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
SORAYA was a phase 3, single-group treatment study evaluating mirvetuximab soravtansine in patients with platinum-resistant, advanced high-grade epithelial ovarian, primary peritoneal, or fallopian tube cancers with high folate receptor-alpha expression.
| Feature | SORAYA |
|---|---|
| Phase | Phase 3 |
| Condition | Epithelial ovarian cancer; peritoneal cancer; fallopian tube cancer |
| Population | Platinum-resistant, advanced high-grade epithelial ovarian, primary peritoneal, or fallopian tube cancers with high folate receptor-alpha expression |
| Design | Single-group |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 106 |
| Number of arms | 1 |
| Intervention | Mirvetuximab soravtansine |
| Status | Completed |
| Start date | 2020-07-23 |
| Primary completion date | 2021-11-16 |
| Lead sponsor | ImmunoGen, Inc. |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT04296890 |
2. Clinical Question
The central question was whether mirvetuximab soravtansine could produce objective tumor responses in patients with platinum-resistant, advanced high-grade epithelial ovarian, primary peritoneal, or fallopian tube cancers with high folate receptor-alpha expression.
Population
Patients with platinum-resistant, advanced high-grade epithelial ovarian, primary peritoneal, or fallopian tube cancers with high folate receptor-alpha expression.
Intervention
Mirvetuximab soravtansine.
Comparator
There was no comparator arm. The study used a single-group design.
Primary question
What percentage of participants achieved a confirmed complete response or partial response according to investigator-assessed RECIST v1.1?
3. Trial Design
Mirvetuximab soravtansine
- Mirvetuximab soravtansine was the study intervention.
- The study was open-label with no masking.
- The design model was single-group.
- The primary purpose was treatment.
Single-group evaluation
- There was one study arm.
- No randomized control group was registered.
- The primary endpoint was an objective response measure rather than a between-group treatment comparison.
- Interpretation therefore centers on the observed response proportion and its statistical uncertainty.
4. Study Timeline
Study start
The registry lists 2020-07-23 as the study start date.
Primary completion
The registry lists 2021-11-16 as the primary completion date.
Study status
The ClinicalTrials.gov record identifies SORAYA as completed and indicates that results have been posted.
5. Primary Endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Objective Response Rate (ORR) | Percentage of participants with objective response as assessed by the investigator using Response Evaluation Criteria in Solid Tumors Version 1.1 (RECIST v1.1). ORR was defined as percentage of participants with a confirmed best overall response (BOR) of complete response (CR) or partial response (PR). | Up to approximately 15 months |
What counted as a response?
The registry defines complete response (CR) as disappearance of all target or non-target lesions, with all pathological or non-pathological lymph nodes, whether target or non-target, reduced in short axis to <10 millimeters (mm).
A partial response (PR) required at least a 30% decrease in the sum of the longest diameters (SoD) of target lesions, taking the baseline SoD as the reference.
ORR is therefore a binary response endpoint at the participant level: each participant is classified according to whether the prespecified response definition is met.
6. Statistical Methodology
Objective response rate as a proportion
ORR is naturally analyzed as a binomial proportion. Each participant contributes a response classification, and the resulting proportion summarizes how frequently the prespecified CR-or-PR criterion was met.
Here, x represents the number of participants classified as responders and n represents the relevant analysis population. The registry defines the endpoint, but the posted record does not provide a formal statistical analysis of the endpoint.
Confidence intervals for ORR
A confidence interval for a response proportion is typically constructed using a binomial method. An exact binomial interval is one common approach, particularly when the response count is not large or when preserving the discrete nature of the endpoint is important.
The important statistical distinction is between the observed response proportion and the uncertainty surrounding that proportion. A point estimate alone describes the observed sample; an interval estimate communicates how precisely that sample estimates the underlying response probability under the chosen statistical framework.
Hypothesis testing in a single-group trial
For a single-group efficacy study, a formal hypothesis test would ordinarily require a prespecified benchmark response rate. The test might compare the observed response probability with that benchmark using a one-sample binomial procedure. The registry record identifies ORR as the primary endpoint but does not state such a benchmark in the information reported here.
