RTOG 0617: Complete Statistical Analysis of Radiation Therapy and Chemotherapy in Stage III Non-Small Cell Lung Cancer

1. Trial at a Glance

RTOG 0617 was a randomized phase 3 oncology trial evaluating radiation therapy dose and cetuximab assignment in patients with newly diagnosed stage III non-small cell lung cancer that could not be removed by surgery. The registry reports a two-by-two factorial design in its statistical analysis text, with radiation therapy dose and cetuximab as the two treatment factors.

2. Clinical Question

The trial addressed two treatment factors within a randomized phase 3 setting: radiation therapy dose and cetuximab assignment. The primary registered endpoint was overall survival, with separate primary analyses comparing the combined 60 Gy radiation groups with the combined 74 Gy radiation groups and the combined cetuximab groups with the combined groups receiving no cetuximab.

3. Trial Design

The registry's statistical analysis text describes the trial as a two-by-two factorial design, with radiation therapy dose as one treatment factor and cetuximab as the other. The registry profile separately lists the design model as crossover. These registry fields should not be silently collapsed into a different design description.

4. Randomization and Analysis Populations

The primary overall-survival analyses were restricted to eligible patients who did not withdraw consent. The registry further specifies that patients enrolled while the high-dose radiation therapy arms were open to accrual were included in the radiation comparison, while patients enrolled while the cetuximab arms were open to accrual were included in the cetuximab comparison.

5. Primary Endpoint

The registered time frame is from randomization to last follow-up. The analysis occurred after all patients were on study for 18 months, with maximum follow-up as specified in the registry.

6. Primary Results: Overall Survival by Radiation Therapy Dose

The first primary analysis compared combined patients receiving 60 Gy RT with combined patients receiving 74 Gy RT. The registry reports a two-sided 95% confidence interval and a log-rank test. The analysis notes describe one-sided testing at an alpha of 0.0125 for each factor, with alpha of 0.0250 for both factors to account for multiple comparisons.

The registry's reference level for the radiation analysis is the 60 Gy group. An HR of 1.38 therefore represents a higher estimated hazard of death in the 74 Gy comparison group relative to the 60 Gy reference group, under the fitted time-to-event comparison.

7. Primary Results: Overall Survival by Cetuximab Assignment

The second primary analysis compared combined patients receiving cetuximab with combined patients from arms receiving no cetuximab.

With the cetuximab comparison expressed as cetuximab versus the groups receiving no cetuximab, the estimated hazard ratio of 1.07 is close to 1. The confidence interval extends from 0.84 to 1.35, so the estimate is compatible with a range of relative hazard values around the null value.

8. Primary Analysis and Multiple Comparisons

The registry's analysis notes explicitly describe RTOG 0617 as a two-by-two factorial design, with radiation therapy dose as one factor and cetuximab as the other. For the primary overall-survival analyses, the notes state that a log-rank test for each factor at a one-sided α of 0.0125 would account for an overall α of 0.0250 for both factors.

This is an important statistical feature of the trial. Two treatment factors were evaluated within the same overall trial, so the interpretation of each primary test belongs to the prespecified multiple-comparison framework rather than being treated as two unrelated hypothesis tests.

9. Secondary Time-to-Event Results: Progression-Free Survival

Progression-free survival was analyzed separately for radiation therapy dose and cetuximab assignment. The registry states that PFS is estimated by the Kaplan-Meier method and that each comparison was tested with a two-sided significance level of 0.05.

The radiation comparison has an HR of 1.19, while the cetuximab comparison has an HR of 0.99. Both confidence intervals include 1.00. The registry explicitly identifies these as secondary comparisons rather than the primary overall-survival endpoint.

10. Secondary Time-to-Event Results: Local-regional Failure

Local-regional failure was estimated using the cumulative incidence method and compared with Gray's test. This is a competing-risks framework rather than an ordinary Kaplan-Meier treatment of every non-target event as simple censoring.

