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
IKEMA was a randomized, parallel-group, single-masked phase 3 study comparing isatuximab, carfilzomib, and dexamethasone (IKd) with carfilzomib and dexamethasone (Kd) in patients with relapsed and/or refractory multiple myeloma. The registry reports 302 enrolled participants, four registered primary endpoint analyses, and eleven posted statistical analyses.
| Feature | IKEMA |
|---|---|
| Trial name | IKEMA |
| Phase | Phase 3 |
| Condition | Plasma Cell Myeloma |
| Population | Relapsed and/or refractory multiple myeloma patients |
| Design | Randomized, parallel |
| Masking | Single |
| Primary purpose | Treatment |
| Enrollment | 302 |
| Primary endpoints registered | 4 |
| Outcome measures posted | 39 |
| Statistical analyses posted | 11 |
| Lead sponsor | Sanofi |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT03275285 |
2. Clinical Question
The central question was whether adding isatuximab to carfilzomib and dexamethasone changed progression-free survival compared with carfilzomib and dexamethasone alone in patients with relapsed and/or refractory multiple myeloma.
Population
Patients with relapsed and/or refractory multiple myeloma, represented in the registry under the condition Plasma Cell Myeloma.
Intervention
Isatuximab SAR650984, carfilzomib, and dexamethasone.
Comparator
Carfilzomib and dexamethasone.
Primary question
Does the isatuximab-containing combination improve progression-free survival relative to carfilzomib plus dexamethasone?
3. Trial Design
Isatuximab combination
- Isatuximab SAR650984
- Carfilzomib
- Dexamethasone
Control combination
- Carfilzomib
- Dexamethasone
The design is particularly useful statistically because the principal efficacy question is a time-to-event comparison. Rather than reducing the outcome to a single binary response at a fixed date, the analysis uses information about the timing of progression or death while accounting for censoring.
4. Endpoints
The registry lists four primary endpoints, all based on progression-free survival as determined by an Independent Response Committee. They represent primary and final analyses of the standard PFS definition and a version in which events occurring more than 8 weeks after the last disease assessment are censored.
| Registered primary endpoint | Registry time frame | Registry definition |
|---|---|---|
| Progression Free Survival (PFS) As Determined by Independent Response Committee (IRC): Primary Analysis | From randomization until the primary analysis data cut-off date of 7 Feb 2020 (median duration of follow-up was 20.73 months) | Time (in months) from randomization to date of 1st documentation of progressive disease (PD)/date of death from any cause, whichever comes 1st. If PD & death are not observed before cut-off date/date of initiation of further anti-myeloma treatment, PFS was censored at date of last valid disease assessment not showing PD performed prior to initiation of further anti-myeloma treatment/cut-off date. |
| Progression Free Survival as Determined by Independent Response Committee: [Event Censored if Occurred >8 Weeks From Last Disease Assessment]: Primary Analysis | From randomization until the primary analysis data cut-off date of 7 Feb 2020 (the median duration of follow-up was 20.73 months) | Time (in months) from randomization to date of 1st PD documentation/death date, whichever is 1st. If PD & death not observed before cut-off date/date of further anti-myeloma treatment initiation, PFS censored at date of last valid disease assessment not showing PD performed prior to initiation of further anti-myeloma treatment/cut-off date. |
| Progression Free Survival as Determined by Independent Response Committee: Final Analysis | From randomization until the final analysis data cut-off date of 14 January 2022 (the median duration of follow-up was 43.96 months) | PFS: time (in months) from randomization to date of first documentation of PD or date of death from any cause, whichever comes first. If PD and death are not observed before analysis cut-off date or date of initiation of further anti-myeloma treatment, PFS was censored at date of last valid disease assessment or analysis cut-off date, whichever comes first. |
| Progression Free Survival as Determined by Independent Response Committee [Event Censored if Occurred >8 Weeks From Last Disease Assessment]: Final Analysis | From randomization until the final analysis data cut-off date of 14 Jan 2022 (the median duration of follow-up was 43.96 months) | Time (in months) from randomization to date of 1st documentation of PD/date of death from any cause, whichever comes 1st. If PD & death are not observed before analysis cut-off date/date of initiation of further anti-myeloma treatment, PFS was censored at date of last valid disease assessment not showing PD/cut-off date, whichever comes 1st. |
The registry also posted secondary analyses covering duration of response, time to progression, time to first response, time to best response, second progression-free survival, and overall response.