Why RECIST classification is statistically important
The response endpoint depends on a prespecified classification rule rather than an unrestricted judgment about whether a patient improved. RECIST v1.1 establishes quantitative thresholds for CR and PR, which makes the binary response endpoint reproducible across participants when assessments are performed according to the protocol-defined framework.
7. Results
The ClinicalTrials.gov record indicates that results have been posted for SORAYA, including the primary endpoint and safety information. However, the posted record contains no formal statistical analysis for the registered primary endpoint.
| Result category | What the registry reports |
|---|---|
| Primary endpoint | Objective Response Rate (ORR) |
| Endpoint status | Results posted |
| Formal statistical analysis | No posted statistical analysis |
| Effect estimate | Not reported in the registry statistical-analysis record |
| Confidence interval | Not reported in the registry statistical-analysis record |
| p-value | Not reported in the registry statistical-analysis record |
How ORR would normally be analyzed
For this endpoint, the central descriptive result would be the proportion of participants with a confirmed BOR of CR or PR. A confidence interval would then quantify uncertainty around that proportion. If the protocol specified a benchmark response rate, a one-sample binomial hypothesis test could additionally assess whether the observed response proportion was statistically distinguishable from that benchmark.
Because the study has a single treatment arm, there is no randomized treatment-versus-control hazard ratio, risk ratio, odds ratio, or between-arm response difference to interpret. The statistical question is instead whether the observed response proportion is sufficiently large and sufficiently precise relative to a prespecified clinical or historical benchmark, if one exists.
The most important statistical feature of the SORAYA efficacy endpoint is its single-group structure. ORR can quantify how often participants achieved the defined tumor-response criterion, but the response proportion alone does not establish how much better mirvetuximab soravtansine performed than an alternative treatment.
A confidence interval, when reported, would describe uncertainty around the estimated response probability. It would not describe the range of responses expected for individual patients.
A p-value, if generated from a prespecified one-sample hypothesis test, would quantify evidence against a particular null response rate. It would not measure the size or clinical importance of the observed response rate.
In a single-group design, comparison with an external historical benchmark can also be affected by differences in eligibility criteria, disease characteristics, response assessment, subsequent treatment, and calendar period. Those issues are distinct from the precision of the observed proportion itself.
8. Safety Results
The registry reports serious adverse events by arm for the mirvetuximab soravtansine study arm. Because SORAYA had one study arm, the safety result is presented as an affected count relative to the number at risk rather than as a randomized between-group comparison.
| Safety measure | Mirvetuximab soravtansine |
|---|---|
| Serious adverse events | 36/106 |
Serious adverse events
36 participants were reported as affected among 106 participants at risk in the mirvetuximab soravtansine arm.
The 36/106 figure is a descriptive safety result. It should not be interpreted as a comparative treatment effect because there was no comparator arm in the registered study design.
A safety count such as 36/106 describes the number of participants meeting the serious-adverse-event criterion among those at risk. It does not by itself describe the severity distribution, timing, duration, attribution to treatment, or clinical consequences of individual events.
It also does not provide a counterfactual comparison. Without a concurrent comparator group, the registry result cannot establish how the serious-adverse-event frequency differs from another treatment under the same study conditions.
9. Statistical Methods Explained
Why is ORR analyzed as a binomial proportion?
Each participant can be classified according to whether a confirmed CR or PR occurred. That produces a binary outcome, making the number of responders naturally modeled as a binomial count under the usual independence framework. The resulting response proportion is the primary descriptive quantity.
Why does a single-group study not produce a treatment hazard ratio?
A hazard ratio compares instantaneous event rates between groups in a time-to-event model. SORAYA has one treatment group and its registered primary endpoint is ORR rather than a time-to-event endpoint. There is therefore no randomized treatment-versus-control hazard ratio for the primary endpoint.
What does a confidence interval for ORR tell us?