The registry labels the outcome as Local-regional Failure (Reported as Two-year Estimates), while the time frame is from randomization to last follow-up. The analysis notes specify that local-regional failure was estimated by cumulative incidence and compared separately by radiation level and cetuximab assignment.

11. Biomarker and Tumor-Volume Analyses

EGFR H-Score

The registry reports Cox regression analyses comparing patients with EGFR H-Score < 200 with patients with EGFR H-Score ≥ 200. These are secondary analyses in eligible patients with EGFR H-Score.

The registry describes both as univariate Cox models. The reference level is EGFR H-Score < 200.

Gross Tumor Volume

Gross tumor volume was evaluated as a continuous variable in eligible patients with GTV data. The registry reports both a univariate and a multivariate Cox analysis.

The multivariate model adjusted for planned radiation therapy dose group and the interaction of GTV and planned dose. The registry reports a P-value of 0.77 for the interaction of GTV and planned dose.

PET Standardized Uptake Value

Pre-treatment PET SUV was evaluated as a continuous variable in eligible patients with baseline PET SUV data. The registry reports three Cox analyses.

These are prognostic analyses rather than randomized treatment comparisons. Their interpretation is therefore different from the primary randomized radiation and cetuximab comparisons.

12. Safety Results

The registry reports serious adverse events by radiation and cetuximab grouping. These results should be read as affected participants divided by participants at risk in the posted analysis.

The registry also contains formal statistical analyses of grade 3-5 toxicity. For esophagitis and pneumonitis, the worst grade at any time was classified into < grade 3 and ≥ grade 3, and only toxicities reported as possibly, probably, or definitely related to treatment were considered.

The esophagitis analysis therefore has a markedly smaller reported p-value than the pneumonitis and other grade 3-5 toxicity analyses. A p-value alone does not specify the magnitude of the absolute toxicity difference, and the ClinicalTrials.gov record does not provide corresponding affected-participant estimates for these three analyses.

EGFR Group and Grade 3+ Adverse Events

The registry also reports a chi-squared analysis of the percentage of patients with grade 3+ adverse events by EGFR H-Score group.

This is an association between an EGFR-defined group and a binary safety outcome. It is not a randomized treatment comparison and should not be interpreted as evidence that changing EGFR H-Score would change toxicity.

13. Patient-Reported and Quality-of-Life Outcomes

Lung Cancer Subscale of FACT-TOI

The registry evaluated the percentage of patients with a decline from baseline to 3 months in the Lung Cancer Subscale of the Functional Assessment of the Cancer Therapy Trial Outcome Index. Changes were grouped using a 2 point decline as the threshold. The comparison was controlled for cetuximab assignment using a Cochran-Mantel-Haenszel test.

Patient-reported Swallowing Score

The swallowing score was summarized as an area under the curve from randomization to 6 weeks after the start of radiation therapy, corresponding to 6-10 weeks from randomization. The analysis population required at least 15 diary entries, including one before radiation therapy and one during the sixth week.

The swallowing score was expressed as score × days, while the EQ5D VAS outcome was expressed as score × months. These analyses illustrate how longitudinal patient-reported measurements can be reduced to an area-under-the-curve summary before comparison between randomized groups.

14. Statistical Methodology

Kaplan-Meier estimation

Overall survival is defined from randomization to death from any cause, with patients last known to be alive censored at their date of last contact. The registry states that overall survival is estimated using the Kaplan-Meier method. Progression-free survival is also estimated by Kaplan-Meier.

Log-rank test

The primary overall-survival comparisons and the secondary PFS comparisons use the log-rank test. Conceptually, the log-rank test compares the observed and expected event experience between groups across the follow-up period rather than comparing only a single time point.

For RTOG 0617, the log-rank framework is especially important because the primary question concerns a time-to-event endpoint and because the registry explicitly records a factorial comparison of two treatment factors.