5. Statistical Methodology
Primary efficacy population
The primary PFS analyses were performed in the intention-to-treat (ITT) population. The registry describes the primary analysis population as participants who gave informed consent and for whom successful allocation of a randomization was confirmed. The final PFS analyses were also performed on the ITT population.
Kaplan-Meier estimation
The registry explicitly identifies Kaplan-Meier analysis for the primary-analysis PFS endpoint with the event-censoring rule and for several secondary time-to-event endpoints. Kaplan-Meier estimation is appropriate when the outcome is the time until an event and some participants have not experienced that event by the analysis cutoff.
Here di represents events at time ti, while ni represents participants at risk immediately before that time. The key point is that participants contribute information until an event or censoring occurs.
Stratified log-rank testing
The primary PFS analysis used a stratified log-rank test. The registry states that stratification was based on number of prior lines of therapy (1 vs. >1) and revised International Staging System stage (I/II vs III vs not classified), according to the interactive response technology used for randomization.
Hazard ratio
The treatment effect was expressed as a hazard ratio. The registry states that the hazard ratio was stratified on number of prior lines of therapy (1 vs. >1) and R-ISS stage (I or II vs. III vs. not classified) according to the randomization system.
A hazard ratio is a relative time-to-event measure. It is not a median survival difference, an absolute probability difference, or the percentage of patients who benefit.
Cochran-Mantel-Haenszel testing
The posted overall-response analysis used a Cochran-Mantel-Haenszel test. The registry states that the one-sided p-value was stratified based on the randomization factors according to the interactive response technology.
Superiority testing
The posted primary and secondary statistical analyses are identified as superiority hypotheses. This is important because the inferential question is whether the treatment groups differ in the prespecified direction under the superiority framework, rather than whether one regimen is merely no worse than another by a non-inferiority margin.
6. Primary Results: PFS at the Primary Analysis
Standard PFS definition
The first primary analysis compared Kd with IKd using a stratified log-rank test in the ITT population. The reported effect measure was a hazard ratio.
Primary PFS hazard ratio
99% two-sided CI: 0.318–0.889 · P = 0.0007
Hypothesis type: Superiority
The HR of 0.531 means that the estimated instantaneous rate of the PFS event under the fitted analysis was approximately 53.1% of the corresponding rate in the comparator group. Equivalently, 1 − 0.531 = 0.469, so the estimated hazard was approximately 48% lower in relative terms.
The HR does not mean that 48% of participants avoided progression or death, nor does it mean that every participant experienced exactly a 48% reduction in risk. It is a model-based relative comparison of event rates over time.
The two-sided 99% confidence interval of 0.318–0.889 quantifies statistical uncertainty around the estimated HR under the analysis framework. Its width matters: the estimate is not known with perfect precision. The interval also remains below 1, which is consistent with the reported superiority result.
The P = 0.0007 value addresses evidence against the null hypothesis under the prespecified testing framework; it does not measure the magnitude or clinical importance of the treatment effect. The registry additionally states that the one-sided nominal significance level to declare overwhelming efficacy at the primary analysis was 0.005, within an overall one-sided type I error of 2.5% controlled using an alpha-spending function.
Because PFS is a time-to-event endpoint, censoring rules and the validity of the proportional-hazards interpretation are important. The reported HR should therefore be interpreted together with the endpoint definition, analysis population, stratification, and confidence interval.
PFS with events censored if occurring >8 weeks from the last disease assessment
A second primary-analysis record used the alternative event-censoring definition specified by the registry and again analyzed the comparison with a stratified log-rank test.
Alternative primary PFS hazard ratio
99.2% two-sided CI: 0.317–0.948 · P = 0.0016
Hypothesis type: Superiority
The HR of 0.548 corresponds to an estimated hazard approximately 54.8% of the comparator hazard, or approximately a 45% lower estimated hazard from the simple transformation 1 − 0.548.
