This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
ORIENT-15 was a randomized, parallel, quadruple-masked, phase 3 treatment trial enrolling 746 subjects with esophageal squamous cell carcinoma. The study compared sintilimab plus chemotherapy with placebo plus chemotherapy, with overall survival designated as the primary endpoint in both the overall population and the PD-L1 positive population.
| Feature | ORIENT-15 |
|---|---|
| Clinical trial | ORIENT-15 |
| ClinicalTrials.gov identifier | NCT03748134 |
| Phase | Phase 3 |
| Status | COMPLETED |
| Condition | Esophageal Squamous Cell Carcinoma |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | QUADRUPLE |
| Primary purpose | TREATMENT |
| Enrollment | 746 |
| Number of arms | 3 |
| Lead sponsor | Innovent Biologics (Suzhou) Co. Ltd. |
| Sponsor type | INDUSTRY |
2. Clinical Question
The registry frames ORIENT-15 around a first-line treatment comparison in subjects with unresectable, locally advanced, recurrent or metastatic esophageal squamous cell carcinoma. The primary statistical question is whether overall survival differs between sintilimab in combination with chemotherapy and placebo in combination with chemotherapy.
Population
Subjects with unresectable, locally advanced, recurrent or metastatic esophageal squamous cell carcinoma receiving first-line treatment.
Intervention
Sintilimab in combination with chemotherapy. The registered interventions include sintilimab, cisplatin, paclitaxel, and fluorouracil.
Comparator
Placebo in combination with chemotherapy. The registered interventions include placebo together with cisplatin, paclitaxel, and fluorouracil.
Primary question
Does sintilimab combined with chemotherapy improve overall survival compared with placebo combined with chemotherapy in the overall population and in the PD-L1 positive population?
3. Trial Design
The parallel structure means subjects are assigned to treatment groups and followed prospectively rather than moving through sequential treatment periods as part of a crossover design. Quadruple masking indicates that the registry classifies the study as having four masked parties. The registry does not provide additional masking-role detail in the record summarized here.
Sintilimab + chemotherapy
- Sintilimab
- Cisplatin
- Paclitaxel
- Fluorouracil
Placebo + chemotherapy
- Placebo
- Cisplatin
- Paclitaxel
- Fluorouracil
4. Primary Endpoints
| Endpoint | Registry definition and time frame | Primary statistical role |
|---|---|---|
| OS in overall population | From date of randomization until the date of death from any cause, assessed up to 40 months. | Primary endpoint |
| OS in PD-L1 positive population | From date of randomization until the date of death from any cause, assessed up to 40 months. | Primary endpoint |
The two primary endpoints are closely related time-to-event outcomes. Both use the same event definition—death from any cause—and the same starting point, date of randomization. Their distinction is the analysis population: the first concerns the overall population, whereas the second concerns the PD-L1 positive population.
Why the time origin matters
Using randomization as time zero establishes a common reference point for the randomized treatment comparison. Each subject contributes follow-up beginning at randomization, with the event of interest being death from any cause. This makes the endpoint a classic overall-survival time-to-event outcome rather than a simple binary endpoint assessed at a fixed visit.
5. Statistical Methodology
The registry identifies overall survival as the primary outcome in two populations but does not post formal statistical analyses or numerical efficacy results for those endpoints. For an endpoint defined as time from randomization to death from any cause, the conventional analysis framework is survival analysis.
Kaplan-Meier estimation
A Kaplan-Meier estimator is typically used to describe the distribution of overall survival over time. It accommodates right-censored observations, allowing a subject who has not died by the last available follow-up to contribute information up to that censoring time.
Here, di is the number of deaths at time ti, and ni is the number of subjects at risk immediately before that time.
Between-group comparison
For randomized treatment groups, a log-rank test is commonly used to compare survival distributions. The test evaluates whether the observed pattern of event times differs between groups over follow-up. Its null hypothesis concerns the survival experience across the groups rather than a particular single time-point survival percentage.
Cox proportional-hazards model
A Cox proportional-hazards model is commonly used to estimate a hazard ratio for a randomized comparison of overall survival. The hazard ratio summarizes the relative instantaneous event rate under the model. A hazard ratio below 1 would correspond to a lower estimated hazard in the numerator treatment group; a hazard ratio above 1 would correspond to a higher estimated hazard.
