Osimertinib Treatment on EGFR T790M Plasma Positive NSCLC Patients (APPLE): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Osimertinib Treatment on EGFR T790M Plasma Positive NSCLC Patients (APPLE) (NCT02856893), a randomized phase 2 study. The primary outcome is PFS Rate at 18 Months.

2. Clinical Question

The trial also addressed the proposed role of liquid biopsy in detecting tumor progression and assessing T790M status during treatment. Its stated objectives included comparing upfront with sequential osimertinib strategies and exploring acquired resistance using an optional biopsy at progression. Those objectives provide context for the treatment arms, but the primary result concerns the 18-month PFS rate while receiving osimertinib.

The primary endpoint is a time-specific proportion, not a direct measure of how long every participant remained free of progression. Its interpretation depends on the RECIST 1.1 event definition, the treatment sequence, and the population included in the analysis. The registry record does not report additional detail about how plasma testing informed individual clinical decisions.

3. Trial Design

APPLE was a completed, randomized, open-label, parallel-group phase II trial sponsored by the European Organisation for Research and Treatment of Cancer - EORTC. The registry record lists an enrollment of 156 and a start date of 2017-10-10, with primary completion on 2022-09-20. The primary endpoint analysis population was per protocol rather than all enrolled participants.

Three treatment strategies were compared. The first used osimertinib until progression according to RECIST 1.1. The second used gefitinib until emergence of positive T790M status by blood testing or progression, followed by osimertinib until a second progression event according to RECIST 1.1. The third used gefitinib until progression according to RECIST 1.1, followed by osimertinib until progression according to RECIST 1.1.

The primary endpoint was formally assessed in the two gefitinib-first strategies. In each, progression or death after switching to osimertinib counted as an event for the endpoint. The estimate for the osimertinib-first strategy was included as a reference; for that group, progression or death was measured from baseline because osimertinib began at the start of treatment.

The posted primary endpoint results include 45 participants in the osimertinib-first reference group, 47 in the T790M-test-guided sequence, and 44 in the progression-guided sequence. These are the analysis counts associated with the posted primary endpoint values; they should not be treated as a full accounting of all enrolled participants. The registry record does not report the trial’s original sample-size justification, planned allocation ratio, or further details of randomization. It also does not report a detailed statistical analysis plan for the primary endpoint.

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.

Before assignment: The patients were first registered (step 1) to the trial by authorized sites. The site immediately sent the blood samples for circulating free DNA EGFR T790M (cfDNA T790M) testing to the central lab. Once the sample was assessed (step 2), the site was notified whether they could proceed to patient randomization. Patients were then randomized (step 3) after verification of all eligibility criteria.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Per-protocol population defined as patients who have started their study treatment and satisfying all eligibility criteria as per medical review

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.

6. Primary Endpoint

The registry lists the primary outcome as PFS Rate at 18 Months, measured over 18 months after randomization.

The primary endpoint, PFS Rate at 18 Months, is the proportion of participants who were alive and had not experienced a progression-free survival event by RECIST 1.1 at 18 months after randomization while receiving osimertinib. For the two gefitinib-first strategies, the endpoint concerns progression or death after the switch to osimertinib. For the osimertinib-first group, the corresponding estimate is measured from baseline and is presented as a reference.

This endpoint is useful for describing the status of participants at a specified follow-up time while accounting for the trial’s treatment sequences. The registry reports Kaplan–Meier survival-rate estimates with 84% confidence intervals for the two formally assessed groups. A Kaplan–Meier estimate can incorporate follow-up that ends before the endpoint time, under assumptions about censoring. It is therefore not necessarily the same quantity as a simple fraction of all participants who had no event by that time.

The endpoint focuses on progression or death, not treatment duration, response rate, overall survival, or all aspects of disease control. Those outcomes are listed among the trial’s secondary measures, but the registry record does not report their numerical results in the registry record used for this account. The per-protocol analysis population also means the primary estimate describes participants who started treatment and met eligibility criteria after medical review, rather than automatically describing everyone enrolled.

