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
CHRONOS-3 was a double-blind (quadruple-masked), randomized, parallel-group phase 3 trial testing whether adding the PI3K inhibitor copanlisib to rituximab prolonged progression-free survival compared with rituximab plus placebo in patients with relapsed indolent B-cell non-Hodgkin lymphoma. The registry reports the primary endpoint at three separate data cut-offs.
| Feature | CHRONOS-3 |
|---|---|
| Official title | Copanlisib and Rituximab in Relapsed Indolent B-cell Non-Hodgkin's Lymphoma (iNHL) |
| Phase | Phase 3 |
| Condition | Lymphoma, Non-Hodgkin |
| Design | Randomized, parallel-group, quadruple-masked, placebo-controlled |
| Arms | Copanlisib + rituximab; placebo + rituximab |
| Enrollment | 458 |
| Primary endpoint | Progression-free survival based on independent central review |
| Primary analysis | Stratified log-rank test; HR from stratified Cox regression; full analysis set |
| Dates | Start 2015-08-03; first randomization 20-Aug-2015; primary completion 2020-08-31 |
| Status | Completed; results posted |
| Lead sponsor | Bayer (industry) |
| ClinicalTrials.gov | NCT02367040 |
2. Clinical Question
Rituximab is an established anti-CD20 backbone for relapsed indolent B-cell lymphoma. CHRONOS-3 asked whether adding copanlisib to rituximab would delay disease progression or death compared with rituximab alone, using a placebo to keep patients, investigators and assessors blinded to assignment.
Population
Patients with relapsed indolent B-cell non-Hodgkin lymphoma; response assessment included criteria for Waldenström macroglobulinemia.
Intervention
Copanlisib (Aliqopa, BAY80-6946) plus rituximab.
Comparator
Placebo plus rituximab.
Primary question
Does copanlisib plus rituximab prolong progression-free survival, assessed by independent central review, compared with placebo plus rituximab?
3. Trial Design
Copanlisib + rituximab
- Copanlisib (Aliqopa, BAY80-6946)
- Rituximab
Placebo + rituximab
- Placebo matched to copanlisib
- Rituximab
The safety denominators of 307 and 146 show that allocation was unequal, with roughly twice as many patients treated in the copanlisib arm. Unequal allocation trades a small loss of statistical efficiency for more experience with the experimental treatment.
4. Endpoints
Primary endpoint
| Endpoint (registry wording) | Definition | Time frame |
|---|---|---|
| Progression Free Survival (PFS) Based on Independent Central Review | Time from randomization to progressive disease (PD) or death due to any cause, whichever was earlier, according to the Lugano Classification and response criteria in patients affected by Waldenström macroglobulinemia | From first randomization (20-Aug-2015) to data cut-off at primary completion (31-Aug-2020), approximately 5 years; 2-year follow-up after primary completion at 31-Aug-2022, up to 7 years; final analysis at 15-Nov-2024, up to 9 years |
Secondary endpoints with posted analyses
| Endpoint | Type | Method | Cut-offs reported |
|---|---|---|---|
| Objective Response Rate (ORR) | Binary | Stratified Cochran-Mantel-Haenszel | Primary completion; 2-year follow-up |
| Complete Response Rate (CRR) | Binary | Stratified Cochran-Mantel-Haenszel | Primary completion; 2-year follow-up |
| Duration of Response (DOR) | Time-to-event | Stratified log-rank; stratified Cox model | Primary completion; 2-year follow-up |
| Disease Control Rate (DCR) | Binary | Stratified Cochran-Mantel-Haenszel | Primary completion; 2-year follow-up |
| Time to Progression (TTP) | Time-to-event | Stratified log-rank; Cox model | Primary completion; 2-year follow-up |
| Overall Survival (OS) | Time-to-event | Stratified log-rank; Cox model | Final analysis (15-Nov-2024) |
| Time to deterioration in DRS-P of at least 3 points (FLymSI-18) | Time-to-event | Stratified log-rank; Cox model | Primary completion; 2-year follow-up |
| Time to improvement in DRS-P of at least 3 points (FLymSI-18) | Time-to-event | Stratified log-rank; Cox model | Primary completion; 2-year follow-up |
5. Analysis Population
All efficacy analyses used the full analysis set (FAS), defined in the registry as all randomized participants. This is the intention-to-treat principle: patients are analysed in the arm to which they were randomized, regardless of how much study treatment they received. Safety was summarised among treated patients (307 and 146 at risk), which explains why the safety total (453) is slightly lower than the enrollment of 458.
6. Primary Endpoint Results: Progression-Free Survival
PFS was compared with a stratified log-rank test, and the hazard ratio (HR) with its 95% CI was estimated from a stratified Cox regression model. The registry reports this analysis at three data cut-offs.