The hazard ratio is a relative, model-based measure of event rate over time. It is not itself a probability, a risk difference, a median survival time, or the percentage of subjects who benefit.
Confidence intervals
A confidence interval around a hazard ratio quantifies statistical uncertainty around the estimated relative effect under the model and sampling framework. A narrow interval indicates greater statistical precision than a wide interval, but neither type of interval describes the range of outcomes that an individual patient could experience.
Analysis of the PD-L1 positive population
The second primary endpoint restricts the population to subjects classified as PD-L1 positive. Statistically, this creates a prespecified population-specific survival comparison. The treatment comparison should be interpreted within that defined population rather than treating the PD-L1 positive analysis as interchangeable with the overall-population analysis.
6. Planned Analysis
No formal statistical analyses were posted to ClinicalTrials.gov. The registry identifies what will be measured through the two primary overall-survival endpoints, but it does not provide numerical estimates, confidence intervals, or p-values for them.
Primary endpoint: OS in overall population
Measured from date of randomization until the date of death from any cause, assessed up to 40 months.
For this type of endpoint, a complete analysis would ordinarily describe the Kaplan-Meier survival estimates over time and compare the randomized treatment groups using an appropriate time-to-event comparison. A Cox proportional-hazards model can provide a hazard ratio and confidence interval when its assumptions are appropriate.
Primary endpoint: OS in PD-L1 positive population
Measured from date of randomization until the date of death from any cause, assessed up to 40 months in the PD-L1 positive population.
The same survival-analysis framework can ordinarily be applied within the PD-L1 positive population. Interpretation should remain population-specific, because the estimand concerns survival among the subjects meeting the PD-L1-positive definition rather than survival in the entire randomized population.
7. Statistical Methods Explained
Why is overall survival a time-to-event endpoint?
Overall survival records not only whether death occurs but also when it occurs relative to randomization. Subjects who remain alive at their last observed follow-up are generally right-censored rather than treated as if they had experienced the event. This structure makes survival methods more informative than simply calculating the proportion dead at an arbitrary time point.
Why use Kaplan-Meier estimation?
Kaplan-Meier estimation provides a way to estimate the probability of remaining alive over time while accommodating different follow-up durations and right censoring. It produces a survival curve rather than reducing all follow-up to a single binary outcome.
What does a hazard ratio measure?
A hazard ratio compares the estimated instantaneous event rates between treatment groups within a time-to-event model. For example, an HR of 0.70 would describe a 30% lower estimated hazard in the treatment group relative to the comparator under the model. It would not mean that 30% of patients avoided death or that survival probability was 30 percentage points higher.
Why is the confidence interval important?
The point estimate alone does not show how precisely the treatment effect was estimated. A confidence interval supplies a range of parameter values compatible with the statistical model and observed information at the stated confidence level. It should be considered alongside the effect estimate rather than treated as an afterthought.
Why does the PD-L1 positive endpoint need separate interpretation?
The registry defines OS separately for the overall population and the PD-L1 positive population. These are different analysis populations, so an effect estimate from one should not automatically be interpreted as the effect estimate in the other. The distinction is especially important when discussing whether a treatment effect appears consistent across populations.
Why is the p-value not a measure of effect size?
If a p-value is reported for a survival comparison, it addresses compatibility with a specified null hypothesis under the statistical testing framework. It does not tell the reader how large the treatment effect is. Effect size is better conveyed using measures such as a hazard ratio together with its confidence interval and, where available, absolute survival estimates.
8. Understanding Censoring in Overall Survival
Right censoring is fundamental to the analysis of an endpoint measured from randomization until death. A subject who remains alive at the end of observed follow-up has not experienced the event, but that subject has still contributed information about survival up to the time of last observation.
Observed death
The subject contributes an event at the observed date of death.
Right censoring
The subject contributes follow-up through the last time survival status is known without an observed death.
Common time origin
The registry defines time zero as the date of randomization.
Assessment window
The registry specifies assessment up to 40 months for both primary OS endpoints.
The statistical consequence is that a valid survival analysis must account for varying amounts of observed follow-up. Simply excluding censored subjects or counting them as deaths would discard or misclassify information.