Again, this is not an absolute risk reduction and does not imply that a fixed percentage of patients necessarily avoided progression or death. It summarizes the relative event-rate comparison under the specified survival-analysis framework.
The 99.2% CI of 0.317–0.948 expresses uncertainty around the HR. The interval is fairly broad relative to the point estimate, which is a reminder that an estimated treatment effect is not a single known quantity.
The reported P = 0.0016 concerns statistical evidence under the specified testing framework. A p-value is not an effect-size measure: a smaller p-value does not mean a larger treatment effect, and a larger p-value does not establish that treatment effects are absent.
The special censoring rule is important. Changing which events are censored can change the statistical dataset and therefore the resulting estimate. The fact that this analysis remains directionally similar to the standard primary PFS analysis provides a useful sensitivity perspective, but the two analyses should not be treated as independent clinical trials.
7. Primary Results: Final PFS Analysis
Final PFS analysis
The final analysis used data through 14 January 2022. The registry reports the analysis in the ITT population and identifies the hazard ratio as stratified by number of prior lines of therapy and R-ISS stage. The statistical method is not reported for this final analysis.
Final PFS hazard ratio
95.4% two-sided CI: 0.418–0.792
Hypothesis type: Superiority
The final HR of 0.576 corresponds to an estimated event hazard approximately 57.6% of that in the comparator group. Expressed as the complementary relative quantity, 1 − 0.576 = 0.424, or approximately a 42% lower estimated hazard.
The 95.4% CI of 0.418–0.792 provides the uncertainty interval reported by the registry. It is important not to silently convert this to a conventional 95% confidence interval: the confidence level is specifically 95.4% in the posted analysis.
No p-value is reported in the statistical-analysis record posted on ClinicalTrials.gov for this final endpoint. Therefore, the final result should be reported using the HR and its confidence interval rather than assigning an unreported significance value.
The final estimate is based on the ITT population and retains the registry's stratification factors. As with other HR analyses, its interpretation depends on the time-to-event framework, censoring, and the proportional-hazards interpretation of a single relative hazard measure.
Final PFS with events censored if occurring >8 weeks after the last disease assessment
The fourth primary analysis used the final data cutoff of 14 January 2022 and the alternative censoring rule. The registry reports Kaplan-Meier analysis in the ITT population and a stratified hazard ratio.
Final alternative PFS hazard ratio
95.4% two-sided CI: 0.424–0.832
Hypothesis type: Superiority
The HR of 0.594 indicates an estimated event hazard approximately 59.4% of the comparator hazard, corresponding to approximately a 41% lower estimated hazard using 1 − 0.594.
The 95.4% CI of 0.424–0.832 describes uncertainty around the estimated relative effect. The confidence interval is more informative about precision than the point estimate alone because it shows a range of effect estimates compatible with the statistical model and data.
No p-value is reported for this final analysis in the ClinicalTrials.gov record. It should therefore not be reconstructed from the HR and confidence interval.
The alternative censoring rule is a methodological sensitivity perspective rather than a new randomized comparison. Its purpose is to define how late events relative to the last disease assessment are handled. Interpretation should remain tied to that exact endpoint definition.
8. Summary of the Four Primary PFS Analyses
| Primary endpoint analysis | HR | Confidence interval | P-value | Analysis population |
|---|---|---|---|---|
| PFS, primary analysis | 0.531 | 99% CI 0.318–0.889 | 0.0007 | ITT |
| PFS, >8-week event-censoring rule, primary analysis | 0.548 | 99.2% CI 0.317–0.948 | 0.0016 | ITT |
| PFS, final analysis | 0.576 | 95.4% CI 0.418–0.792 | Not reported | ITT |
| PFS, >8-week event-censoring rule, final analysis | 0.594 | 95.4% CI 0.424–0.832 | Not reported | ITT |
The four posted primary analyses show a consistent direction of the estimated PFS treatment effect: all four hazard ratios are below 1. The point estimates range from 0.531 to 0.594. The registry supplies formal p-values for the two primary-analysis records but not for the two final-analysis records in the ClinicalTrials.gov record.
