Efficacy Study of Avastin® With Pemetrexed +/- Carboplatin to Treat Elderly Patients With Non-small Cell Lung Cancer: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Efficacy Study of Avastin® With Pemetrexed +/- Carboplatin to Treat Elderly Patients With Non-small Cell Lung Cancer (NCT00976456), a randomized phase 3 study. The primary outcome is Progression Free Survival.

2. Clinical Question

The central statistical question is not simply which group has the larger median. The comparison also depends on the uncertainty around each estimate, the reported hazard ratio and its confidence interval, and the method used to assess the groups. These results should be read as evidence from this particular randomized study, rather than as a definitive description of every patient or setting.

3. Trial Design

This was a phase 3, randomized, open-label, parallel-group study. It compared bevacizumab plus pemetrexed with bevacizumab plus pemetrexed plus carboplatin as first-line treatment for patients with advanced or recurrent non-squamous non-small cell lung cancer. The registry record describes the study as multicenter and completed. Enrollment was 271.

Random allocation is intended to make treatment groups comparable, on average, with respect to both measured and unmeasured characteristics at the point of assignment. That design supports a comparative interpretation, although the record excerpt does not specify how the random sequence was generated, whether allocation was concealed, or what procedures were used to implement allocation. Those details are not reported in the registry record.

The study was open-label, so participants and study personnel were not masked to treatment assignment. This can matter for outcomes involving judgment or decisions about assessment and care. Progression-free survival includes an assessment of disease progression, but the registry record does not describe any independent or blinded review process. Such a process is not reported in the registry record.

The treatment descriptions specify a minimum of 4 cycles of bevacizumab with chemotherapy, followed by bevacizumab maintenance, with pemetrexed if appropriate, until disease progression. The posted primary endpoint has a timeframe of 42 months. The registry record lists start as 2009-09 and primary completion as 2014-10. It does not report the original sample-size justification or the trial's own detailed statistical analysis plan, so the planning illustrations on this page should not be mistaken for the assumptions used to design the study.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Overall Study

Reasons for not completing this period:

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

Population: 271 patients were enrolled and randomized, due to Screening failure and subject withdrawn consent the intent-to-treat (ITT) population comprised 253 evaluable patients (BP n=119; BPC n=134).

6. Primary Endpoint

The registry lists the primary outcome as Progression Free Survival, measured over 42 months.

The primary endpoint is progression-free survival, reported as a median in months. The registry definition starts the clock on the day of first treatment and ends it at death from any cause or disease progression, whichever occurs first. If neither progression nor death is observed during the study, the endpoint uses the day of the last response assessment. The listed timeframe is 42 months.

This type of endpoint combines disease progression and death into a time-to-event measure. In general, time-to-event outcomes can account for the fact that participants may be followed for different lengths of time and that some have not experienced the event by the end of observation. Those participants contribute information up to their last known assessment rather than being treated as if an event occurred then. The registry definition indicates such a last-assessment endpoint, but the record does not give the number or handling details for censored observations.

A median summarizes the time by which half of the relevant event-time distribution has been reached, under the assumptions of the analysis. It is a useful compact description, but it does not show the full pattern of event timing or establish that the groups differ throughout follow-up. The hazard ratio is a separate summary used for the reported comparison. Its interpretation relies on how the event and groups were coded, details that are not fully described in the registry record.

Progression-free survival is used in clinical trials because it captures time until worsening or death, rather than requiring death alone as the endpoint. That is a general explanation of the endpoint, not a claim about why this particular trial selected it; the specific rationale is not reported in the registry record.

7. Results Posted in the Registry

• Bevacizumab + Pemetrexed vs Bevacizumab + Pemetrexed + Carboplatin: hazard ratio 1.29 (95% CI 0.99–1.68), reported p = 0.058 (as reported in the registry).

8. Secondary Efficacy Endpoints

The registry also posts results for secondary outcomes. For each outcome below, the endpoint type, the test applied to the posted summary statistics, the result and its interpretation are given in turn. Secondary endpoints are usually not adjusted for multiple testing, so they support rather than confirm the primary finding.

