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
RTOG 0724 was a randomized, parallel-group, phase 3 trial in cervical cancer. The registered primary endpoint was disease-free survival, analyzed as a time-to-event outcome using Kaplan-Meier estimation and a log-rank comparison between the two treatment groups.
| Feature | RTOG 0724 |
|---|---|
| Trial name | RTOG 0724 |
| Phase | Phase 3 |
| Condition | Cervical Cancer |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 236 |
| Primary endpoint | Disease-free Survival (Percentage of Participants Alive Without Disease) |
| Primary endpoint type | Time-to-event |
| Primary analysis method | Log-rank test |
| Effect measure | Hazard ratio |
| Hypothesis type | Superiority |
| ClinicalTrials.gov | NCT00980954 |
2. Clinical Question
The central question was whether adding 4 additional cycles of carboplatin and paclitaxel after concurrent radiation and weekly cisplatin could improve disease-free survival compared with cisplatin and radiation therapy alone in patients with high-risk early-stage cervical cancer after radical hysterectomy.
Population
Patients with high-risk early-stage cervical cancer after radical hysterectomy. The registered primary hypothesis specifically describes patients with cervical carcinoma with positive nodes and/or positive margins after a radical hysterectomy.
Intervention
Cisplatin and radiation therapy followed by additional carboplatin and paclitaxel.
Comparator
Cisplatin and radiation therapy without the additional carboplatin/paclitaxel treatment.
Primary question
Does the additional chemotherapy improve disease-free survival relative to cisplatin/radiation therapy alone?
3. Trial Design
Cisplatin / Radiation Therapy
- Cisplatin
- Radiation therapy
- Radiation could be intensity-modulated radiation therapy or standard external beam radiation therapy
- Optional brachytherapy boost
Cisplatin / Radiation Therapy + Carboplatin/Paclitaxel
- Cisplatin
- Radiation therapy
- Additional carboplatin and paclitaxel
- Radiation could be intensity-modulated radiation therapy or standard external beam radiation therapy
- Optional brachytherapy boost
The registered interventions therefore create a clinically interpretable randomized comparison: both groups receive the cisplatin/radiation backbone, while Arm II adds carboplatin and paclitaxel. Statistically, this isolates the incremental treatment question as far as the randomized design permits.
4. Trial Timing and Status
Trial start
The registered trial start was 2009-09.
Primary completion
The registered primary completion date was 2023-12-14.
Registry status
The trial is listed as COMPLETED, with 236 participants enrolled.
5. Primary Endpoint
| Endpoint | Registry definition / time frame | Analysis |
|---|---|---|
| Disease-free Survival (Percentage of Participants Alive Without Disease) | From randomization to first failure (local, regional, or distant metastases failure or death due to any cause) or last follow-up (censored). | Kaplan-Meier estimation; log-rank comparison; hazard ratio |
The registry defines disease-free survival as the time from randomization to the first DFS failure, where failure can be local, regional, or distant metastatic failure or death from any cause. Participants without a qualifying event by the end of observation are censored at their last follow-up.
6. Secondary Endpoint With a Reported Analysis
| Endpoint | Time frame | Analysis |
|---|---|---|
| Overall Survival (Percentage of Participants Alive) | From randomization to death or last follow-up. Maximum follow-up time at time of analysis was 12.8 years. The 2- and 4-year survival rates are part of the registered outcome measure. | Log-rank test; hazard ratio |
The overall-survival endpoint uses the same basic time origin as DFS—randomization—but has a different event definition: death. A participant who has not died by the last follow-up contributes observed survival time and is censored at that point.
7. Statistical Methodology
Kaplan-Meier estimation
The registered DFS definition states that disease-free survival is estimated by the Kaplan-Meier method. This is the standard nonparametric framework for estimating a survival or event-free probability over time when some participants are censored before experiencing the event.
Here, di is the number of events at time ti, while ni is the number of participants at risk immediately before that event time.
Log-rank test
The registry reports a log-rank test for comparing the distribution of DFS between the two randomized arms. The same method is reported for overall survival.
The log-rank test is fundamentally a comparison of event timing across groups. At each observed event time, it compares the number of events actually observed in each group with the number expected under the null hypothesis of no difference in the underlying time-to-event distributions.