It describes statistical uncertainty around the estimated response probability. A narrower interval indicates greater statistical precision under the selected interval procedure, while a wider interval indicates greater uncertainty. The interval is not a prediction interval for individual patients.
Why would a historical benchmark be important?
In a single-arm trial, a response proportion has no concurrent control group against which it can be directly compared. A prespecified benchmark can provide the reference value for a one-sample hypothesis test, but the validity of that comparison depends on how well the historical benchmark represents the population, treatment setting, eligibility criteria, and response-assessment process of the current study.
Why is RECIST v1.1 important?
RECIST v1.1 converts tumor imaging observations into predefined response categories. The SORAYA registry definition requires a confirmed best overall response of CR or PR, with specific quantitative requirements for lesion disappearance or reduction. This standardization reduces ambiguity in the binary endpoint.
Why does the lack of a comparator matter?
Without randomization to a concurrent control treatment, observed response cannot be separated as cleanly from patient selection, disease characteristics, assessment timing, and other differences between the study population and a potential alternative-treatment population. A single-arm response result is therefore descriptive and benchmark-dependent rather than a direct randomized treatment-effect estimate.
10. Understanding the Primary Endpoint Statistically
ORR is attractive as a clinical-trial endpoint because it reduces complex longitudinal tumor measurements to an interpretable binary outcome. For each participant, the investigator's RECIST assessment ultimately contributes to the numerator or denominator of the response calculation.
Numerator
Participants whose confirmed best overall response is complete response or partial response.
Denominator
The relevant participants in the response analysis population.
Clinical meaning
The proportion quantifies how frequently the prespecified tumor-response criterion was achieved.
Statistical meaning
The estimate is subject to sampling uncertainty and should ordinarily be accompanied by an appropriate confidence interval.
Response is not the same as disease control
A response endpoint specifically counts participants meeting the CR or PR definitions. Participants who do not meet those response definitions are not counted as responders, even though their disease course could involve other clinically relevant outcomes. ORR therefore answers a narrower question than a composite disease-control endpoint.
Response is not the same as survival
ORR measures tumor-response status according to the specified assessment framework. Overall survival, by contrast, is a time-to-event outcome. A response proportion therefore cannot be translated directly into a survival probability, median survival time, or hazard ratio.
11. What a Confidence Interval Would Add
Suppose an observed response proportion were accompanied by a 95% confidence interval. The interval would provide a range of values that are compatible with the data under the specified statistical procedure and confidence level.
Confidence interval = uncertainty around that estimate
These quantities answer different questions. The point estimate describes the observed sample, while the interval communicates statistical precision.
For a single-arm endpoint, the width of the confidence interval depends strongly on the number of participants and the response proportion. With 106 enrolled participants, the study has a finite amount of statistical information; precision cannot be separated from the number of observations contributing to the endpoint.
The confidence interval also does not solve the central causal limitation of a single-arm study. A highly precise estimate of ORR is still an estimate from one treatment group. Precision and comparative validity are different properties.
12. Single-Group Design and Causal Interpretation
The absence of randomization changes the statistical interpretation of the study. In a randomized controlled trial, treatment assignment creates a framework for comparing outcomes between groups while balancing measured and unmeasured baseline factors in expectation. SORAYA does not have that randomized comparison.
| Question | Single-group ORR can address | Single-group ORR cannot directly establish |
|---|---|---|
| Observed response | How frequently participants met the prespecified response definition | — |
| Precision | How uncertain the estimated response proportion is, using an appropriate confidence interval | — |
| Comparative efficacy | Requires an external benchmark or other comparison framework | Randomized superiority versus another treatment |
| Treatment effect | Can be assessed relative to a prespecified benchmark if scientifically justified | A concurrent randomized treatment effect |
| Individual benefit | Provides a population-level response proportion | The probability that a particular individual will respond |
13. Endpoint Definition in Detail
Confirmed best overall response
The registered endpoint does not simply count any observed tumor shrinkage. It requires a confirmed best overall response of CR or PR. This is important because the endpoint is based on the prespecified RECIST classification rather than on a continuous tumor-size measurement alone.
