This page separates reported registry information from statistical interpretation. ClinicalTrials.gov reports that results are posted for the registered primary endpoints, but the ClinicalTrials.gov record contains no formal statistical analyses, estimates, confidence intervals, or p-values for those endpoints.
1. Trial at a Glance
SAVANNAH is a randomized, parallel-group, quadruple-masked phase 2 oncology trial of osimertinib, savolitinib, and placebo in patients described by the registry as having EGFRm+/MET+ NSCLC following prior osimertinib.
| Feature | SAVANNAH |
|---|---|
| Trial name | SAVANNAH |
| NCT identifier | NCT03778229 |
| Phase | Phase 2 |
| Status | Active, not recruiting |
| Start | 2019-01-09 |
| Primary completion | 2024-08-23 |
| Enrollment | 367.0 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Lead sponsor | AstraZeneca |
| Sponsor type | Industry |
| Conditions | Carcinoma |
2. Clinical Question
The registry identifies the trial as evaluating osimertinib plus savolitinib in EGFRm+/MET+ NSCLC following prior osimertinib. The registered primary outcomes are objective response rate (ORR) by investigator assessment in two specified analysis populations.
Population
Patients described in the trial title as having EGFRm+/MET+ NSCLC following prior osimertinib.
Interventions
The registry identifies osimertinib, savolitinib, and placebo as study interventions.
Design question
The trial is randomized, parallel, and quadruple-masked, allowing response to be assessed within a controlled experimental framework.
Primary question
What percentage of subjects achieved a confirmed objective response according to investigator assessment in the registered analysis populations?
3. Trial Design
Osimertinib + savolitinib combinations
- Osimertinib is listed as a study intervention.
- Savolitinib is listed as a study intervention.
- Placebo is also listed as a study intervention.
Randomized parallel comparison
- 2 study arms
- Randomized allocation
- Quadruple masking
- Primary purpose: treatment
4. Trial Timeline and Registry Status
Trial start
The registry lists 2019-01-09 as the study start date.
Primary completion
The registry lists 2024-08-23 as the primary completion date.
Active, not recruiting
The ClinicalTrials.gov record identifies the study status as ACTIVE_NOT_RECRUITING.
5. Primary Endpoints
| Registered endpoint | Time frame | Type |
|---|---|---|
| Objective Response Rate (ORR) by Investigator Assessment; Target Population Analysis Set | Time from randomization/first dose of study treatment to progression or last evaluable assessment in the absence of progression until the end of the study, or an approximate maximum of 5.6 years. | Binary |
| Objective Response Rate (ORR) by Investigator Assessment; 300 mg QD Safety Analysis Set | Time from randomization/first dose of study treatment to progression or last evaluable assessment in the absence of progression until the end of the study, or an approximate maximum of 5.6 years. | Binary |
Registry definition of objective response
The registry defines objective response as the percentage of subjects with a confirmed Objective Response (OR) per Response Evaluation Criteria In Solid Tumors (RECIST 1.1). Objective response includes complete response, defined as disappearance of all target lesions, or partial response, defined as a ≥30% decrease in target lesion diameters.
The registry further states that the objective response must be recorded at one visit and confirmed by repeat imaging at least 4 weeks later, with no disease progression between the initial assessment and confirmation.
For each evaluable subject, the primary outcome can be represented as a binary indicator: confirmed objective response versus no confirmed objective response. The group-level ORR is the proportion of subjects classified as responders under the prespecified response definition.
6. Analysis Populations
The two registered primary endpoints explicitly identify different analysis sets. This is statistically important because the denominator used for an ORR calculation determines the population to which the response percentage applies.
| Analysis set | Primary endpoint using the set | Statistical implication |
|---|---|---|
| Target Population Analysis Set | ORR by Investigator Assessment | The reported response proportion applies to the subjects included in this prespecified population. |
| 300 mg QD Safety Analysis Set | ORR by Investigator Assessment | The response proportion is evaluated in the specified safety analysis population rather than automatically being interchangeable with the target population analysis. |
The ClinicalTrials.gov record does not provide the numerical sizes of these two analysis sets. Therefore, an ORR percentage, responder count, or confidence interval cannot be reconstructed from the available information.