This pattern is statistically informative without requiring the estimates to be treated as four independent confirmatory tests. Two records represent different PFS definitions at the primary analysis, while the other two represent corresponding final analyses. They are therefore related analyses of the same randomized comparison.
9. Secondary Endpoint Results
Duration of Response
Duration of Response was analyzed in the subset of participants with PR or better, defined as responders, within the ITT population. The registry reports a stratified hazard ratio.
Duration of Response
95% two-sided CI: 0.269–0.672
Analysis: responders in the ITT population; stratified by number of prior lines of therapy and R-ISS stage.
Because DOR begins among participants who achieved PR or better, its analysis population differs conceptually from an ITT analysis of the primary PFS endpoint. This distinction matters: DOR describes the durability of response among responders rather than the treatment effect from randomization across all randomized participants.
Time to Progression
Time to Progression
95% two-sided CI: 0.324–0.757
Primary analysis data cutoff: 7 Feb 2020
The TTP analysis was performed in the ITT population using Kaplan-Meier analysis. TTP is distinct from PFS because the event definition centers on progression rather than combining progression and death. The statistical interpretation of the HR remains a relative time-to-event comparison.
Time to First Response
Time to First Response
95% two-sided CI: 0.888–1.471
Primary analysis data cutoff: 7 Feb 2020
The point estimate above 1 contrasts with the PFS hazard ratios below 1. That does not mean the analyses are contradictory: they concern different events. The time-to-first-response analysis asks about the timing of achieving a response, while PFS combines progression and death under the registered endpoint definition.
Time to Best Response
Time to Best Response
95% two-sided CI: 0.740–1.233
Primary analysis data cutoff: 7 Feb 2020
The confidence interval for time to best response includes 1. The appropriate interpretation is that the posted estimate is uncertain with respect to a relative hazard difference around the null value; it should not be converted into a claim that the treatments are equivalent.
Second Progression-Free Survival: Final Analysis
PFS2 at 14 January 2022
95% two-sided CI: 0.496–0.941
ITT population; Kaplan-Meier method reported.
PFS2 is a later time-to-event endpoint measured from randomization. The registry describes the analysis as stratified by number of prior lines of therapy and R-ISS stage. No additional significance level or multiplicity margin is specified in the ClinicalTrials.gov record for this analysis.
Second Progression-Free Survival: Overall Survival Analysis
PFS2 at 7 February 2023
95% two-sided CI: 0.491–0.895
ITT population; Kaplan-Meier method reported.
This later PFS2 analysis uses a different data cutoff from the 14 January 2022 PFS2 analysis. The two estimates should therefore be viewed as separate timepoints rather than averaged or combined.
Overall Response
Cochran-Mantel-Haenszel test
One-sided p-value; ITT population; stratified by randomization factors according to IRT.
The registry identifies overall response as a binary endpoint and reports a Cochran-Mantel-Haenszel analysis. The ClinicalTrials.gov record does not provide the response percentages or a corresponding effect estimate, so the formal result is reported here as the posted p-value without reconstructing an unreported response difference.
10. Secondary Results at a Glance
| Secondary endpoint | Effect measure | Estimate | Confidence interval | P-value |
|---|---|---|---|---|
| Duration of Response | Stratified HR | 0.425 | 95% CI 0.269–0.672 | Not reported |
| Time to Progression | Stratified HR | 0.495 | 95% CI 0.324–0.757 | Not reported |
| Time to First Response | Stratified HR | 1.143 | 95% CI 0.888–1.471 | Not reported |
| Time to Best Response | Stratified HR | 0.955 | 95% CI 0.740–1.233 | Not reported |
| PFS2, final analysis | Stratified HR | 0.683 | 95% CI 0.496–0.941 | Not reported |
| PFS2, overall survival analysis | Stratified HR | 0.663 | 95% CI 0.491–0.895 | Not reported |
| Overall Response | Cochran-Mantel-Haenszel test | Not reported | Not reported | 0.1930, one-sided |
The secondary endpoints illustrate why the analysis target must always be identified before interpreting a hazard ratio. A hazard ratio for duration of response is not directly comparable with a hazard ratio for PFS because the populations and event definitions differ.
