation. A hazard ratio compares those rates between groups over follow-up. It is not a risk ratio, a ratio of the medians, or a direct measure of the probability that a particular participant will die.

The reported hazard ratio is 0.70, with a 95% confidence interval from 0.57 to 0.85. Because the estimate is below one, the reported direction is consistent with a lower hazard in the tislelizumab group. The registry record reports a one-sided log-rank test and p< 0.001. The log-rank test compares the event-time patterns between groups over follow-up. A p-value describes how compatible the observed data are with a specified null model under the test assumptions; it is not the probability that the null hypothesis is true, nor a measure of the size or clinical importance of an effect.

The registry record does not report the full details of the survival model, including whether a proportional-hazards assumption was assessed or whether covariate adjustment was used. In general, a hazard ratio is most straightforward to interpret as a constant relative comparison when hazards are proportional over time. If the relative event rate changes over follow-up, one overall hazard ratio can average over patterns that differ by time. This is a general consideration, not a claim that such a change occurred in this trial.

What is an unadjusted recalculation?

For the secondary response outcomes, the posted group percentages are accompanied by simple unadjusted recalculations based on the posted summary statistics. Such recalculations compare the observed response proportions without reproducing any sponsor-specific adjustment or model. They can help describe the size and uncertainty of the observed contrast, but they do not replace a prespecified sponsor analysis. The registry record does not list sponsor-reported analyses for these secondary outcomes.

The intent-to-treat population includes all randomized participants, as defined in the record. This preserves the randomized group assignment in the primary comparison. Other outcomes use narrower populations, including a PD-L1-positive analysis set. Definitions and handling of participants with unavailable biomarker status are not reported in the registry record. Differences in analysis populations can affect how directly outcomes can be compared with the primary result.

Confidence intervals are discussed further in the precision interpretation: in general, a confidence interval expresses the range of estimates compatible with the data and the method used, rather than a range containing every participant's individual outcome. Its width reflects statistical precision, while interpretation also depends on study conduct and analysis assumptions.

The planning illustrations use the observed effect as a reference and consider stated fractions of that effect. For binary response outcomes, the illustrations are expressed as participants per arm; for time-to-event outcomes, they are expressed as total events. These are hypothetical planning calculations under the listed assumptions, not reconstructions of the trial's original design. The trial's own original sample-size assumptions are not reported in the registry record.

Over-dispersion is a general statistical issue that arises when observed variability exceeds what a simple model assumes. In binary outcomes, for example, clustering or unmodeled heterogeneity can make the data more variable than a basic independent-observation model anticipates. If over-dispersion is present but ignored, standard errors may be too small and confidence intervals may be too narrow. The registry record does not report an over-dispersion assessment for the posted outcomes, so this is a general methodological point rather than a finding about this study.

11. Confidence Intervals and Precision

The primary median estimates are 8.6 months for tislelizumab, with a 95% confidence interval from 7.5 to 10.4 months, and 6.3 months for investigator-chosen chemotherapy, with a 95% confidence interval from 5.3 to 7.0 months. These intervals express uncertainty in the estimated medians. They are not ranges for individual survival times and do not mean that the true median has a 95% probability of lying within the particular interval after the data are observed. In general, a confidence interval is produced by a method with stated long-run coverage properties under its assumptions.

The hazard ratio estimate of 0.70 has a 95% confidence interval from 0.57 to 0.85. The interval gives a range of values compatible with the data under the analysis method and is informative about precision as well as direction. It does not remove uncertainty about study conduct, model assumptions, or whether the relative hazard is stable over follow-up. The registry record does not report whether proportional hazards were assessed.

The intervals for the two medians overlap, while the reported hazard-ratio interval does not include one. These summaries concern different statistical quantities and should not be treated as interchangeable tests. Overlap between separate confidence intervals is not, by itself, a formal test of the between-group contrast. The reported comparison is the hazard ratio from the one-sided log-rank analysis.

For the secondary response outcomes, interval widths differ across the overall intent-to-treat and PD-L1-positive sets. In general, estimates based on smaller analysis populations can be less precise, though the actual precision also depends on the observed data and method. The response estimates should be interpreted with their own confidence intervals and population definitions, rather than inferred from the primary survival interval.