Overall Survival

• Endpoint type: Time to event: compared as a hazard ratio.
• Test: Hazard ratio as reported by the sponsor (Wilcoxon (Mann-Whitney)); it was not recalculated.
• Result:
  • Bevacizumab + Pemetrexed vs Bevacizumab + Pemetrexed + Carboplatin: hazard ratio 1.09 (95% CI 0.79–1.50), reported p = 0.387 (as reported in the registry).
• Interpretation:
  • The 95% confidence interval includes no difference (a ratio of 1), so these data do not show a statistically significant difference between Bevacizumab + Pemetrexed vs Bevacizumab + Pemetrexed + Carboplatin and the control. The interval gives the range of effects that remain compatible with the data.

Overall survival is the one posted secondary outcome. Across the outcome summaries, the point estimates show a broadly similar descriptive pattern in which the combination including carboplatin has the larger reported median. The secondary comparison is less precise and does not provide a clear statistical signal against no relative difference. These secondary findings are directionally consistent with the primary outcome's descriptive pattern, but they do not resolve the uncertainty in the primary comparison or establish a treatment effect.

Secondary outcomes are useful for understanding whether the primary result sits alongside other measures of benefit, but they usually carry less weight for a confirmatory claim unless the trial's testing strategy accounts for them. If many outcomes or comparisons are tested, the chance of at least one apparently unusual result can increase. Statistical procedures for controlling that multiplicity can help, but the registry record does not report whether adjustment was planned or applied.

The record lists one secondary outcome and two outcome measures and analyses, but it does not describe a hierarchical testing plan, an alpha-allocation strategy, or whether the secondary analysis was intended to be confirmatory. Those details are not reported in the registry record. Accordingly, secondary results are best treated as supporting context rather than independent proof of the primary comparison. Their interpretation should also account for the fact that the registry provides summary estimates, not a full account of follow-up, missing observations, or analysis populations.

Secondary outcomes registered for this trial, for reference (only those above were recalculated):

• Overall Survival

9. Statistical Interpretation

10. Statistical Methods Explained

The primary endpoint is time-to-event: each participant's time is measured from first treatment until progression or death, whichever occurs first, or until the last response assessment if neither event is observed. Time-to-event analysis is useful when participants have varying follow-up or have not had an event at the time observation ends. In general, such observations are censored: their available follow-up contributes information, but their eventual event time is unknown. The registry gives the endpoint definition but does not report the amount of censoring or the detailed censoring rules used in the analysis.

What does a median time-to-event estimate describe?

A median is a point in the event-time distribution, not an average duration for every participant. It can be understood as the time by which half of the modeled or observed group has had the event, subject to the estimator and censoring assumptions. Reporting a median in months gives a compact summary, but it does not show the complete distribution. Participants may experience events earlier or later than the median, and the two groups' event-time patterns may vary across follow-up.

The confidence intervals beside the medians communicate uncertainty in those estimates. They are not the same as a confidence interval for the difference between the two medians. In general, comparing two separate intervals by eye is not a formal test of the treatment contrast. The registry separately reports a hazard ratio and p-value for the comparison; these are the direct posted comparative statistics.

How should the hazard ratio be read?

A hazard ratio is a relative summary of event rates over time, conditional on the analysis and its assumptions. In general, a value of 1 represents equal hazards under the chosen group ordering and event coding. Values on either side indicate a relative difference in the instantaneous event rate as defined by that analysis. The numerical direction cannot be interpreted safely without knowing which group is the reference and which event is coded as the event. The registry lists the comparison order and gives a hazard ratio, but its detailed coding convention is not reported in the registry record.

The hazard ratio is not a ratio of the two medians, and it does not say that every participant has the same relative change in time to event. It summarizes the comparison rather than describing individual outcomes. A single hazard ratio is most straightforward to interpret when the relative hazards are reasonably stable over time. Whether that assumption was checked, and whether alternative summaries were considered, is not reported in the registry record.