7. Results and Statistical Reporting Status
ClinicalTrials.gov indicates that results are posted for both registered primary endpoints. However, the ClinicalTrials.gov record contains no formal statistical analyses. In particular, it provides no treatment-effect estimate, confidence interval, formal comparison, or p-value for either primary endpoint.
| Primary endpoint | Results posted? | Formal statistical analysis reported? |
|---|---|---|
| ORR; Target Population Analysis Set | Yes | No statistical analysis entry reported |
| ORR; 300 mg QD Safety Analysis Set | Yes | No statistical analysis entry reported |
8. Planned Analysis
Because the ClinicalTrials.gov record does not contain formal statistical analyses, the appropriate educational approach is to describe the analysis that would ordinarily accompany these registered binary endpoints without presenting an unreported result.
Objective response rate
For a binary ORR endpoint, the basic estimand is the proportion of subjects with a confirmed objective response under the RECIST 1.1 definition:
The numerator counts subjects meeting the complete-response or partial-response definition with the required confirmation. The denominator is determined by the specified analysis population.
Between-group comparison
If a formal comparison between randomized groups is performed, the primary statistical question for a binary endpoint is whether the observed response proportions differ between treatment groups under the prespecified analysis plan. Depending on the protocol and estimand, this can be expressed as a risk difference, risk ratio, odds ratio, or another prespecified measure, with an appropriate confidence interval and hypothesis test.
The ClinicalTrials.gov record does not identify which comparative effect measure, confidence-interval method, or hypothesis-testing procedure was used for SAVANNAH. Those details therefore are not attributed to the trial.
Confidence interval for ORR
A confidence interval would quantify statistical uncertainty around the estimated response proportion or treatment contrast. It would not describe the range of responses that individual patients could experience.
Confirmation requirement
The confirmation rule is central to the endpoint definition. A transient initial tumor response is not sufficient by itself under the registry definition; the response must be confirmed by repeat imaging at least 4 weeks later, with no disease progression between the initial assessment and confirmation.
9. Statistical Methods Explained
Why is ORR a binary endpoint?
Each subject can ultimately be classified according to whether a confirmed objective response occurred under the registry definition. That produces a binary outcome suitable for estimating a response proportion. The statistical challenge is then to quantify uncertainty around that proportion and, where specified, compare proportions between randomized groups.
Why does the analysis population matter?
The two primary endpoints use different named analysis sets. A response rate calculated in the Target Population Analysis Set answers a different statistical question from a response rate calculated in the 300 mg QD Safety Analysis Set if their included subjects differ. Comparing percentages without preserving their denominators can therefore be misleading.
Why is response confirmation important?
RECIST-based response is not simply any observed decrease in tumor size. The registry requires the response to be recorded at one visit and confirmed by repeat imaging at least 4 weeks later, with no disease progression between the initial assessment and confirmation. This reduces the chance that a short-lived or uncertain radiologic change is counted as a confirmed response.
What does randomization contribute statistically?
Randomization creates the basis for a controlled comparison because treatment assignment is determined by the trial design rather than chosen by investigators or participants. If the randomized groups are analyzed according to the prespecified framework, differences in outcomes can be interpreted within the causal structure created by the randomized allocation.
What does quadruple masking contribute?
Quadruple masking is intended to reduce opportunities for knowledge of treatment assignment to influence trial conduct or assessment. This is particularly relevant when an endpoint involves investigator assessment, because knowledge of treatment assignment could otherwise be a potential source of assessment bias.
Why should an ORR percentage not be treated as a survival effect?
ORR measures whether a confirmed tumor response occurred. It does not measure time to progression, time to death, or overall survival. A response endpoint therefore answers a different clinical and statistical question from a time-to-event endpoint, even when both are collected in the same trial.
10. Interpreting a Future ORR Result
If a formal ORR comparison is reported in a later version of the registry or source analysis, several statistical components should be read together rather than in isolation.
Effect estimate
The response proportions describe the frequency of confirmed objective response in the specified analysis populations. A comparative effect measure would describe the difference or relative contrast between groups.
Confidence interval
The confidence interval describes uncertainty around the estimated effect under the selected statistical framework. A narrower interval generally indicates greater statistical precision than a wider interval.
P-value
A p-value addresses compatibility with a specified null hypothesis under the test assumptions. It is not a measure of the magnitude or clinical importance of an effect.