11. Multiplicity and Type I Error
The primary PFS analysis includes explicit alpha-spending language. The registry states that the nominal significance levels at the primary analysis were determined using an alpha-spending function to control the overall one-sided type I error at 2.5%.
Primary PFS boundary
The one-sided nominal significance level to declare overwhelming efficacy at the primary analysis was 0.005 for the analysis based on 103 PFS events.
Alternative PFS analysis
For the alternative primary PFS analysis, the registry reports a one-sided nominal significance level of 0.004.
Closed testing
A closed test procedure was used to control the Type I error rate from the primary efficacy endpoints sequentially through the secondary efficacy endpoints.
Why the hierarchy matters
Sequential testing means that the interpretation of later efficacy endpoints depends on whether the prespecified earlier testing conditions were satisfied.
This is an important distinction between an individual p-value and a trial-wide inferential framework. A p-value is calculated for a particular statistical test, while multiplicity procedures determine how repeated or sequential tests should be interpreted without exceeding the prespecified type I error budget.
12. Statistical Methods Explained
Why was a stratified log-rank test used?
A log-rank test is designed for comparing time-to-event distributions between randomized groups. The stratified version allows the comparison to account for prespecified factors used in the randomization framework. In IKEMA, the registry identifies number of prior lines of therapy and R-ISS stage as the stratification factors for the PFS hazard-ratio analysis.
What does an HR of 0.531 mean?
It means the estimated instantaneous event rate in the IKd group was approximately 53.1% of that in the Kd group under the fitted time-to-event comparison. The complementary quantity, 1 − 0.531, is approximately 0.469, or a 48% lower estimated hazard. This is a relative model-based measure, not an absolute percentage of patients who benefit.
Why does the confidence interval matter?
The point estimate is only one estimate of the underlying treatment effect. The confidence interval communicates statistical precision around that estimate. For example, the first primary PFS analysis reports a 99% CI of 0.318–0.889 around an HR of 0.531. A reader should therefore consider both the location and width of the interval rather than focusing only on the HR.
Why is ITT analysis important?
The primary PFS analyses were conducted in the ITT population. An ITT comparison retains participants according to their randomized treatment assignment rather than selectively analyzing only those who remained on treatment or complied with treatment. This preserves the comparison created by randomization and is particularly important when interpreting a randomized superiority trial.
Why is the p-value not an effect size?
The p-value describes the compatibility of the observed data with a specified null hypothesis under the statistical testing framework. It does not tell us how large the treatment effect is. The HR describes relative effect magnitude, while the confidence interval communicates uncertainty. These quantities answer different statistical questions.
Why should the alternative PFS censoring rule be interpreted separately?
The alternative endpoint changes how certain late progression or death events are handled relative to the last disease assessment. Because censoring rules determine which observations contribute events and which contribute follow-up time without an event, changing the rule can change the estimated survival distribution and treatment effect. The resulting HR is therefore tied to that specific endpoint definition.
13. Hazard Ratios, Censoring, and Proportional Hazards
The four primary PFS analyses all use hazard ratios. A hazard is an instantaneous event rate conditional on remaining event-free immediately before a given time. The HR compares those rates between treatment groups under the model used for the analysis.
An HR below 1 indicates a lower estimated event hazard in the IKd group relative to Kd. For the primary estimate of 0.531, the simple relative-hazard interpretation corresponds to approximately a 48% lower estimated hazard.
It does not directly tell you the difference in median PFS, the difference in the proportion progressing by a particular date, or the probability that an individual patient will benefit. Those require different summaries.
A single Cox-model HR is easiest to interpret as a stable relative hazard when the proportional-hazards assumption is reasonable. The ClinicalTrials.gov record does not provide a diagnostic assessment of that assumption, so the HR should be understood as the reported model-based summary rather than as a universal constant risk ratio for every point in time.
For PFS, censoring is also central. The registry definitions specify that participants without PD or death before the relevant cutoff or initiation of further anti-myeloma treatment can be censored at an appropriate last disease assessment or cutoff date. This means the analysis uses both observed events and follow-up information from participants who have not yet experienced an event.