7. Results Posted in the Registry

• Gefitinib Till + Blood Test/Progression Than Osimertinib: 68.1% (32/47) versus 51.1% (23/45) on control; risk difference 17.0 percentage points (95% CI -2.8 to 36.7), risk ratio 1.33 (0.94–1.88); unadjusted two-proportion z-test p = 0.097.
• Gefitinib Till Progression Than Osimertinib: 54.5% (24/44) versus 51.1% (23/45) on control; risk difference 3.4 percentage points (95% CI -17.3 to 24.2), risk ratio 1.07 (0.72–1.58); unadjusted two-proportion z-test p = 0.746.

8. Statistical Interpretation

9. Statistical Methods Explained

The primary endpoint is a time-to-event outcome expressed as a rate at a specified follow-up time. The registry reports Kaplan–Meier survival-rate estimates for the two gefitinib-first strategies, each with an 84% confidence interval. A separate reference estimate is posted for the osimertinib-first strategy. The distinction between these estimates and the recalculated comparisons matters: the former are Kaplan–Meier summaries, while the latter are simple unadjusted calculations from posted counts.

What does a Kaplan–Meier estimate represent?

In general, Kaplan–Meier methods estimate the probability of remaining free of the event over time while allowing participants to have different lengths of observed follow-up. Participants whose event status becomes unknown before the endpoint time may be censored at their last known follow-up. The method relies on assumptions about censoring, including that censoring does not systematically conceal a different event experience. The registry record does not report further details about censoring patterns or checks of those assumptions.

For this trial, the endpoint event is progression according to RECIST 1.1 or death while receiving osimertinib, with the timing defined differently according to the treatment sequence. In the two gefitinib-first groups, the relevant period begins after switching to osimertinib. The osimertinib-first group is measured from baseline and serves as a reference. Consequently, the estimates are not simply three measurements made from a common treatment start point.

What do the recalculated comparisons add?

The recalculated comparisons use event and denominator counts from the posted summaries. For the comparison with the T790M-test-guided strategy, the counts were 23 of 45 in the osimertinib-first group and 32 of 47 in the gefitinib-first group. For the comparison with the progression-guided strategy, they were 23 of 45 and 24 of 44, respectively. These calculations treat the counts as unadjusted proportions and do not reproduce the Kaplan–Meier analysis.

A risk difference describes the difference between two proportions. A risk ratio compares the proportions as a ratio, and an odds ratio compares the odds of the event between groups. These measures summarize different scales of association; they are not interchangeable. The posted recalculated comparisons include risk differences, risk ratios, odds ratios, confidence intervals, and unadjusted p-values. Because they are unadjusted, they do not account for possible differences in prognostic characteristics or other covariates.

A confidence interval describes the range of estimates compatible with the data under the statistical method and its assumptions. In general, a wider interval means less precision. When a confidence interval for a risk difference includes zero, the data are compatible with no difference as well as with differences in either direction at the interval’s confidence level. This does not prove equality, and an interval excluding zero would not by itself establish clinical importance.

What does the p-value say?

A p-value evaluates how compatible the observed data are with a specified null model, under the assumptions of the test. The recalculated unadjusted p-values are 0.097 for the T790M-test-guided comparison and 0.746 for the progression-guided comparison. They are not probabilities that the null hypothesis is true, nor do they indicate the probability that an observed difference will be reproduced. The registry record does not report a sponsor p-value for the posted Kaplan–Meier estimates.

The two recalculated comparisons are exploratory descriptions based on the posted summary counts. The registry record does not report whether the primary endpoint comparisons were intended to be confirmatory, whether there was a prespecified testing hierarchy, or whether multiplicity was addressed. Those details are important when interpreting several comparisons because repeated testing can increase the chance of a false-positive result if no adjustment or other control is used.

Why can over-dispersion matter?

In general, over-dispersion occurs when observed variability is greater than a statistical model expects. It can make standard errors too small and intervals too narrow if the model fails to account for that extra variability. The posted analyses use Kaplan–Meier estimates for the sponsor results and unadjusted count-based recalculations; the registry record does not report an over-dispersion model or a need to apply one. This general concept should not be mistaken for a claim that over-dispersion was present in APPLE.