Analysis population
The population must be read alongside the percentage. The registered endpoint explicitly distinguishes the Target Population Analysis Set from the 300 mg QD Safety Analysis Set.
For a binary endpoint such as ORR, the central statistical object is a proportion. A treatment-group response percentage tells us how frequently confirmed response occurred in that analyzed population. It does not tell us how long the response lasted, whether patients lived longer, or whether every responding patient experienced the same magnitude of benefit.
A comparative estimate, if reported, would quantify the contrast between randomized groups. Its confidence interval would communicate precision. The p-value, if reported, would address the prespecified hypothesis test; it would not measure effect size. Because no formal statistical analyses entry is reported here, none of these comparative quantities can responsibly be inserted.
11. Safety Results
The ClinicalTrials.gov record provides serious adverse-event counts by reported safety group. These are presented as affected subjects divided by subjects at risk.
| Safety group | Serious adverse events | Affected / at risk |
|---|---|---|
| Savo 300mg QD + Osi | Serious adverse events | 62/196 |
| Savo 300mg BID + Osi | Serious adverse events | 31/101 |
| Savo 600mg QD + Osi | Serious adverse events | 20/44 |
| Savo 300mg BID Monotherapy | Serious adverse events | 8/24 |
| Cross-over | Serious adverse events | 1/14 |
The affected/at-risk counts provide descriptive safety information for the reported groups. They should not be treated as a randomized efficacy comparison, and the ClinicalTrials.gov record does not identify a common randomized denominator structure that would justify collapsing these groups into a single treatment-versus-control safety statistic.
12. Safety Interpretation
A notation such as 62/196 means that 62 subjects were affected among 196 subjects at risk in the specified safety group. It is a descriptive frequency. The denominator is essential because the same affected count can represent very different proportions in populations of different sizes.
Clinical trials commonly define separate populations for efficacy and safety because the scientific questions differ. Efficacy is often linked closely to randomized treatment assignment, while safety is tied to treatment exposure. The SAVANNAH registry data explicitly names a 300 mg QD Safety Analysis Set for one primary endpoint and separately provides serious-adverse-event denominators by safety group.
13. Crossover and Its Statistical Importance
The safety data contains a distinct Cross-over: 1/14 group. The ClinicalTrials.gov record therefore documents a crossover category, but it does not specify the crossover rules, timing, treatment sequence, or the statistical method used to account for crossover.
Crossover can complicate interpretation because treatment received after the original randomized assignment can differ from the treatment initially assigned. For an efficacy analysis, the distinction between assignment and actual treatment received becomes important. An intention-to-treat comparison preserves the randomized treatment contrast, whereas an as-treated analysis can be affected by the reasons and circumstances that led participants to receive different treatment.
14. Randomization and Masking
Randomization
The trial is explicitly classified as RANDOMIZED with a PARALLEL design. Randomization is statistically valuable because it establishes treatment assignment before outcome assessment and, in expectation, balances measured and unmeasured baseline factors across randomized groups.
Quadruple masking
The registry classifies SAVANNAH as QUADRUPLE masked. Masking is especially relevant when outcome assessment involves investigator judgment. In this trial, the primary endpoint is ORR by investigator assessment, making protection against knowledge of treatment assignment an important design feature to understand.
Randomization reduces selection bias
Treatment assignment is determined by the randomized design rather than by a participant's or investigator's treatment preference.
Masking reduces assessment influence
Keeping treatment assignment masked can reduce the opportunity for knowledge of assignment to influence assessment or trial conduct.
15. What the Registry Does and Does Not Establish
| Question | Supported by the ClinicalTrials.gov record? |
|---|---|
| Was the trial randomized? | Yes |
| Was the design parallel? | Yes |
| Was the trial quadruple masked? | Yes |
| Was ORR a registered primary endpoint? | Yes; two registered primary ORR endpoints |
| Were results posted? | Yes |
| Are numerical primary efficacy estimates reported? | No |
| Are confidence intervals posted on ClinicalTrials.gov for primary efficacy results? | No |
| Are p-values posted on ClinicalTrials.gov for primary efficacy results? | No |
| Are formal statistical analyses reported? | No the posted statistical analyses |
| Are serious adverse-event counts reported? | Yes |
This distinction is important for an educational statistical record. A registry can indicate that results have been posted without the ClinicalTrials.gov recordset containing every statistical-analysis object needed to reconstruct a treatment comparison.