Cox proportional-hazards model

The registry uses Cox regression for the EGFR, GTV, and PET SUV analyses. The hazard ratio is the effect measure for these analyses. In the GTV analysis, the multivariate model adjusts for planned radiation therapy dose group and the interaction between GTV and planned dose.

Gray's test and competing risks

Local-regional failure was estimated using the cumulative incidence method and compared with Gray's test. This is appropriate to a competing-risks setting because a competing event can prevent the target event from subsequently being observed in the same way.

Chi-squared testing

Binary toxicity outcomes were compared with chi-squared tests. The registry specifies two-sided testing with a significance level of 0.05 for the radiation comparison of esophagitis and pneumonitis.

Cochran-Mantel-Haenszel testing

The FACT-TOI-LCS analysis used a Cochran-Mantel-Haenszel test to compare decline versus no decline by radiation therapy level while controlling for cetuximab assignment. This illustrates the distinction between an unadjusted two-group comparison and a stratified categorical comparison.

t-tests

The patient-reported swallowing score and EQ5D VAS area-under-the-curve outcomes were analyzed with two-sided t-tests. Both comparisons used a two-sided significance level of 0.05 according to the registry.

15. Statistical Methods Explained

Why use a log-rank test for overall survival?

Overall survival is a time-to-event endpoint. Patients can have different follow-up durations, and some patients may remain alive at the end of observation. The log-rank test uses the timing of deaths and the numbers at risk over follow-up rather than treating every patient as if the same fixed follow-up time were available.

What does an overall-survival HR of 1.38 mean?

For the reported radiation comparison, the HR of 1.38 is expressed with the 60 Gy group as the reference level. It means the estimated instantaneous hazard of death in the comparison group was 1.38 times that of the reference group. It does not mean that 38% more patients died or that each patient's probability of death changed by 38%.

Why does the confidence interval matter?

The HR is an estimate rather than a known population quantity. The two-sided 95% confidence interval describes statistical uncertainty around that estimate under the analysis framework. For the radiation primary analysis, the interval is 1.09–1.76; for the cetuximab primary analysis, it is 0.84–1.35. The interval therefore conveys substantially more information than the point estimate alone.

Why is Gray's test used for local-regional failure?

Local-regional failure can be considered in a competing-risks framework because other events can prevent observation of the target event. The registry therefore uses cumulative incidence estimation and Gray's test rather than simply applying an ordinary Kaplan-Meier framework to the target event.

Why was a Cochran-Mantel-Haenszel test used for the FACT-TOI-LCS endpoint?

The outcome was converted into decline versus no decline using a 2 point decline threshold. The Cochran-Mantel-Haenszel procedure then compared these categorical outcomes by radiation therapy level while controlling for cetuximab assignment. It therefore matches the registry's categorical endpoint and its specified adjustment factor.

Why does the factorial structure matter?

The registry's analysis notes describe two treatment factors rather than one simple two-arm comparison. Radiation therapy dose and cetuximab assignment were evaluated separately. This structure affects both the interpretation of the comparisons and the multiple-testing framework, including the stated one-sided alpha of 0.0125 for each factor.

Why should the biomarker analyses not be treated like the randomized treatment comparison?

EGFR H-Score, GTV, and PET SUV analyses use selected analysis populations and regression models rather than the original randomized treatment contrast. They are therefore primarily prognostic or covariate analyses. Even a statistically small p-value in such an analysis would not by itself demonstrate that modifying the biomarker would improve an outcome.

16. Primary-Analysis Interpretation: Effect Size, Precision, and P-values

For RTOG 0617, these distinctions are particularly important. The two primary hazard ratios point in different directions, and the confidence intervals provide different levels of information about uncertainty. At the same time, the analyses cannot be interpreted independently of the factorial structure and its stated alpha allocation.

17. Crossover, Accrual Changes, and Internal Validity

The registry profile lists the design model as crossover. Separately, the statistical analysis text describes the trial as a two-by-two factorial design. The ClinicalTrials.gov record does not provide additional details sufficient to characterize the meaning of the crossover design field beyond that recorded classification.