What does the reported p-value convey?

A p-value measures how compatible the observed data are with a specified null model, under the test and assumptions used. It is not the probability that the null hypothesis is true, and it does not measure the probability that the result occurred by chance. The posted p-value is 0.058. The planning assumptions on this page list a two-sided significance level of 0.05; against that conventional cutoff, the result is above the threshold. A p-value just above a cutoff does not make the observed effect disappear, and a p-value below a cutoff would not establish clinical importance. The estimate, interval, design, and limitations all matter.

The registry identifies the method as Wilcoxon (Mann-Whitney) while the reported parameter is a hazard ratio. Those labels are reproduced as posted. The statistical method's precise implementation, the estimand it targeted, and the rationale for pairing that method with the reported parameter are not described in the registry record. The method label should therefore not be expanded into assumptions about a particular model or test procedure.

Why does over-dispersion matter in some analyses?

In general, over-dispersion occurs when observed variability is greater than a statistical model expects. If it is ignored, standard errors can be too small and confidence intervals or tests can appear more precise than warranted. This is a general statistical concept, not a description of the analysis in this trial. The primary outcome is time-to-event, and the registry does not report an over-dispersion adjustment or a count model for this endpoint.

More broadly, statistical methods rely on assumptions about the data, endpoint definition, follow-up, and analysis population. A registry summary cannot establish whether every assumption was checked. The protocol-level analysis details, model diagnostics, missing-data procedures, and sensitivity analyses are not reported in the registry record. Interpretation should consequently stay close to the estimands and results that are actually posted.

11. Confidence Intervals and Precision

The primary comparison has a reported hazard ratio of 1.290 and a 95% confidence interval from 0.989 to 1.682. In general, a 95% confidence interval is a range generated by a method that, over repeated comparable studies under its assumptions, would cover the target parameter in 95% of cases. It is not a statement that there is a 95% probability that this particular interval contains the true value.

The interval is useful because it shows the uncertainty around the point estimate. Its lower end is close to 1, while its upper end is farther from 1. Since 1 represents no relative difference for a hazard ratio under the chosen coding, the interval includes that value. The interval also extends to values consistent with a more substantial relative contrast. This range limits how precise a conclusion can be drawn from the point estimate alone. Interpretation of direction still depends on reference-group and event coding, which are not fully described in the registry record.

The median progression-free survival estimates have their own 95% confidence intervals: 4.3 to 6.0 months for bevacizumab plus pemetrexed and 5.8 to 7.4 months for bevacizumab plus pemetrexed plus carboplatin. These intervals quantify uncertainty around each median separately. They are not the uncertainty interval for a between-group median difference, and their overlap or lack of overlap should not substitute for the reported comparative analysis.

Precision is distinct from whether a result crosses a significance threshold. A wide interval can leave several effect sizes compatible with the data, even if the point estimate appears clear. Conversely, a narrow interval can be precise without establishing that an effect matters clinically. The registry record does not report a clinically important threshold for the primary endpoint, a non-inferiority margin, or a target interval width. Those omissions limit judgments about whether the precision was adequate for the trial's intended decision.

12. Planning a Similar Trial: Sample Size

This is a planning illustration, not a re-evaluation of the original trial. It asks: if a new study wanted to detect a difference like the one observed, or a smaller one, how large would it need to be? It is deliberately not “observed power” of the finished trial, which is not a meaningful quantity.

The sample-size table and curve are planning illustrations derived from the observed effect, not the original design calculation for this trial. Because the endpoint is time-to-event, the table reports the total number of events needed, not participants per arm. Its planning assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation; each scenario targets the stated fraction of the observed effect. These assumptions describe the illustration, while the trial's original design assumptions are not reported in the registry record.