14. Stratification in IKEMA
The registry identifies two stratification factors for the PFS hazard-ratio analyses: number of prior lines of therapy and R-ISS stage. Specifically, the strata were number of prior lines of therapy (1 vs. >1) and R-ISS stage (I or II vs. III vs. not classified).
| Stratification factor | Categories reported | Statistical role |
|---|---|---|
| Number of prior lines of therapy | 1 vs. >1 | Stratification factor for the PFS hazard-ratio analysis |
| R-ISS stage | I or II vs. III vs. not classified | Stratification factor for the PFS hazard-ratio analysis |
Stratification does not mean that separate independent trials were conducted within each category. Rather, the statistical comparison incorporates the prespecified strata when estimating or testing the treatment effect. This can improve alignment between the analysis and the factors used in the randomization process.
15. Serious Adverse Events
The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by the number at risk. These are reported here exactly in that format.
| Treatment arm | Serious adverse events |
|---|---|
| Carfilzomib + Dexamethasone (Kd) | 74/122 |
| Isatuximab + Carfilzomib + Dexamethasone | 129/177 |
The denominators in this safety summary should not be substituted for randomized arm sizes. The trial-level enrollment is 302, while the serious-adverse-event summary is specifically reported as affected/at risk with denominators of 122 and 177.
16. Primary and Secondary Analysis Architecture
| Analysis family | Endpoint examples | Population / method | Primary statistical concept |
|---|---|---|---|
| Primary PFS | PFS, standard definition | ITT; stratified log-rank | Superiority; HR |
| Alternative primary PFS | PFS with >8-week event-censoring rule | ITT; stratified log-rank | Superiority; HR |
| Final PFS | PFS at 14 January 2022 cutoff | ITT; stratified HR | Superiority; HR |
| Secondary time-to-event | DOR, TTP, time to response, PFS2 | Primarily ITT; Kaplan-Meier where reported | Stratified HR |
| Binary response | Overall Response | ITT | Cochran-Mantel-Haenszel test |
This architecture illustrates an important principle in clinical-trial statistics: the endpoint definition, analysis population, statistical model, and testing framework are linked. A numerical result cannot be interpreted correctly without knowing what event was analyzed, who was included, how censoring was handled, and what hypothesis was being tested.
17. Trial Timeline
Study start
The registry lists 25 October 2017 as the trial start date.
Primary PFS analysis cutoff
The first two registered primary PFS analyses use a 7 February 2020 data cutoff. The registry reports median follow-up durations of 20.73 months and 20.7 months in the corresponding records.
Final PFS analysis cutoff
The final standard PFS and alternative-censoring analyses use a 14 January 2022 data cutoff. The corresponding registry records report median follow-up values beginning with 4 and 43.96 months, respectively, in the registry-reported extract.
Later PFS2 analysis
A posted secondary PFS2 analysis uses an overall survival analysis data cutoff of 7 February 2023.
Trial status
The ClinicalTrials.gov record identifies the IKEMA trial as completed.
18. How to Read the Primary PFS Results Together
The primary-analysis HRs are 0.531 and 0.548, while the corresponding final-analysis HRs are 0.576 and 0.594. The direction is therefore consistent across the four registered primary PFS analyses, even though the exact point estimate changes with the data cutoff and endpoint censoring definition.
Consistency of direction
All four reported HRs are below 1, indicating a lower estimated PFS event hazard for IKd relative to Kd under each specified analysis.
Point estimates are not identical
The estimates differ because the analyses use different data cutoffs and, for two records, a different event-censoring rule.
Confidence levels differ
The primary-analysis records use 99% and 99.2% two-sided confidence intervals, while the final-analysis records use 95.4% two-sided confidence intervals.
P-values are not available for every analysis
The ClinicalTrials.gov record provides p-values for the primary analyses but not for the two final PFS analyses.
A useful statistical reading therefore focuses on the entire pattern rather than treating the first HR as the only relevant number. The repeated direction across the standard and alternative PFS definitions is part of the evidence structure, while the differing confidence levels and data cutoffs must remain explicit.