Complete response
The registry defines CR as disappearance of all target or non-target lesions. It additionally specifies that all pathological or non-pathological lymph nodes, whether target or non-target, must have reduction in short axis to <10 mm.
Partial response
The registry defines PR as at least a 30% decrease in the sum of the longest diameters of target lesions, using baseline sum of longest diameters as the reference.
14. Analysis Populations and Denominators
The registry reports enrollment of 106 participants and a serious-adverse-event denominator of 106. For ORR, the conceptual denominator is the relevant response-analysis population, but the ClinicalTrials.gov record does not provide a separate response-analysis sample size or a posted responder count.
| Population concept | Information reported |
|---|---|
| Total enrollment | 106 |
| Study arms | 1 |
| Serious adverse-event population at risk | 106 |
| ORR response-analysis denominator | Not separately reported in the registry statistical-analysis record |
| ORR responder count | Not reported in the registry statistical-analysis record |
This distinction matters because an endpoint's denominator determines the meaning of its percentage. A response rate should be interpreted together with the population from which the response classifications were derived.
15. Statistical Interpretation of Safety
Safety and efficacy have different statistical structures. The primary efficacy endpoint is a binary response measure based on RECIST v1.1, whereas the registry's safety result is a participant count affected by serious adverse events.
Efficacy
ORR asks whether each participant achieved a confirmed CR or PR and summarizes those classifications as a proportion.
Safety
The serious-adverse-event result counts participants affected among those at risk in the study arm.
No direct trade-off statistic
The two results should not be collapsed into a single numerical benefit-risk measure without an explicit statistical framework.
No randomized safety contrast
Because the trial has one arm, the serious-adverse-event result does not estimate a comparative risk versus another treatment.
16. Limitations
- Single-group design: there is no concurrent randomized comparator against which the observed response proportion can be directly contrasted.
- Benchmark dependence: a formal efficacy hypothesis test in a single-arm study generally requires a prespecified reference response rate, and the ClinicalTrials.gov record does not report one.
- Response classification: ORR depends on investigator-assessed RECIST v1.1 classifications and therefore depends on the quality, timing, and consistency of tumor assessments.
- Confirmation requirement: the endpoint requires a confirmed best overall response, so an isolated observation of tumor shrinkage is not necessarily sufficient to qualify as a response.
- Denominator considerations: the registry reports 106 enrolled participants and 106 at risk for the reported serious-adverse-event measure, but does not separately provide the ORR response-analysis denominator in the ClinicalTrials.gov record.
- Limited comparative inference: a response proportion from one treatment group cannot by itself establish superiority or inferiority relative to another treatment.
- Safety interpretation: the reported 36/106 serious-adverse-event result is descriptive and cannot provide a randomized comparative safety estimate.
- Statistical reporting: the registry indicates that results were posted but does not post a formal statistical analysis for the registered primary endpoint.
17. Why This Trial Matters Statistically
SORAYA is a useful teaching example because it illustrates how the statistical questions change when a phase 3 study uses a single-group design and a binary tumor-response endpoint rather than a randomized time-to-event comparison.
| Concept | How it appears in SORAYA |
|---|---|
| Phase 3 trial | The study is registered as phase 3. |
| Single-group design | One study arm received mirvetuximab soravtansine. |
| No masking | The registered masking status is none. |
| Binary efficacy endpoint | ORR is based on whether a confirmed CR or PR occurred. |
| RECIST v1.1 | Response is assessed using standardized tumor-response criteria. |
| Binomial estimation | ORR can be represented as a response proportion with an appropriate binomial confidence interval. |
| Historical benchmark | A formal one-sample hypothesis test would ordinarily require a prespecified reference response rate. |
| Comparative inference | No randomized treatment effect can be estimated because there is no comparator arm. |
| Safety analysis | Serious adverse events are reported as 36/106 in the single treatment arm. |
| Statistical reporting | The registry indicates posted results but contains no posted formal statistical analysis for the primary endpoint. |
18. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary endpoint is a binary response measure summarized as a proportion. Its precision would ordinarily be described with a binomial confidence interval, while a formal hypothesis test would require a prespecified benchmark.