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 requirements for future study planning. The time-to-event illustration is expressed as total events, while the response illustrations are expressed as participants per arm. These are different units: an event target is not a target enrollment, and a per-arm participant figure applies separately to each arm.

The calculations use a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. The effect to detect is set to the stated fraction of the observed effect. The time-to-event calculations use the Schoenfeld approach. The trial's original design assumptions and original sample-size justification are not reported in the registry record, so these illustrations should not be mistaken for the design plan used to conduct this study.

The 80% power column gives the number of participants per arm, or total events for the time-to-event outcome, associated with an 80% chance of detecting the specified effect under the planning assumptions if that effect is real. The 90% power column expresses the corresponding planning target for a 90% chance. Power is a property of a design under assumed conditions; it does not describe the probability that a result is true.

Each set of entries considers planning for 100%, 75%, and 50% of the observed effect. A smaller effect is harder to distinguish from random variation, so more participants or events are needed to maintain the same target power under the same assumptions. These sample-size figures are planning illustrations, not observed power.

For the primary time-to-event comparison, the listed targets are events rather than people. For objective response rate, the entries are participants per arm. A reader should therefore keep the outcome-specific unit in mind and avoid comparing the raw planning figures across outcome types as though they represented the same quantity.

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 = 0.700, so ln(HR) = -0.3567.
2. d = 4 × (1.96 + 0.842)² / [ln(0.700)]² = 246.8, rounded up to 247 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 <- 0.7                              # 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(3430843)
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 = 247 (80% power), 331 (90% power)
 75% of observed effect: events = 439 (80% power), 588 (90% power)
 50% of observed effect: events = 988 (80% power), 1,322 (90% power)
Simulated power with 247 events: 0.810 (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 375 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 512 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 <- 6.3          # control-arm median, months (converted from the posted unit)
hr          <- 0.7          # 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: 246        p (1-sided)   0.0001
 Events: 112       ~HR at bound   0.4946
   Month: 16   P(Cross) if HR=1   0.0001
             P(Cross) if HR=0.7   0.0338
   IA 2: 67%                  Z   2.5114
      N: 376        p (1-sided)   0.0060
 Events: 223       ~HR at bound   0.7139
   Month: 25   P(Cross) if HR=1   0.0060
             P(Cross) if HR=0.7   0.5603
       Final                  Z   1.9930
      N: 376        p (1-sided)   0.0231
 Events: 334       ~HR at bound   0.8039
   Month: 60   P(Cross) if HR=1   0.0250
             P(Cross) if HR=0.7   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 survival comparison is sponsor-reported in the registry record: a hazard ratio of 0.70, with a 95% confidence interval from 0.57 to 0.85, and p< 0.001 from a one-sided log-rank test. The median survival estimates are also posted for each treatment group. There is no discrepancy flagged between the posted primary analysis and the recalculation status in the record, and the sponsor-reported analysis is the relevant inferential result for this endpoint.

The response outcomes have posted group percentages and simple unadjusted recalculations, but the registry record lists no sponsor-reported analyses for them. Those recalculations use the posted summary statistics and should not be presented as though they were the sponsor's prespecified analysis. In particular, their unadjusted p-values do not account for the wider testing context or any analysis details that are not reported in the registry record.

For overall survival, the sponsor's method is identified as a one-sided log-rank test, and the reported effect measure is a hazard ratio. The record does not give a fuller account of the test specification, censoring rules, stratification, covariate adjustment, or assumption checks. Accordingly, the sponsor-reported estimate and test should be described as reported, without inventing missing methodological detail or treating an independent recalculation as a replacement.

16. Multiplicity, Missing Data and Analysis Populations

Multiplicity arises when several outcomes, populations, or comparisons are tested. Even when each individual test uses a conventional significance threshold, the chance of at least one apparently positive result across a family of tests can increase as more tests are performed. This is why prespecified testing order, alpha allocation, or another multiplicity-control strategy can matter when interpreting multiple secondary outcomes.