19. Limitations
- Registry-level reporting: the ClinicalTrials.gov record contains selected posted analyses rather than every detail of the underlying statistical analysis plan.
- Incomplete p-value reporting across analyses: formal p-values are not posted on ClinicalTrials.gov for the final PFS analyses or most secondary endpoints.
- Different endpoint definitions: the four primary records include both the standard PFS definition and an alternative rule censoring events occurring more than 8 weeks from the last disease assessment.
- Different confidence levels: the primary analyses use 99% and 99.2% two-sided confidence intervals, while the final analyses use 95.4% two-sided intervals.
- Hazard-ratio interpretation: a single HR summarizes a time-to-event comparison and should not automatically be interpreted as a constant relative risk at every time point.
- Analysis-population differences: DOR is analyzed among responders with PR or better, whereas the primary PFS analyses use the ITT population. These estimates therefore describe different populations.
- Multiplicity: the trial used alpha spending and a closed testing procedure, so individual p-values should be interpreted within the prespecified testing hierarchy.
- Safety denominators: the serious-adverse-event summary uses affected/at-risk denominators of 122 and 177, which should not be confused with total trial enrollment.
20. Why This Trial Matters Statistically
IKEMA is a useful teaching case because its posted registry results bring together several core concepts in survival analysis and randomized clinical-trial inference.
| Concept | How it appears in IKEMA |
|---|---|
| Randomization | The trial uses randomized allocation with two parallel treatment arms. |
| Intention-to-treat analysis | The primary PFS analyses were conducted in the ITT population. |
| Kaplan-Meier estimation | Used for the primary alternative PFS analysis and several secondary time-to-event analyses. |
| Hazard ratio | Primary and secondary time-to-event effects are expressed using HRs. |
| Confidence intervals | Primary analyses report 99% or 99.2% two-sided CIs; final analyses report 95.4% two-sided CIs. |
| Stratified analysis | PFS hazard ratios were stratified by number of prior lines of therapy and R-ISS stage. |
| Stratified log-rank test | Used for the primary PFS comparison. |
| Cochran-Mantel-Haenszel test | Used for the posted overall-response analysis. |
| Alpha spending | The primary analysis used an alpha-spending function to control the overall one-sided type I error at 2.5%. |
| Closed testing | The registry states that a closed test procedure controlled Type I error sequentially through secondary efficacy endpoints. |
| Censoring rules | The registered PFS definitions specify how participants without PD or death are censored and provide an alternative >8-week event-censoring rule. |
| Longitudinal analysis | The registry contains primary and final PFS analyses plus later PFS2 analysis. |
The strongest statistical lesson is that a treatment-effect estimate is inseparable from its endpoint definition and analysis framework. In IKEMA, even closely related PFS analyses can have different hazard-ratio estimates because the data cutoff and censoring rules differ.
21. Related Statistical Tutorials
Learn more about the methods used in this trial:
22. Related Statistical Calculators
23. Sources
- ClinicalTrials.gov: IKEMA, NCT03275285.
- PubMed: PMID 34097854.
- PubMed: PMID 36372355.
- PubMed: PMID 36239134.
- PubMed: PMID 42363804.
- PubMed: PMID 39067465.
Continue through the Clinical Biostats statistical library
Connect the endpoints and methods used in IKEMA to deeper statistical tutorials, calculators, and additional clinical-trial analyses.
24. Record Summary
IKEMA provides a detailed example of randomized time-to-event analysis in a phase 3 trial. The registry reports four primary PFS analyses, all using the ITT population and all producing hazard ratios below 1. The two primary-analysis records report HRs of 0.531 and 0.548, while the two final-analysis records report HRs of 0.576 and 0.594. The corresponding confidence intervals, data cutoffs, and censoring rules differ and therefore need to remain attached to each estimate.
The trial also illustrates the importance of statistical hierarchy. The primary PFS analysis used stratified log-rank testing and alpha spending, while the posted secondary analyses include stratified hazard ratios for several time-to-event endpoints and a Cochran-Mantel-Haenszel analysis for overall response. These results should be interpreted within their respective populations, endpoint definitions, and multiplicity framework rather than treated as interchangeable measures of treatment effect.