The 80% power column describes a planning target: under the specified assumptions and effect, a study with the listed total number of events would have an 80% chance of rejecting the null hypothesis at the stated significance level, if the effect and model assumptions were correct. The 90% power column uses a higher probability of detecting that specified effect, so it calls for a larger event target. These are prospective planning quantities, not a description of what happened in this trial.

The rows show planning for 100% of the observed effect, 75% of the observed effect, and 50% of the observed effect. As the effect to be detected becomes smaller, more events—and generally more participants followed long enough to experience events—are needed to distinguish it from no effect under the same assumptions. The displayed event totals are not the number of people enrolled in this study, and they should not be read as a claim that the study reached or failed to reach those targets.

This is not observed power. It is a planning illustration using the observed effect as an input. It cannot tell whether the trial was adequately designed because the original design effect, assumptions, event target, and stopping rules are not reported in the registry record. Read the plotted curve and table as sensitivity illustrations about how planning targets change with the assumed effect and desired power.

13. How the Sample Size Is Derived

Power is the probability that a study of a given size rejects the null hypothesis when a specific true effect exists. It depends on four things: the effect to detect, the variability of the outcome, the significance level (here a two-sided α of 0.05, so z_1−α/2 = 1.960) and the desired power (z_1−β = 0.842 for 80% and 1.282 for 90%).

Formula

d is the total number of events needed (Schoenfeld's formula) with 1:1 allocation and proportional hazards.

Worked calculation for this trial

1. Hazard ratio HR = 1.290, so ln(HR) = 0.2546.
2. d = 4 × (1.96 + 0.842)² / [ln(1.290)]² = 484.2, rounded up to 485 events in total for 80% power.

The smaller-effect rows in the table above repeat the same calculation with the effect multiplied by 0.75 and 0.5 (for ratios, applied to the logarithm of the ratio).

R code

alpha <- 0.05
z_a   <- qnorm(1 - alpha / 2)          # two-sided
hr <- 1.29                              # observed hazard ratio
events_needed <- function(power, frac = 1) ceiling(round(4 * (z_a + qnorm(power))^2 / log(hr^frac)^2, 6))   # Schoenfeld, 1:1 allocation
for (f in c(1, 0.75, 0.5))
  cat(sprintf("%3.0f%% of observed effect: events = %s (80%% power), %s (90%% power)\n", 100 * f,
              format(events_needed(0.80, f), big.mark = ","), format(events_needed(0.90, f), big.mark = ",")))

# Simulation: given d total events under proportional hazards, events in the treatment arm are Binomial(d, hr / (1 + hr)).
# The score (log-rank type) statistic is (D1 - d / 2) / sqrt(d / 4).
set.seed(976456)
d    <- events_needed(0.80)
reps <- 20000
D1 <- rbinom(reps, d, hr / (1 + hr))
z  <- (D1 - d / 2) / sqrt(d / 4)
cat(sprintf("Simulated power with %s events: %.3f (%d replicates)\n", format(d, big.mark = ","), mean(abs(z) > z_a), reps))

Output from running the code

100% of observed effect: events = 485 (80% power), 649 (90% power)
 75% of observed effect: events = 861 (80% power), 1,153 (90% power)
 50% of observed effect: events = 1,937 (80% power), 2,593 (90% power)
Simulated power with 485 events: 0.785 (20000 replicates)

The simulation uses the fact that, with d events in total and proportional hazards, the number of events in the treatment arm is binomial. It is an approximation that conditions on the total number of events, and it should land close to the target power.

14. Analyses Reported by the Sponsor

The registry's reported primary analysis gives a hazard ratio of 1.290, a 95% confidence interval from 0.989 to 1.682, and p = 0.058. The comparative-results section gives the same estimate, interval, and p-value. The analysis is labeled Wilcoxon (Mann-Whitney), and its analysis type is listed as non-inferiority or equivalence.

The posted results do not flag a discrepancy between the comparison and the sponsor-reported analysis. The figures shown in the results tables are therefore the sponsor's reported analysis, not an independently adjusted estimate. The registry also includes medians and their confidence intervals by group; these are descriptive summaries and do not replace the sponsor-reported comparative statistic.