The registry record lists 14 secondary outcomes, including outcomes in the intent-to-treat set and PD-L1-positive sets. It does not report a multiplicity adjustment or a hierarchical testing plan. The secondary response recalculations are unadjusted, so their p-values should not be read as though they controlled the chance of false-positive findings across all secondary tests. The sponsor-reported primary survival analysis is distinct from those unadjusted secondary calculations.

Missing-data handling is another important issue. The primary analysis set includes all randomized participants, but this definition alone does not specify how censoring, loss to follow-up, or incomplete outcome information was handled. Specific survival censoring rules and sensitivity analyses are not reported in the registry record. For response outcomes, the record gives group denominators and response estimates, but does not fully describe how missing assessments or unevaluable participants were treated.

Population definitions also affect interpretation. The primary outcome uses the intent-to-treat set, while some secondary outcomes use PD-L1-positive analysis sets. Details on biomarker testing, availability, and population inclusion are not reported in the registry record. The resulting estimates therefore refer to the named analysis populations, and should not be assumed to apply identically across all participants.

The registry record does not report interim-analysis plans, stopping rules, or a complete statistical analysis plan. Those omissions prevent a full assessment of whether all analyses followed a prespecified plan. They do not alter the numbers posted, but they limit the strength of conclusions that can be drawn about the broader set of outcomes.

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 3% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: pneumonia (7.5%), dysphagia (5.1%), neutrophil count decreased (4.2%), febrile neutropenia (3.3%), diarrhoea (3.3%).

The registry record reports serious adverse events and other adverse events in separate groups. For serious adverse events, the at-risk count was 255 in the tislelizumab group and 240 in the investigator-chosen chemotherapy group; the affected counts were 109 and 106, respectively. These counts describe participants recorded as affected within the registry's adverse-event reporting fields. They should be interpreted using the record's group denominators and definitions rather than combined into a new summary.

The most frequently listed serious adverse-event terms include pneumonia at 7.5%, dysphagia at 5.1%, and neutrophil count decreased at 4.2%. Febrile neutropenia, diarrhoea, and white blood cell count decreased are each listed at 3.3%. The excerpt does not provide a complete arm-by-arm breakdown for each of these terms, so they should not be attributed to a particular arm unless the registry's corresponding data do so.

The other-adverse-event affected counts were 232 of 255 at risk for tislelizumab and 231 of 240 at risk for investigator-chosen chemotherapy. Death counts were 233 in each group. Counts alone do not establish whether an event was caused by treatment, its severity beyond the serious-event classification, or how the groups compare after accounting for exposure time. Detailed definitions, event duration, and exposure-adjusted analyses are not reported in the registry record.

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

The posted primary result describes a randomized comparison of tislelizumab with investigator-chosen chemotherapy in advanced unresectable or metastatic esophageal squamous cell carcinoma after first-line therapy. The estimated median overall survival was 8.6 months in the tislelizumab group and 6.3 months in the chemotherapy group. The sponsor-reported hazard ratio was 0.70, with a 95% confidence interval from 0.57 to 0.85, and the reported one-sided log-rank p-value was < 0.001. These estimates describe the groups studied and the analysis reported in the registry record.

The result does not mean that every participant receiving tislelizumab lived longer than every participant receiving chemotherapy, or that an individual participant's outcome can be predicted from a group median or hazard ratio. It also does not mean the hazard ratio is a direct measure of months gained. The median and hazard ratio summarize different features of time-to-event data.

The trial was open-label, and the chemotherapy arm allowed investigator choice among several regimens. The comparison therefore reflects the trial's treatment strategies and conduct, not a comparison against one fixed chemotherapy schedule. The record does not report enough detail to assess every possible source of bias or the generalizability of the findings to populations outside the trial.

Secondary outcomes, including response in the overall and PD-L1-positive sets, add context but do not replace the primary endpoint. Their analyses are unadjusted in the available recalculations, and the multiplicity plan is not reported. Safety counts describe recorded events but do not by themselves establish causation or comparative risk.

Overall, the posted data show a primary survival estimate favoring tislelizumab in this trial, with uncertainty represented by the confidence interval. They do not establish an individual prognosis, resolve all questions about safety, or answer every question about treatment selection. The statistical result should be interpreted with the study design and reporting limitations in view.