9. Hazard Ratios and Proportional Hazards
The Cox proportional-hazards model is a standard framework for estimating relative effects on time-to-event outcomes. Its interpretation depends on the proportional-hazards assumption: the relative hazard between treatment groups is assumed to be reasonably stable over time, conditional on the model.
A hazard ratio summarizes a relative comparison of instantaneous event rates. It does not directly provide an absolute survival probability, median survival time, or individual-level probability of benefit.
If treatment effects change substantially over time, one constant hazard ratio can compress a complex time-varying relationship into a single number. In that setting, survival curves and time-specific estimates can provide important additional information.
The confidence interval indicates the statistical precision of the hazard-ratio estimate. It does not establish that every individual experiences an effect inside that interval.
10. Multiplicity and the Two Primary Endpoints
ORIENT-15 has two registered primary endpoints: OS in the overall population and OS in the PD-L1 positive population. When a trial has multiple primary hypotheses, the statistical design generally needs a prespecified approach to controlling the overall type I error rate if both hypotheses are intended to support confirmatory conclusions.
| Endpoint | Population | Time frame | Statistical consideration |
|---|---|---|---|
| OS in overall population | Overall population | From randomization to death from any cause, assessed up to 40 months | Primary time-to-event hypothesis |
| OS in PD-L1 positive population | PD-L1 positive population | From randomization to death from any cause, assessed up to 40 months | Primary population-specific time-to-event hypothesis |
The registry information available here does not specify the alpha-allocation, hierarchical testing procedure, gatekeeping strategy, or other multiplicity procedure used for the two primary endpoints. Those details matter when interpreting multiple confirmatory tests and should be obtained from the prespecified statistical analysis plan when available.
11. Interim Analysis and Other Design Features
The available registry information identifies randomization, parallel allocation, quadruple masking, treatment purpose, enrollment, and the two primary OS endpoints. It does not report a prespecified interim-analysis schedule, alpha-spending approach, non-inferiority margin, crossover procedure, factorial structure, Bayesian method, or missing-data/imputation method.
| Design topic | What the registry information establishes |
|---|---|
| Randomization | RANDOMIZED allocation |
| Parallel design | PARALLEL design model |
| Masking | QUADRUPLE masking |
| Primary purpose | TREATMENT |
| Non-inferiority margin | Not reported in the available registry record |
| Crossover | Not reported in the available registry record |
| Factorial design | Not reported; the registry identifies a PARALLEL design |
| Interim analysis | Not reported in the available registry record |
| Missing-data/imputation method | Not reported in the available registry record |
| Bayesian methods | Not reported in the available registry record |
This distinction is important because absence of a methodological detail from the registry summary does not establish that the method was not used. It means that the available registry information does not specify the method.
12. Study Timeline
Study start
The registered start date for ORIENT-15 was 2018-12-24.
Primary completion
The registered primary completion date was 2021-09-09.
Current registry status
The ClinicalTrials.gov record identifies the study status as COMPLETED.
13. What the Registry Does and Does Not Report
The ClinicalTrials.gov record establishes a completed phase 3 randomized trial with 746 enrolled subjects and three arms, using a parallel design and quadruple masking. It identifies overall survival as the primary endpoint in both the overall population and the PD-L1 positive population, with both endpoints assessed from randomization to death from any cause up to 40 months.
The registry record does not post formal statistical analyses for these primary endpoints. Consequently, numerical hazard ratios, confidence intervals, p-values, median overall survival estimates, survival percentages, subgroup estimates, and adverse-event results are not reported here as trial outcomes.
14. Limitations and Interpretation Issues
- No posted statistical analyses: the ClinicalTrials.gov record does not provide formal numerical analyses for the two primary OS endpoints.
- Multiple primary endpoints: two primary OS populations are registered, but the available registry information does not specify how multiplicity was controlled.
- Population definition: the PD-L1 positive endpoint applies to a specific population, so its interpretation is not identical to the overall-population endpoint.
- Hazard-ratio assumptions: if a Cox model is used, interpretation of a single hazard ratio depends on the proportional-hazards assumption.