6a. Primary completion analysis (cut-off 31-Aug-2020)
Hazard ratio for progression or death
Two-sided 95% CI: 0.393–0.688 · P = 0.000002
Copanlisib + rituximab vs placebo + rituximab · Full analysis set
What the estimate means. An HR of 0.520 means that, under the stratified Cox model and over the follow-up to the primary cut-off, the estimated instantaneous rate of progression or death with copanlisib plus rituximab was about 52% of that with placebo plus rituximab, a 48.0% lower estimated hazard.
What it does not mean. It does not mean that 48% fewer patients progressed, that median PFS was roughly doubled, or that each patient's risk fell by 48%. The HR summarises the relative event rate across the whole follow-up; converting it into absolute benefit requires Kaplan-Meier estimates at specific time points, which should be read from the registry's outcome tables rather than inferred from the HR.
Precision. The 95% CI of 0.393–0.688 lies entirely below 1. Even the least favourable value compatible with the data at this cut-off corresponds to a hazard reduction of about 31%.
The p-value. P = 0.000002 indicates that data this extreme would be very unlikely if the two regimens had identical PFS distributions. It does not measure how large the benefit is; a trial with many events can produce a tiny p-value for a modest HR.
Cautions. PFS is sensitive to assessment schedules and to how censoring is handled for patients who start new therapy or miss scans. Independent central review and blinding mitigate, but do not eliminate, these issues. The HR also assumes approximately proportional hazards.
6b. 2-year follow-up analysis (cut-off 31-Aug-2022)
Hazard ratio for progression or death
Two-sided 95% CI: 0.431–0.722 · P = 0.000003
Additional two years of follow-up after primary completion
What the estimate means. With two more years of follow-up, the estimated hazard of progression or death remained lower with copanlisib, by an estimated 44.3%. The interval, 0.431–0.722, is slightly narrower in relative terms than at the primary cut-off, consistent with more events contributing information.
What it does not mean. This is not an independent replication. The same patients contribute to both analyses, with additional events added. The two HRs are highly correlated and should be read as one evolving estimate.
Cautions. Updated analyses after the primary completion date are descriptive. Their p-values do not carry a newly allocated type I error, so they confirm durability of the estimate rather than providing a second confirmatory test.
6c. Final analysis (cut-off 15-Nov-2024)
Hazard ratio for progression or death
Two-sided 95% CI: 0.502–0.823 · P = 0.000199
Up to approximately 9 years from first randomization
What the estimate means. At the final analysis the estimated PFS hazard was 35.7% lower with copanlisib plus rituximab, and the 95% CI of 0.502–0.823 still excludes 1.
Why the HR moved toward 1. The point estimate rose from 0.520 to 0.557 to 0.643 across the three cut-offs. Several explanations are compatible with this pattern: the relative benefit may be largest early and diminish later (non-proportional hazards); late events may accumulate in the copanlisib arm after treatment discontinuation; or the early estimate may have been somewhat optimistic by chance. The registry does not report tests of proportional hazards, so these explanations cannot be distinguished from the HRs alone.
The p-value. P = 0.000199 is larger than earlier p-values despite more events, because the estimated effect is smaller. This illustrates that the p-value reflects both effect size and information, and is not itself a measure of benefit.
Cautions. A single HR summarises the average relative hazard over the whole follow-up. When hazards are not proportional, restricted mean survival time or landmark analyses give a more descriptive picture.
7. Secondary Endpoint Results
Secondary endpoints were reported at the primary completion and 2-year follow-up cut-offs (OS at the final analysis). The registry does not describe a hierarchical testing procedure for these endpoints, so they are best interpreted as supportive.
Response rates (stratified CMH)
| Endpoint | Cut-off | Difference, percentage points (copanlisib − placebo) | 95% CI | P-value |
|---|---|---|---|---|
| Objective response rate | Primary completion | 32.99 | 23.95 to 42.03 | <0.000001 |
| Objective response rate | 2-year follow-up | 30.67 | 21.63 to 39.72 | <0.000001 |
| Complete response rate | Primary completion | 19.27 | 11.57 to 26.96 | <0.000001 |
| Complete response rate | 2-year follow-up | 18.92 | 11.11 to 26.73 | 0.000001 |
| Disease control rate | Primary completion | 4.43 | −2.26 to 11.12 | 0.097339 |
| Disease control rate | 2-year follow-up | 4.43 | −2.26 to 11.12 | 0.097339 |
The ORR and CRR differences are large and precisely estimated, with confidence intervals well above zero at both cut-offs. The DCR difference of 4.43 percentage points has an interval that includes zero. This pattern is informative: disease control counts stable disease as well as responses, and rituximab alone already controls disease in many patients, so the added contribution of copanlisib appears mainly in converting stable disease into objective and complete responses. The DCR results are identical at both cut-offs, which is expected if disease-control status was already determined for all patients by the primary cut-off.