19. Why This Trial Matters Statistically

This trial is statistically notable for pairing a time-to-event primary endpoint with a randomized comparison in a defined second-line setting. Overall survival preserves information about event timing and follow-up, and the reported hazard ratio provides a comparative summary alongside group-specific medians. Reading both measures together helps prevent the common mistake of treating a hazard ratio as a difference in median survival.

The trial also illustrates how an open-label study can still report an objective endpoint such as death from any cause, while leaving other outcomes more open to assessment and reporting issues. Because chemotherapy was investigator-chosen, the comparator represents a treatment strategy rather than a single standardized regimen. That feature is important for interpreting the estimand: the result is about tislelizumab versus the chemotherapy approach used in this study.

Its secondary outcomes show why endpoint hierarchy and multiplicity matter. Response outcomes may offer useful context, but multiple endpoints and analysis populations create a broader inferential landscape than the primary analysis alone. Without a reported multiplicity strategy, unadjusted secondary p-values warrant caution.

Finally, the record distinguishes reported sponsor analyses from simple unadjusted recalculations. This distinction is important in statistical education: recalculation from summary data can be useful for exploration, but it does not reconstruct an unavailable protocol or replace the sponsor's primary analysis. The record also makes clear that planning illustrations based on observed effects are not evidence of observed power and cannot be mistaken for the trial's original design rationale.

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 open-label design is a trial-specific limitation. Participants and study personnel knew treatment assignment, which may affect treatment delivery, co-interventions, reporting, or decisions about assessments. Overall survival is less directly vulnerable to subjective assessment than response or quality-of-life outcomes, but open-label conduct can still influence the broader study process. The registry record does not report whether particular outcome assessments were independently blinded.

The chemotherapy comparator allowed paclitaxel, docetaxel, or irinotecan, with specified schedules. This reflects investigator-selected treatment but makes the comparator less uniform than a single fixed regimen. The registry record does not report how often each chemotherapy option was used or whether use differed by relevant participant characteristics.

The primary completion date is reported as 2020-12-01, while the response timeframe refers to the end-of-trial analysis cutoff date of 28-Dec-2022 and up to approximately 5 years. The available record excerpt does not explain the full follow-up schedule or how follow-up duration varied across participants. Details of censoring, proportional-hazards assessment, and survival sensitivity analyses are not reported.

Interpretation of secondary outcomes is limited by the absence of sponsor-reported analyses for those outcomes in the record and by the lack of a reported multiplicity strategy. The PD-L1-positive population definition and handling of missing biomarker data are not described in detail. Missing-data methods for response and quality-of-life outcomes are also not reported.

The registry record does not include the full protocol or statistical analysis plan, original sample-size justification, interim-analysis plan, or detailed adverse-event attribution and exposure-adjusted analyses. These limitations restrict assessment of prespecification, robustness, and the full safety profile. They do not invalidate the posted estimates, but they define how cautiously the results should be interpreted.

21. Related Tutorials

22. Related Statistical Calculators

23. Sources

• ClinicalTrials.gov: A Study of Tislelizumab (BGB-A317) Versus Chemotherapy as Second Line Treatment in Participants With Advanced Esophageal Squamous Cell Carcinoma — NCT03430843.
• Registry record last updated 2024-10-26; status last verified 2024-10.
• Results first posted 2023-12-29.
• 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

NCT03430843 is a completed randomized, controlled, open-label, global Phase 3 parallel-group study of tislelizumab versus investigator-chosen chemotherapy as second-line treatment for advanced unresectable or metastatic esophageal squamous cell carcinoma. The primary outcome was overall survival in the intent-to-treat analysis set.

The posted median overall survival was 8.6 months for tislelizumab and 6.3 months for investigator-chosen chemotherapy. The sponsor-reported hazard ratio was 0.70, with a 95% confidence interval from 0.57 to 0.85; the reported one-sided log-rank p-value was < 0.001.

The record also lists response, progression-free survival, duration of response, and health-related quality-of-life outcomes. Secondary analyses should be interpreted cautiously because the multiplicity strategy and several analysis details are not reported in the registry record. Safety counts describe recorded events but do not alone establish causality or comparative risk.