16. Statistical Methodology
Binary endpoint analysis
The primary endpoints are classified as binary. The fundamental analysis begins with a count of subjects meeting the confirmed-response definition and a denominator defined by the specified analysis set.
Here, x represents the number of subjects meeting the confirmed OR definition and n represents the number of subjects in the applicable analysis population.
Comparative effect measures
For two randomized groups, a binary endpoint can be summarized using several measures. A risk difference describes the absolute difference between response proportions. A risk ratio describes their ratio. An odds ratio compares the odds of response between groups. The choice should be prespecified because the measures answer different questions.
These are general statistical formulations, not reported SAVANNAH results.
Confidence intervals
A confidence interval around an ORR or comparative effect communicates sampling uncertainty. For proportions, the interval method should be selected with attention to the binary outcome and sample size. For comparative measures, the interval is typically constructed on an appropriate statistical scale and then transformed back to the clinically interpretable scale.
Hypothesis testing
A formal hypothesis test would require a prespecified null hypothesis, alternative hypothesis, test statistic, significance level, and analysis population. The registry-reported SAVANNAH data does not identify these details, so a particular p-value procedure is not attributed to the trial.
RECIST confirmation
The statistical endpoint begins with a clinical and radiologic classification rule. Because the registry requires confirmation at least 4 weeks later and absence of disease progression between assessments, the binary response indicator depends on longitudinal assessment rather than a single imaging measurement.
17. Statistical Concepts That Should Not Be Invented
The ClinicalTrials.gov record does not report several design and analysis features that are sometimes present in phase 2 oncology trials. These should remain separate from the documented trial characteristics.
| Methodological topic | Status in reported SAVANNAH data |
|---|---|
| Non-inferiority margin | Not reported |
| Factorial design | Not reported; the registry identifies the design model as parallel |
| Multiplicity adjustment | Not reported |
| Interim analysis | Not reported |
| Missing-data imputation | Not reported |
| Bayesian methods | Not reported |
| Stratification factors | Not reported |
| Formal comparative effect measure | Not reported |
The absence of these items from the ClinicalTrials.gov record does not establish that they were absent from the underlying protocol or statistical analysis plan. It means only that they are not available as documented inputs for this analysis.
18. Understanding the Primary Endpoint Structure
Same endpoint, different population
Both registered primary endpoints measure investigator-assessed ORR, but they use different named analysis populations.
Binary outcome
The endpoint is classified as binary, so the primary descriptive quantity is a response proportion.
Confirmed response
Response requires RECIST 1.1 criteria plus confirmation by repeat imaging at least 4 weeks later.
Time frame
The registry specifies a time frame beginning at randomization/first dose and continuing to progression or the last evaluable assessment in the absence of progression until the end of the study, or an approximate maximum of 5.6 years..
19. Limitations
- No formal statistical analyses reported: no formal statistical analyses were posted, so no numerical treatment-effect estimates, confidence intervals, or p-values can be reported from the ClinicalTrials.gov record.
- Posted results without registry-reported numerical results: ClinicalTrials.gov identifies results as posted, but in the registry record does not contain the corresponding numerical primary efficacy outcomes.
- Analysis-set sizes are unavailable: the ClinicalTrials.gov record names the Target Population Analysis Set and 300 mg QD Safety Analysis Set but does not provide their sample sizes.
- Randomized arm sizes are unavailable: total enrollment is 367.0 and the design has 2 arms, but the ClinicalTrials.gov record does not provide the randomized sample size of each arm.
- Comparative statistical method is unavailable: the registry data does not identify the formal effect measure or hypothesis-testing procedure used for the primary ORR comparison.
- Crossover details are incomplete: a Cross-over group of 1/14 is reported for serious adverse events, but the ClinicalTrials.gov record does not describe crossover timing, eligibility, treatment sequence, or statistical adjustment.
- Safety denominators differ: serious adverse events are reported across several safety groups with different numbers at risk. These should not automatically be treated as mutually exclusive randomized efficacy groups.