Time-to-event secondary endpoints
| Endpoint | Cut-off | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Time to progression | Primary completion | HR 0.476 | 0.357–0.635 | <.000001 |
| Time to progression | 2-year follow-up | HR 0.505 | 0.387–0.659 | <.000001 |
| Duration of response | Primary completion | 0.700 (effect measure listed as Kaplan-Meier estimates) | 0.481–1.018 | 0.030371 |
| Duration of response | 2-year follow-up | HR 0.761 | 0.547–1.059 | 0.051976 |
| Overall survival | Final analysis | HR 0.972 | 0.691–1.368 | 0.436458 |
| Time to deterioration in DRS-P (≥3 points) | Primary completion | HR 1.060 | 0.843–1.331 | 0.692610 |
| Time to deterioration in DRS-P (≥3 points) | 2-year follow-up | HR 1.047 | 0.841–1.302 | 0.661145 |
| Time to improvement in DRS-P (≥3 points) | Primary completion | HR 0.996 | 0.732–1.355 | 0.510038 |
| Time to improvement in DRS-P (≥3 points) | 2-year follow-up | HR 1.036 | 0.768–1.398 | 0.405970 |
Time to progression shows a slightly stronger effect than PFS (HR 0.476 vs 0.520 at primary completion). TTP counts only progression and censors deaths without progression, so the similarity to PFS suggests that most PFS events were progressions rather than deaths.
Overall survival at the final analysis shows no detectable difference: HR 0.972 with a 95% CI of 0.691–1.368. The interval is wide, compatible with a 31% reduction or a 37% increase in the hazard of death. In an indolent lymphoma with long survival and multiple effective subsequent therapies, OS differences are difficult to detect, and post-progression treatment can dilute any effect of first-line randomization.
Patient-reported symptoms (DRS-P on the FLymSI-18) show HRs near 1 for both deterioration and improvement, with intervals centred on no difference. The PFS advantage was therefore not accompanied by a detectable difference in time to symptom change.
8. Safety: Serious Adverse Events
| Measure | Copanlisib + rituximab | Placebo + rituximab |
|---|---|---|
| Patients with a serious adverse event / at risk | 161 / 307 (52.4%) | 32 / 146 (21.9%) |
The proportion of patients with a serious adverse event was more than twice as high with copanlisib. The PFS benefit therefore came with a substantially greater burden of serious toxicity. The safety comparison should be weighed alongside the absence of an OS or symptom difference, rather than folded into a single benefit metric. Longer exposure in the copanlisib arm, if treatment continued longer before progression, can also raise cumulative event counts; the registry's SAE counts are not adjusted for exposure time.
9. Statistical Methodology
Stratified log-rank test
The primary comparison used a stratified log-rank test, which compares observed and expected events within each randomization stratum and sums the results. Stratification respects the randomization design and prevents imbalance in prognostic factors across strata from distorting the comparison. The registry does not list the stratification factors.
Os and Es are observed and expected events in the copanlisib arm within stratum s, and Vs is the variance. A negative Z indicates fewer events than expected in the copanlisib arm.
Stratified Cox proportional-hazards model
The HR and its 95% CI for PFS and DOR came from a stratified Cox model, which allows a separate baseline hazard in each stratum while estimating a common treatment effect. The registry describes the TTP, OS and DRS-P HRs as based on a Cox regression model (with stratified log-rank tests).
The proportional-hazards assumption means exp(β) is constant over time within strata. The drift of the PFS HR from 0.520 to 0.643 across cut-offs is a reason to examine that assumption.
Kaplan-Meier estimation
PFS, TTP, DOR and OS distributions are summarised with Kaplan-Meier estimates, which account for right-censoring: patients without an event at the data cut-off contribute follow-up time up to their censoring date.
Cochran-Mantel-Haenszel test for response rates
ORR, CRR and DCR were compared with a stratified Cochran-Mantel-Haenszel (CMH) test. CMH combines 2×2 tables across strata, testing for an association between treatment and response while controlling for the stratification factors. The reported effect measure is a difference in rates, in percentage points, with a two-sided 95% CI.
Multiple data cut-offs
The registry reports PFS at primary completion, after 2 additional years of follow-up, and at a final analysis. Only the primary-completion analysis is the natural confirmatory test; the later analyses update the estimate using the same randomized cohort.
10. Statistical Methods Explained
Why did the PFS hazard ratio increase from 0.520 to 0.643 over time?