Clinical interpretation
The endpoint asks whether participants achieved a confirmed complete or partial response under RECIST v1.1. That provides a clinically interpretable tumor-response measure, but it does not by itself establish comparative efficacy.
This distinction is especially important for single-arm oncology studies. Statistical significance, precision, response magnitude, and comparative effectiveness are related but different concepts.
19. What ORR Does — and Does Not — Mean
An ORR estimate represents the proportion of participants classified as having a confirmed best overall response of CR or PR according to the registered RECIST v1.1 definition.
It does not mean that every responder experienced the same magnitude or duration of benefit. It also does not mean that participants who were not classified as responders necessarily experienced immediate disease progression or no clinical benefit.
A confidence interval around ORR would quantify uncertainty in the estimated response probability. It would not describe the range of tumor responses among individual participants.
Even a precisely estimated response proportion from a single treatment group does not provide a direct randomized estimate of how the treatment compares with another therapy. Comparative inference requires an appropriate reference group or benchmark.
20. Interpreting the 36/106 Safety Result
The registry reports serious adverse events affecting 36 of 106 participants in the mirvetuximab soravtansine arm. This is a participant-level safety summary, not an estimate of the number of individual serious adverse events.
A participant can experience more than one adverse event. Therefore, an affected-participant count should not be interpreted as the total number of serious adverse-event events.
The result also does not by itself establish causality. A serious adverse event recorded during a study is a safety outcome; determining treatment attribution requires the corresponding clinical assessment and safety definitions.
21. Broader Statistical Lessons
SORAYA demonstrates several principles that generalize to clinical-trial analysis.
| Statistical lesson | Practical implication |
|---|---|
| Define the estimand first | The meaning of the analysis depends on whether the target is a response probability, a treatment contrast, a time-to-event quantity, or another estimand. |
| Respect the study design | A single-arm design should not be interpreted as though it were randomized. |
| Separate point estimates from uncertainty | An observed proportion and its confidence interval provide complementary information. |
| Separate efficacy from safety | Response and serious-adverse-event measures address different outcomes and should not be combined without an explicit framework. |
| Use endpoint definitions precisely | RECIST thresholds determine which participants qualify as CR or PR. |
| Distinguish descriptive and inferential statistics | A response proportion describes the observed study population; a hypothesis test requires a prespecified comparison or null value. |
| Interpret denominators carefully | The denominator must correspond to the population represented by the reported endpoint. |
22. Sources
- ClinicalTrials.gov: NCT04296890 — SORAYA.
- PubMed: PubMed record.
- PubMed: PubMed record.
- PubMed: PubMed record.
Continue through the Clinical Biostats clinical-trial library
Explore additional clinical trial analyses, statistical tutorials, and quantitative tools for understanding how clinical-trial evidence is designed and analyzed.
23. Record Summary
SORAYA is a phase 3, single-group study of mirvetuximab soravtansine in platinum-resistant, advanced high-grade epithelial ovarian, primary peritoneal, or fallopian tube cancers with high folate receptor-alpha expression. Its registered primary endpoint is objective response rate, defined as the percentage of participants with a confirmed best overall response of complete response or partial response according to investigator-assessed RECIST v1.1, with a time frame of up to approximately 15 months.
Statistically, the study is best understood as a single-arm estimation problem rather than a randomized treatment-comparison problem. ORR is naturally represented as a binomial proportion, with an appropriate confidence interval providing information about precision. A formal one-sample hypothesis test would require a prespecified benchmark response rate. The registry indicates that results were posted, but no formal statistical analyses were posted for the primary endpoint. The reported safety result of 36/106 serious-adverse-event-affected participants provides a descriptive safety measure for the single study arm.