Hazard ratio
The reported effect measure is the hazard ratio (HR). A hazard ratio below 1 for Arm II relative to Arm I would indicate a lower estimated instantaneous event rate in Arm II under the model used for the analysis. Conversely, an HR above 1 indicates a higher estimated instantaneous event rate.
An HR is not a probability, not a percentage of patients cured, and not a direct measure of absolute risk. Its interpretation concerns relative event rates over time under the assumptions of the time-to-event analysis.
Analysis population
The posted statistical analyses identify eligible participants as the analysis population. This is important because the reported effect estimates are not described as being calculated from every enrolled participant regardless of eligibility.
Superiority framework
The statistical analyses are identified as superiority analyses. The question is therefore whether the additional chemotherapy produces evidence of a difference favoring the intervention rather than whether it can be shown to be no worse than a prespecified non-inferiority margin.
8. Sample-Size Hypothesis and Statistical Design
The registered analysis notes that the sample-size calculation was based on the hypothesis that 4 additional cycles of carboplatin and paclitaxel following concurrent radiation and weekly cisplatin would increase 4-year DFS from 80% to 90% for patients with cervical carcinoma with positive nodes and/or positive margins after a radical hysterectomy.
Targeted effect
The planning hypothesis was an increase in 4-year DFS from 80% to 90%.
Alpha
The analysis notes specify a one-sided alpha of 0.05.
Power
The analysis notes specify statistical power of 80% for the primary hypothesis.
Accrual assumption
The planning notes describe 5.5 years of accrual with 4 years of follow-up in the sample-size calculation.
These are design assumptions, not observations from the completed analysis. A planning assumption that the additional chemotherapy would raise 4-year DFS from 80% to 90% should not be read as evidence that either rate was observed.
9. Results: Disease-Free Survival
Disease-free survival was the registered primary endpoint. The formal analysis used the log-rank test in eligible participants, comparing Arm I, cisplatin/radiation therapy, with Arm II, cisplatin/radiation therapy plus carboplatin/paclitaxel.
Primary DFS hazard ratio
90% two-sided CI: 0.65–1.68 · P = 0.56
Log-rank test · Superiority hypothesis · Eligible participants
| Primary endpoint | Arm I | Arm II | Effect estimate |
|---|---|---|---|
| Disease-free Survival | Cisplatin/Radiation Therapy | Cisplatin/Radiation Therapy + Carboplatin/Paclitaxel | HR 1.05 90% CI 0.65–1.68 P = 0.56 |
The estimated DFS hazard ratio of 1.05 compares the instantaneous rate of a DFS failure in Arm I with the corresponding rate in Arm II as reported by the registry. An HR of 1.05 is close to 1, so the point estimate does not indicate a large relative separation between the randomized groups in the direction of Arm II.
The estimate does not mean that exactly 5% more participants experienced a DFS event, and it does not translate directly into an absolute difference in the percentage of participants who remained disease-free. A hazard ratio summarizes relative event rates over time rather than a simple difference in cumulative proportions.
The 90% two-sided confidence interval of 0.65–1.68 is fairly broad relative to the point estimate. It includes 1, so values representing a lower or higher hazard are compatible with the interval. The interval therefore communicates substantial uncertainty around the estimated treatment effect.
The P = 0.56 does not measure the size of the treatment effect or the probability that the treatment works. It measures the compatibility of the observed test statistic with the null hypothesis under the specified testing framework. The p-value should therefore be read alongside the hazard ratio and its confidence interval.
Because DFS is a time-to-event endpoint, interpretation also depends on censoring and on the assumptions underlying the hazard-ratio representation. A single HR is most straightforward when the relative hazards are reasonably described by a common proportional effect over time. The ClinicalTrials.gov record does not provide additional diagnostics for that assumption.
10. Results: Overall Survival
Overall survival was a reported secondary endpoint. The registry defines it from randomization to death or last follow-up and reports a maximum follow-up time of 12.8 years at the time of analysis.