The non-inferiority or equivalence label should be read cautiously. A label does not give the decision margin or show that a formal criterion was met. The margin, prespecified hypotheses, analysis populations, and detailed analysis plan are not reported in the registry record. Without those items, the reported hazard ratio and p-value can be described, but the registry summary alone cannot establish a non-inferiority or equivalence conclusion.

15. Multiplicity, Missing Data and Analysis Populations

Multiplicity arises when a study conducts multiple statistical tests, for example across several endpoints, time points, or subgroup analyses. In general, testing many hypotheses without an appropriate strategy increases the chance of obtaining at least one result that crosses a nominal significance threshold even when the corresponding effects are absent. Common approaches include prespecifying a hierarchy or adjusting the error rate, but the appropriate approach depends on the trial's aims and design.

The registry lists progression-free survival as the primary outcome and overall survival as a secondary outcome. It records two outcome measures and two analyses. It does not report whether the primary test was protected from other testing, whether an adjustment was made, how endpoints were prioritized, or whether any additional analyses were conducted. A multiplicity plan is not reported in the registry record. Therefore, the posted p-values should not be assumed to have been adjusted, nor should they be described as unadjusted unless the record explicitly says so.

Missing outcome information also affects interpretation. For time-to-event analysis, participants without a recorded event at the end of their available follow-up may be censored. In general, valid interpretation depends on appropriate follow-up and censoring assumptions. The registry definition specifies use of the last response assessment when progression or death is not observed, but it does not report how many participants were censored, how missing assessments were handled, or whether sensitivity analyses examined departures from the assumptions.

The registry outcome summaries list analysis-group sizes, but the analysis population definition is not reported in the registry record. It is not clear from this summary whether all randomized participants were included, whether participants were analyzed according to assigned group, or how protocol deviations and incomplete follow-up were handled. The record also does not report interim analyses, stopping rules, or a detailed statistical analysis plan. These omissions do not invalidate the posted results, but they limit independent assessment of their robustness and of the formal inferential claims.

16. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 0% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: dyspnea (6.7%).

The posted safety summary lists serious adverse events in 8 of 119 participants in the bevacizumab plus pemetrexed group and 7 of 134 participants in the bevacizumab plus pemetrexed plus carboplatin group. It also lists 61 and 97 participants, respectively, in the other adverse-event count category. These are counts in the registry summary; the definitions and observation periods for these categories are not described in the record excerpt.

Dyspnea is listed as the top serious adverse-event term at 6.7%. The record does not clarify the denominator or group attribution for that percentage in the summary shown. Death counts are not reported in the safety summary. The figures should not be interpreted as proving that an event was caused by a treatment, nor do the counts alone establish comparative safety. Such conclusions would require fuller information about event definitions, severity, exposure, follow-up, and the analysis used. The safety data are descriptive registry results, not medical advice.

17. What the Result Does, and Does Not, Mean

The posted results describe a randomized comparison of two bevacizumab-based regimens in patients with advanced or recurrent non-squamous non-small cell lung cancer. For the primary endpoint, the registry reports group-specific median progression-free survival, a hazard ratio, a confidence interval, and a p-value. The point estimates are informative summaries, while the interval and test show that the comparative estimate remains uncertain.

The primary p-value is above the conventional two-sided 0.05 threshold used in the planning illustration, and the hazard-ratio confidence interval includes 1. This means the posted primary analysis does not provide a statistically significant result at that threshold. It does not demonstrate that the regimens are identical, that either regimen is ineffective, or that a clinically important difference is impossible. Equivalence and non-inferiority require prespecified margins and methods that are not reported in the registry record.

The results also do not establish which regimen is preferable for an individual. The registry summary does not provide the full clinical context, patient-level data, complete safety profile, or the information needed to assess every possible benefit and harm. Nor can a single trial summary show that its findings apply equally to people or settings outside the studied population.