More directly relevant to the posted analyses are the changes in treatment availability during accrual. Cetuximab arms were added after study opening, and the first 34 patients were excluded from the cetuximab analysis. The high-dose radiation arms closed early at n=474, after which later patients enrolled to the low-dose radiation arm were excluded from the radiation analysis.

18. Design Topics Supported by the Registry

19. What the Secondary Analyses Add

The secondary analyses broaden the statistical story beyond overall survival. PFS evaluates another time-to-event outcome; local-regional failure uses competing-risk methodology; EGFR, GTV, and PET SUV analyses evaluate prognostic or predictive relationships; and safety and patient-reported outcomes examine treatment effects beyond survival.

20. Limitations

• Arm availability changed during accrual: cetuximab arms were added after study opening, excluding the first 34 patients from the cetuximab analysis.
• High-dose radiation arms closed early: the high-dose arms closed at n=474, and later patients enrolled to the low-dose radiation arm were excluded from the radiation analysis.
• Different analysis populations: the denominator for a specific comparison therefore differs from the overall enrollment of 544.
• Hazard-ratio interpretation: a Cox-type hazard ratio is a model-based relative time-to-event measure and should not be interpreted as an individual-level risk ratio.
• Censoring: the registered overall-survival definition censors patients last known to be alive at the date of last contact. Time-to-event conclusions therefore depend on the censoring framework.
• Secondary analyses: biomarker and tumor-volume analyses use selected populations with available measurements and should not be interpreted as randomized treatment effects.
• Competing risks: local-regional failure requires interpretation within the cumulative-incidence and competing-risks framework reported by the registry.
• Multiple endpoints and analyses: the primary factorial comparisons have an explicitly stated alpha framework, while the ClinicalTrials.gov record does not provide a complete multiplicity plan for all secondary analyses.
• Incomplete methodological detail: the ClinicalTrials.gov record does not specify a missing-data/imputation strategy, Bayesian methods, or a detailed interim-analysis procedure.
• Registry-level evidence: this page is restricted to the ClinicalTrials.gov-derived data and does not add numerical results from publications.

21. Why This Trial Matters Statistically

RTOG 0617 is a useful statistical teaching case because the trial combines a factorial treatment structure with multiple time-to-event endpoints, competing-risk analysis, covariate modeling, categorical safety testing, and patient-reported outcomes. The design also illustrates how changes in accrual can affect the population available for a particular treatment-factor comparison.

22. Related Tutorials

Learn more about the methods used in this trial:

23. Related Calculators

24. Sources

• ClinicalTrials.gov: RTOG 0617, NCT00533949.
• PubMed: PMID 31841363.
• PubMed: PMID 38935373.
• PubMed: PMID 36842666.
• PubMed: PMID 36738929.
• PubMed: PMID 36436615.

25. Record Summary

RTOG 0617 provides a compact example of several important clinical-trial statistical principles. The registry describes a randomized phase 3 trial with a two-by-two factorial structure, radiation therapy dose and cetuximab as treatment factors, and overall survival as the registered primary endpoint. The primary analyses used log-rank tests and hazard ratios, with the analysis notes specifying one-sided alpha allocation across the two treatment factors.

The radiation comparison reported an overall-survival HR of 1.38 with a 95% CI of 1.09–1.76 and P = 0.0042. The cetuximab comparison reported an HR of 1.07 with a 95% CI of 0.84–1.35 and P = 0.29. Secondary analyses extended the statistical framework to PFS, local-regional failure, EGFR H-Score, GTV, PET SUV, toxicity, and patient-reported outcomes.

The most important interpretive lesson is that these numbers cannot be read independently of their analysis populations and methods. Radiation and cetuximab comparisons had different accrual histories; local-regional failure required a competing-risks approach; continuous covariates were modeled with Cox regression; and categorical toxicity and patient-reported outcomes required different statistical procedures.