Overall-survival hazard ratio
90% two-sided CI: 0.49–1.69 · P = 0.40
Log-rank test · Superiority hypothesis · Eligible participants
| Secondary endpoint | Arm I | Arm II | Effect estimate |
|---|---|---|---|
| Overall Survival | Cisplatin/Radiation Therapy | Cisplatin/Radiation Therapy + Carboplatin/Paclitaxel | HR 0.91 90% CI 0.49–1.69 P = 0.40 |
The overall-survival hazard ratio of 0.91 is below 1. Under the reported time-to-event comparison, the point estimate corresponds to a lower estimated hazard of death in Arm II relative to Arm I when expressed in that direction. Equivalently, the estimate is relatively close to 1 and therefore does not represent a large estimated separation in mortality hazards.
The estimate does not mean that mortality was reduced by exactly 9%, nor does it mean that 91% of participants survived. Hazard ratios are relative time-to-event measures, not absolute survival probabilities.
The 90% two-sided confidence interval of 0.49–1.69 spans 1 and is wide. It therefore permits a range of relative effects, including values below and above the null value. The precision of the estimate is consequently limited by the information summarized in the registry result.
The P = 0.40 is not a measure of clinical importance or effect size. A p-value describes evidence against the null hypothesis within the specified test framework; it does not state how large or small the treatment effect is.
The registry identifies the endpoint as a time-to-event analysis and reports the log-rank test. Interpretation must therefore account for censoring and the time dimension rather than treating overall survival as a simple binary outcome at a single date.
11. Comparing the Two Reported Time-to-Event Results
| Endpoint | HR | 90% two-sided CI | P-value | Role |
|---|---|---|---|---|
| Disease-free Survival | 1.05 | 0.65–1.68 | 0.56 | Primary |
| Overall Survival | 0.91 | 0.49–1.69 | 0.40 | Secondary |
The two reported hazard ratios answer different endpoint questions. DFS counts the first local, regional, or distant metastases failure or death, whereas overall survival counts death. It is therefore possible for the two endpoints to have different numerical hazard ratios even though they arise from the same randomized treatment comparison.
The confidence intervals for both reported effects include 1. That is the null value for a hazard ratio. The registry-reported p-values are also not small under their respective testing frameworks. The statistically important point is not that one HR is "better" than the other, but that each estimate must be interpreted in relation to its endpoint definition, uncertainty interval, and prespecified role in the trial.
12. Statistical Methods Explained
Why was a Kaplan-Meier method used?
DFS and OS are measured as times from randomization until an event or censoring. Kaplan-Meier estimation is designed for this structure because participants can have different amounts of observed follow-up. Someone who has not experienced an event by the last follow-up is not treated as if the event never could occur; instead, that participant is censored at the observed follow-up time.
Why was a log-rank test used?
The log-rank test compares the timing of events between randomized groups across follow-up. It is therefore aligned with the time-to-event nature of DFS and OS. Rather than comparing only one fixed-time percentage, it uses the sequence of observed events and the numbers at risk over time.
What does an HR of 1.05 mean for DFS?
An HR of 1.05 is a relative time-to-event estimate. With the comparison as reported for DFS, it is close to the null value of 1. It does not mean a 5-percentage-point difference in DFS, and it does not identify the proportion of individual participants who benefited or were harmed.
What does the 90% confidence interval mean?
The reported 90% two-sided interval describes uncertainty around the hazard-ratio estimate under the statistical framework used to generate it. For DFS, the interval is 0.65–1.68; for OS, it is 0.49–1.69. Because both include 1, neither interval excludes the null hazard ratio.
Why does the p-value not measure effect size?
A p-value summarizes how compatible the observed test statistic is with a null hypothesis under the specified testing framework. It depends on both the estimated effect and the amount of statistical information available. The hazard ratio and confidence interval are needed to understand the estimated magnitude and precision of the treatment comparison.
Why is the primary hypothesis described as one-sided while the confidence interval is two-sided?
The registry's sample-size notes specify a one-sided alpha of 0.05 for the primary hypothesis. The posted DFS and OS estimates, however, are accompanied by 90% two-sided confidence intervals. A confidence interval and a hypothesis-test alpha allocation are related but are not interchangeable descriptions of the entire statistical design.
13. Interim Analysis and Alpha Spending
The primary DFS analysis notes explicitly identify interim analysis / alpha spending as a statistical concept in the analysis. The same notes specify a one-sided alpha of 0.05 and statistical power of 80% for the primary hypothesis.