- Censoring: survival analyses require appropriate handling of subjects whose death has not been observed during available follow-up.
- Unreported statistical details: the available registry information does not specify the interim-analysis framework, missing-data method, Bayesian methods, or other detailed SAP procedures.
- Three-arm structure: the record identifies three arms but the available summary does not provide enrollment counts or sufficiently detailed regimen descriptions for each individual arm.
- Generalizability: the population represented by a clinical trial is determined by its eligibility criteria and study setting; the available registry summary does not provide a full baseline-characteristics table.
15. Why This Trial Matters Statistically
ORIENT-15 is a useful statistical teaching example because its primary objectives illustrate how a randomized oncology trial can define the same time-to-event outcome across two clinically distinct populations. The design also provides a clear setting for understanding randomization, masking, censoring, survival estimation, hazard ratios, confidence intervals, and multiplicity.
| Concept | How it appears in ORIENT-15 |
|---|---|
| Randomization | The registry classifies allocation as RANDOMIZED. |
| Parallel design | The registered design model is PARALLEL. |
| Masking | The registry classifies the study as QUADRUPLE masked. |
| Overall survival | Both primary endpoints measure time from randomization to death from any cause. |
| Kaplan-Meier analysis | A standard method for estimating survival distributions with right censoring. |
| Hazard ratio | A standard model-based measure for comparing event rates over time. |
| Confidence interval | Provides statistical uncertainty around an estimated treatment effect. |
| Multiplicity | Two distinct primary OS populations require attention to the joint testing framework. |
| Population-specific analysis | The second primary endpoint specifically concerns the PD-L1 positive population. |
| Follow-up | The registered time frame for both primary OS endpoints extends up to 40 months. |
16. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The registry defines two primary time-to-event endpoints and a randomized comparison involving sintilimab or placebo with chemotherapy. A formal analysis would quantify the survival distributions and relative treatment effect while accounting for censoring and the prespecified testing framework.
Clinical interpretation
The registered clinical question concerns first-line treatment of unresectable, locally advanced, recurrent or metastatic esophageal squamous cell carcinoma. Whether the observed treatment effect supports a clinical conclusion requires numerical outcome data and its appropriate statistical context.
Keeping these interpretations separate prevents a study design from being mistaken for an observed result. Randomization establishes the framework for a causal treatment comparison; the eventual magnitude, precision, and clinical meaning of that comparison depend on the observed data and prespecified analysis.
17. The Statistical Logic of the Primary Comparison
The basic inferential chain can be expressed as follows:
Randomization provides the foundation for comparing treatment assignments. Follow-up supplies event and censoring information. Survival methods then estimate the underlying time-to-event distributions. A treatment-effect model such as Cox regression can summarize the relative event rate, while confidence intervals quantify statistical precision and hypothesis testing addresses the prespecified null hypothesis.
The second primary endpoint repeats this logic within the PD-L1 positive population. Because two primary hypotheses are specified, the final inferential interpretation also depends on whatever multiplicity-control strategy was prespecified for the trial.
18. Record Summary
ORIENT-15 is a completed phase 3 randomized clinical trial with 746 enrolled subjects, three arms, a parallel design, and quadruple masking. The registered intervention framework includes sintilimab or placebo with chemotherapy consisting of cisplatin, paclitaxel, and fluorouracil. Its two primary endpoints are overall survival in the overall population and overall survival in the PD-L1 positive population, both measured from randomization to death from any cause and assessed up to 40 months.
From a statistical perspective, the central issues are those of a time-to-event randomized comparison: defining the time origin, handling right censoring, estimating survival over time, quantifying relative treatment effects, assessing precision with confidence intervals, and interpreting two primary populations within an appropriate multiplicity framework. The ClinicalTrials.gov record does not post formal statistical analyses or numerical results for these endpoints, so the registry establishes the planned statistical question without supplying an observed treatment effect.
19. Sources
- ClinicalTrials.gov: ORIENT-15 (NCT03748134).
- PubMed: PubMed record.
- PubMed: PubMed record.
- PubMed: PubMed record.
Continue through the Clinical Biostats clinical-trial library
Clinical trial analysis connects study design and endpoint definitions with the statistical methods used to estimate treatment effects and quantify uncertainty.