Each later analysis includes more events and longer follow-up. If the relative benefit of copanlisib is concentrated early, perhaps while patients are on treatment, later events dilute the average HR toward 1. A changing HR across cut-offs is a practical signal that hazards may not be proportional, and that time-specific summaries may describe the effect better than one number.
Why is overall survival not significant when PFS clearly is?
OS in indolent lymphoma requires very long follow-up and is influenced by all subsequent treatments. A PFS benefit can be real without translating into a detectable OS difference, and the OS interval (0.691–1.368) is too wide to rule out either a meaningful benefit or harm.
What does an ORR difference of 32.99 percentage points mean?
It is an absolute difference: out of every 100 patients, about 33 more achieved an objective response with copanlisib plus rituximab than with placebo plus rituximab. The 95% CI of 23.95 to 42.03 means the data are compatible with an absolute gain anywhere from about 24 to 42 percentage points.
Why use a Cochran-Mantel-Haenszel test rather than a simple chi-square?
Randomization was stratified, and the CMH test respects that structure by comparing treatment within strata and pooling the evidence. This gives a valid test even if response rates differ across strata, and is the binary-outcome counterpart of the stratified log-rank test used for PFS.
How can a p-value be below 0.05 when the 95% CI includes 1?
For the primary-completion DOR analysis, P = 0.030371 accompanies a two-sided 95% CI of 0.481–1.018. A two-sided test and a two-sided 95% CI would normally agree. The discrepancy is consistent with a one-sided p-value, which is roughly half the two-sided value. Knowing the sidedness is essential before comparing a p-value with 0.05.
Why is duration of response harder to interpret than PFS?
DOR is measured only in patients who responded, and more patients responded with copanlisib. The responder groups are therefore not randomized comparisons of like with like, and differences in DOR can reflect who responded as much as the effect of treatment.
11. Limitations and Interpretation Issues
- Non-proportional hazards: the PFS HR drifted from 0.520 to 0.643 across cut-offs; a single HR may not describe the time course of benefit well.
- Later cut-offs are descriptive: updated analyses reuse the same patients and do not carry independent confirmatory weight.
- Overall survival: HR 0.972 with a wide CI; subsequent therapies and long survival in indolent lymphoma limit what OS can show.
- Secondary multiplicity: many secondary analyses at two cut-offs are posted without a described hierarchy.
- P-value sidedness: several posted p-values appear to be one-sided and are not labelled as such.
- Responder-based endpoints: DOR compares post-randomization subgroups.
- Safety burden: serious adverse events were reported in 52.4% vs 21.9% of treated patients, unadjusted for exposure time.
- Heterogeneous population: indolent B-cell NHL includes several histologies with different natural histories; the registry does not report histology-specific results.
12. Why This Trial Matters Statistically
CHRONOS-3 is a useful case study in how a time-to-event effect evolves with follow-up, and in the gap that can open between a strong PFS result and neutral overall survival and patient-reported outcomes.
| Concept | How it appears in CHRONOS-3 |
|---|---|
| Time-to-event endpoints | PFS, TTP, DOR, OS and time to symptom deterioration or improvement |
| Kaplan-Meier estimation | Survival distributions with right-censoring at each data cut-off |
| Log-rank test | Stratified log-rank test for the primary and time-to-event secondary endpoints |
| Hazard ratio | PFS HR 0.520, 0.557 and 0.643 across three cut-offs |
| Cochran-Mantel-Haenszel test | Stratified comparison of ORR, CRR and DCR |
| Confidence intervals | Distinguish precise PFS and ORR effects from uncertain OS and DOR effects |
| P-values | Apparent one-sided p-values illustrate why sidedness must be known |
| Intention-to-treat analysis | Full analysis set of all randomized participants |
13. Related Tutorials
Learn more about the methods used in this trial:
14. Related Calculators
15. Sources
- ClinicalTrials.gov: NCT02367040 – Copanlisib and Rituximab in Relapsed Indolent B-cell Non-Hodgkin's Lymphoma (iNHL).
- PubMed: PMID 33848462.
- PubMed: PMID 38226495.
Explore more trial analyses
Each trial page connects its endpoints and methods to statistical tutorials and calculators.
16. Record Summary
CHRONOS-3 randomized 458 patients with relapsed indolent B-cell non-Hodgkin lymphoma to copanlisib or placebo, each with rituximab. PFS by independent central review favoured copanlisib at every cut-off, with HRs of 0.520, 0.557 and 0.643, all with confidence intervals excluding 1. Objective and complete response rates were higher by 32.99 and 19.27 percentage points at primary completion. Overall survival (HR 0.972) and patient-reported symptom endpoints showed no detectable difference, and serious adverse events were more frequent with copanlisib (52.4% vs 21.9%). A balanced reading combines the strength and attenuation of the PFS effect, the neutral OS and symptom results, and the added toxicity.