Why interim analysis matters
An interim analysis permits efficacy information to be examined before the trial's final information set. Without an appropriate error-control strategy, repeated testing can increase the overall probability of a false-positive conclusion.
Why alpha spending matters
Alpha spending is a framework for controlling the cumulative type I error across interim looks. The ClinicalTrials.gov record identifies alpha spending as part of the analysis but does not provide enough detail here to reconstruct a specific spending function or boundary.
This distinction is important when interpreting a reported p-value: a p-value from a design involving interim monitoring should be understood in the context of the prespecified sequential testing framework rather than as though the analysis were necessarily a single unplanned final test.
14. One-Sided Testing and Superiority
The primary analysis notes describe a superiority hypothesis and specify a one-sided alpha of 0.05. The planning question was whether the additional chemotherapy would increase 4-year DFS from 80% to 90%.
In a superiority framework, the direction of the prespecified hypothesis matters. The design is not asking whether the two treatments are merely similar; it is testing whether the additional chemotherapy produces an improvement in the targeted endpoint according to the planned statistical criterion.
The reported DFS estimate of 1.05, together with its 90% two-sided confidence interval of 0.65–1.68 and P = 0.56, should therefore be interpreted against the trial's stated superiority objective rather than against a non-inferiority margin.
15. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized arm as the number affected relative to the number at risk.
| Safety measure | Arm I | Arm II |
|---|---|---|
| Serious adverse events | 14/107 | 25/101 |
| Arm | Cisplatin/Radiation Therapy | Cisplatin/Radiation Therapy + Carboplatin/Paclitaxel |
These serious-adverse-event counts are reported descriptively here. The ClinicalTrials.gov record does not provide a formal statistical comparison, confidence interval, severity breakdown, exposure-adjusted incidence rate, or detailed adverse-event definitions. It would therefore be inappropriate to turn the two counts into a formal comparative safety conclusion beyond describing what the registry reports.
16. What the Primary Result Does — and Does Not — Mean
The DFS HR of 1.05 is a relative estimate of the event hazard between the randomized groups. It does not equal an absolute percentage-point difference in disease-free survival and does not mean that 5% more patients experienced failure.
The 90% two-sided DFS CI of 0.65–1.68 shows that the point estimate should not be interpreted in isolation. The interval includes the null value of 1 and spans both lower and higher hazard ratios.
The DFS P = 0.56 describes the result of the reported statistical test under its testing framework. It does not quantify the magnitude of treatment effect, the probability that the null hypothesis is true, or the probability that the treatment is ineffective.
DFS and OS incorporate follow-up time and censoring. A participant who has not experienced an event by the end of observed follow-up is not equivalent to a participant known never to experience the event; the Kaplan-Meier framework accounts for the participant's observed information up to censoring.
17. Limitations
- Limited numerical reporting in the ClinicalTrials.gov record: the available trial data provide hazard ratios, 90% two-sided confidence intervals, and p-values for DFS and OS, but do not provide median DFS, median OS, event counts, or time-specific observed survival estimates.
- No baseline table in the ClinicalTrials.gov record: baseline demographic and clinical characteristics cannot be assessed from the ClinicalTrials.gov record.
- No subgroup results: the ClinicalTrials.gov record does not contain subgroup hazard ratios or interaction tests, so treatment-effect heterogeneity cannot be evaluated.
- No reconstructed Kaplan-Meier curves: summary statistics are insufficient to reproduce participant-level event and censoring histories.
- Hazard-ratio interpretation: the HR is a model-based relative time-to-event measure and should not be treated as an absolute risk difference. The ClinicalTrials.gov record does not provide diagnostics for proportional hazards.
- Interim-analysis detail: alpha spending is identified in the analysis notes, but the ClinicalTrials.gov record does not specify the complete interim-monitoring schedule or numerical boundaries.
- Safety comparison: serious adverse events are reported as affected/at-risk counts by arm, without a formal comparative safety analysis.
- Analysis population: the posted statistical analyses specify eligible participants. The ClinicalTrials.gov record does not provide additional detail on exclusions or deviations from the randomized population.