A balanced reading keeps four points together: the randomized design supports a comparative question; the study was open-label; the primary result has uncertainty; and essential details about analysis populations, missing data, multiplicity, and the original design assumptions are not reported in the registry record. The posted results can inform statistical understanding of this trial, but they are not a stand-alone treatment recommendation.

18. Why This Trial Matters Statistically

This trial illustrates why a clinical result should not be reduced to a single number or a binary significant/not-significant label. Its primary endpoint is a time-to-event outcome, so the reported medians summarize each group while the hazard ratio gives a comparative summary. Those quantities are related but are not interchangeable. The confidence intervals add essential information about how much uncertainty surrounds the estimates.

The record also shows the importance of checking what an analysis label can and cannot establish. Although the primary analysis is categorized as non-inferiority or equivalence, the margin and decision rules are not reported. A statistical category is not a substitute for the prespecified framework needed to support that kind of claim. Likewise, the method and reported parameter are both named, but the detailed implementation is absent.

Open-label treatment assignment, incomplete reporting about outcome assessment, and missing details on populations and censoring all shape how confidently the summary can be interpreted. These are not reasons to disregard the results; they are reasons to distinguish clearly between what the registry reports and what it does not establish. The planning illustration reinforces a separate principle: power calculations depend on assumed effects and design conditions, and post hoc illustrations are not evidence of observed power or of the original design's adequacy.

For biostatisticians and clinical researchers, this record is a useful example of reading endpoint summaries, comparative estimates, confidence intervals, and planning quantities as parts of one argument. Statistical interpretation depends not only on numerical results but also on the estimand, analysis plan, follow-up, and reporting transparency.

19. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

Several limitations are specific to the available record. First, the study was open-label, and the registry summary does not describe whether progression assessments were reviewed independently or without knowledge of treatment assignment. That leaves uncertainty about safeguards around an endpoint that includes disease progression.

Second, the primary endpoint is reported at a timeframe of 42 months, but the record excerpt does not describe the follow-up distribution, the number of events, or the amount and pattern of censoring. Those details are important for understanding a time-to-event analysis. The summary also does not specify whether the reported hazard ratio relied on an assumption of proportional hazards or whether that assumption was assessed.

Third, the registry labels the analysis non-inferiority or equivalence but does not report the margin, hypotheses, or detailed analysis populations needed to assess such a claim. The method is named as Wilcoxon (Mann-Whitney), while the reported parameter is a hazard ratio; the precise statistical implementation is not reported. This limits independent scrutiny of the comparative analysis.

Finally, the original sample-size justification, interim analyses, multiplicity strategy, missing-data procedures, and detailed statistical plan are not reported in the registry record. The planning illustrations on this page cannot fill those gaps because they are based on the observed effect and stated planning assumptions, not the original design. The safety summary is also brief and does not provide complete event definitions or follow-up context. These limitations concern what can be established from the registry summary; they do not, on their own, determine the quality of unreported study procedures.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: Efficacy Study of Avastin® With Pemetrexed +/- Carboplatin to Treat Elderly Patients With Non-small Cell Lung Cancer — NCT00976456.
• Registry record last updated 2016-03-10; status last verified 2016-02.
• Results first posted 2016-03-10.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

23. Record Summary

This completed phase 3 study randomized 271 patients in an open-label, parallel-group comparison of bevacizumab plus pemetrexed with the same regimen plus carboplatin. The primary outcome was progression-free survival, defined from first treatment to progression or death, with last response assessment used when neither event was observed. The posted timeframe was 42 months.

The registry reports median progression-free survival and a hazard ratio comparison, with a p-value above the two-sided 0.05 planning threshold and a confidence interval that includes 1. The record also lists overall survival as a secondary outcome. The results should be read with the uncertainty in the estimates and the missing analysis details in mind. The original design assumptions, multiplicity strategy, missing-data procedures, analysis populations, and non-inferiority margin are not reported in the registry record.