- Unreported design features: the ClinicalTrials.gov record does not establish the use or absence of interim analysis, multiplicity adjustment, missing-data imputation, stratification, non-inferiority logic, or Bayesian methods.
20. Why This Trial Matters Statistically
SAVANNAH is useful as a statistical teaching case because it illustrates an important distinction between trial design information and reported statistical evidence. The registry documents randomization, parallel allocation, quadruple masking, binary primary endpoints, and multiple analysis populations. At the same time, the ClinicalTrials.gov record does not contain the formal statistical analyses needed to quantify the primary treatment comparison.
| Concept | How it appears in SAVANNAH |
|---|---|
| Randomization | The allocation is explicitly randomized. |
| Parallel design | The design model is PARALLEL with 2 arms. |
| Blinding | The registry classifies masking as QUADRUPLE. |
| Binary endpoint | Both registered primary endpoints are classified as binary. |
| Response assessment | ORR is assessed by investigator assessment using RECIST 1.1. |
| Response confirmation | Confirmed response requires repeat imaging at least 4 weeks later and no progression between assessments. |
| Analysis populations | The two primary endpoints use separately named analysis sets. |
| Safety analysis | Serious adverse events are reported as affected subjects among subjects at risk in several safety groups. |
| Crossover | A distinct Cross-over safety group is reported as 1/14. |
| Statistical reporting | Results are posted, but no formal statistical analyses are reported in the registry record. |
21. A Practical Statistical Reading of SAVANNAH
A disciplined analysis of SAVANNAH begins by asking four questions in sequence.
1. Who is analyzed?
Identify the precise analysis population before interpreting an ORR percentage.
2. What counts as response?
Apply the RECIST 1.1 definition, including the required confirmation rule.
3. What is the comparison?
Identify the prespecified effect measure and hypothesis test rather than assuming one from the endpoint name alone.
4. How precise is it?
Read the point estimate together with its confidence interval and the analysis population that generated it.
This sequence prevents several common errors: treating ORR as a time-to-event endpoint, comparing percentages from incompatible denominators, interpreting a p-value as an effect-size measure, or assuming an analysis method that is not documented.
22. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The documented primary endpoints are binary investigator-assessed ORR measures in two named analysis populations. The ClinicalTrials.gov recordset does not contain the numerical comparative estimates or formal statistical tests.
Clinical interpretation
The trial was designed to evaluate objective tumor response in a defined EGFRm+/MET+ NSCLC population following prior osimertinib. The ClinicalTrials.gov record does not establish the magnitude or direction of the observed efficacy result.
23. What a Complete Primary-Efficacy Table Would Contain
If the underlying registry results or statistical report registry-reported the missing numerical analyses, a complete statistical presentation would normally place the following information side by side:
| Component | Purpose |
|---|---|
| Analysis population | Defines who contributes to the endpoint. |
| Responders | Provides the numerator for confirmed ORR. |
| Total evaluable subjects | Provides the denominator. |
| ORR | Describes the observed confirmed-response proportion. |
| Comparative effect estimate | Quantifies the contrast between groups if prespecified. |
| 95% confidence interval | Describes uncertainty around the estimate. |
| P-value | Addresses the prespecified statistical hypothesis. |
| Analysis method | Identifies how the estimate and inference were generated. |
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Statistical Calculators
26. Sources
- ClinicalTrials.gov: SAVANNAH, NCT03778229.
- PubMed: PubMed record for PMID 42460617.
- PubMed: PubMed record for PMID 41308232.
- PubMed: PubMed record for PMID 32027845.
Continue through the Clinical Biostats statistical library
Explore the underlying clinical-trial concepts and statistical tools connected with randomized, masked studies and binary response endpoints.
27. Record Summary
SAVANNAH is a randomized, parallel, quadruple-masked phase 2 trial with 367.0 enrolled participants and 2 study arms. Its two registered primary endpoints are investigator-assessed objective response rate in the Target Population Analysis Set and in the 300 mg QD Safety Analysis Set. Both endpoints are binary and use a RECIST 1.1 confirmed-response definition requiring repeat imaging at least 4 weeks later without disease progression between assessments.
The ClinicalTrials.gov record states that results are posted, but contains no formal statistical analyses and no numerical efficacy estimates, confidence intervals, or p-values.