- Generalizability: the trial concerns high-risk early-stage cervical cancer after radical hysterectomy; conclusions should be understood within the population represented by the registered trial.
18. Why This Trial Matters Statistically
RTOG 0724 is a useful teaching example because its central question is structured around an incremental treatment effect and a time-to-event primary endpoint. The statistical story can be understood by separating the design hypothesis, the time-to-event estimation, the between-arm test, and the uncertainty around the effect estimate.
| Statistical concept | How it appears in RTOG 0724 |
|---|---|
| Randomization | Participants were randomized between two parallel treatment arms. |
| Superiority testing | The trial's hypothesis type is identified as superiority. |
| Time-to-event endpoint | DFS is the registered primary endpoint and OS is a reported secondary endpoint. |
| Kaplan-Meier estimation | DFS is estimated using Kaplan-Meier methods. |
| Log-rank test | The distributions of DFS and OS are compared using the log-rank test. |
| Hazard ratio | DFS HR = 1.05; OS HR = 0.91. |
| Confidence interval | Both reported HRs have 90% two-sided confidence intervals. |
| One-sided alpha | The primary sample-size notes specify one-sided alpha = 0.05. |
| Statistical power | The primary hypothesis was planned with 80% statistical power. |
| Interim analysis / alpha spending | The primary analysis notes explicitly identify interim analysis / alpha spending. |
| Censoring | Participants without an event by last follow-up are censored in the DFS definition; OS likewise uses death or last follow-up. |
19. A Statistical Reading of the Trial
The most informative way to read the primary result is to consider three quantities together:
- The point estimate: DFS HR 1.05.
- The uncertainty: 90% two-sided CI 0.65–1.68.
- The hypothesis test: log-rank P = 0.56 under the reported superiority framework.
The point estimate alone could be overinterpreted in either direction. An HR of 1.05 is close to the null value, but the confidence interval is broad enough to include materially lower and higher relative hazards. The p-value then provides the formal test statistic's compatibility with the null hypothesis within the specified framework.
This illustrates why a clinical-trial result should not be reduced to a single p-value. A statistically literate interpretation asks what was estimated, how precisely it was estimated, what hypothesis was tested, what population was analyzed, and how the endpoint was defined.
20. Primary vs Secondary Analysis
| Endpoint | Role | Estimate | Interpretive focus |
|---|---|---|---|
| Disease-free Survival | Primary | HR 1.05; 90% CI 0.65–1.68; P = 0.56 | Primary superiority comparison of time to first DFS failure or death |
| Overall Survival | Secondary | HR 0.91; 90% CI 0.49–1.69; P = 0.40 | Time from randomization to death or last follow-up |
The distinction between primary and secondary endpoints is important because it reflects the role each endpoint was assigned in the trial's statistical plan. A secondary result should not automatically be treated as though it carried the same confirmatory status as the primary endpoint.
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov: RTOG 0724, NCT00980954.
Continue with Clinical Biostats statistical tutorials
Explore the statistical concepts behind randomized trials, survival analysis, confidence intervals, hypothesis testing, and time-to-event endpoints.
24. Record Summary
RTOG 0724 provides a focused example of statistical analysis for a randomized phase 3 oncology trial with a time-to-event primary endpoint. The trial randomized participants between cisplatin/radiation therapy and cisplatin/radiation therapy plus additional carboplatin/paclitaxel, with disease-free survival as the registered primary endpoint.
The primary DFS analysis used Kaplan-Meier estimation and a log-rank test, with a reported hazard ratio of 1.05, a 90% two-sided confidence interval of 0.65–1.68, and P = 0.56. Overall survival was reported as a secondary time-to-event endpoint with an HR of 0.91, a 90% two-sided confidence interval of 0.49–1.69, and P = 0.40. The analysis notes also identify interim analysis / alpha spending, one-sided testing, a one-sided alpha of 0.05, and 80% statistical power in the primary design.
The central statistical lesson is that these results should be interpreted as a combination of effect estimates, confidence intervals, hypothesis tests, and endpoint definitions. The hazard ratio communicates relative event rates, the confidence interval communicates uncertainty, and the p-value addresses the prespecified hypothesis test. None of these quantities, considered alone, describes the complete clinical meaning of the trial.