The registry identifies a per-protocol population defined by treatment initiation and medical review of eligibility. The primary endpoint counts are reported for that population. The record does not report the reasons for exclusion from this analysis, the amount or handling of missing endpoint data, or sensitivity analyses. Without those details, the robustness of the primary endpoint estimates to exclusions or missingness cannot be assessed from the registry information alone.

10. Confidence Intervals and Precision

The primary endpoint estimates are accompanied by 84% confidence intervals. For the T790M-test-guided sequence, the posted estimate is 67.2%, with an interval from 56.4% to 75.9%. For the progression-guided sequence, the estimate is 53.5%, with an interval from 42.3% to 63.5%. The osimertinib-first reference estimate is 51.1%, with an interval from 40.2% to 61.0%. These intervals convey uncertainty in the estimated 18-month rates; they are not ranges containing a fixed proportion of individual participant outcomes.

The recalculated risk-difference intervals for both comparisons include zero. In general, that means the data are compatible with no difference on the risk-difference scale, as well as with differences in either direction within the interval. The intervals are broad, so the recalculations do not pin down the size or direction of a difference precisely. The unadjusted p-values, 0.097 and 0.746, are consistent with that uncertainty but should not replace direct consideration of the estimates and intervals.

Confidence intervals depend on the statistical method and assumptions used. The registry reports an 84% confidence level for the Kaplan–Meier endpoint results; the recalculated comparison intervals are separate and are based on unadjusted summary counts. These confidence intervals should not be treated as if they were calculated by the same method or targeted the same estimand.

Precision is also affected by the analysis group sizes: the primary endpoint results include 45, 47, and 44 participants across the three strategies. The registry record does not report further information about censoring, follow-up completeness, or missing endpoint data. Those factors could affect precision and the assumptions underlying the time-to-event estimates, but their impact cannot be quantified from the registry record.

11. 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 values are planning illustrations based on the observed effect, not the trial’s original design calculations. For these binary-outcome comparisons, the figures are participants per arm, not total events. The planning assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation; the effect to detect is the stated fraction of the observed effect. The registry record does not report the trial’s own original sample-size assumptions or justification.

The 80% power column gives the planned participants per arm for a hypothetical study to have an 80% chance of detecting the specified effect under those assumptions. The 90% power column gives the corresponding number for a 90% chance. These are planning targets under a statistical model, not predictions of what a future study will achieve.

For the T790M-test-guided comparison, planning for 100% of the observed effect gives 131 participants per arm for 80% power and 174 for 90% power. Planning for 75% of the observed effect gives 236 per arm for 80% power and 315 for 90% power. Planning for 50% of the observed effect gives 538 per arm for 80% power and 719 for 90% power.

For the progression-guided comparison, planning for 100% of the observed effect gives 3,316 participants per arm for 80% power and 4,438 for 90% power. Planning for 75% of the observed effect gives 5,901 per arm for 80% power and 7,899 for 90% power. Planning for 50% of the observed effect gives 13,288 per arm for 80% power and 17,788 for 90% power. Smaller effects need more participants to detect under the same assumptions. These figures are not observed power.

12. 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

n is the number of participants in each arm, p₀ and p₁ are the control and treatment proportions, and p̄ is their average.

Worked calculation for this trial

1. Control proportion p₀ = 0.511; treatment proportion p₁ = 0.681; average p̄ = 0.596.
2. First term: z_1−α/2 × √[2p̄(1−p̄)] = 1.96 × 0.6940 = 1.3601.
3. Second term: z_1−β × √[p₀(1−p₀) + p₁(1−p₁)] = 0.842 × 0.6835 = 0.5752.
4. n = (1.3601 + 0.5752)² / (0.1697)² = 130.0, rounded up to 131 participants per arm 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
p0 <- 0.511111111111111                          # control proportion
p1 <- 0.680851063829787                          # treatment proportion
n_per_arm <- function(power, frac = 1) {
  q1 <- p0 + (p1 - p0) * frac               # smaller effect = fraction of the observed difference
  pbar <- (p0 + q1) / 2
  z_b <- qnorm(power)
  ceiling(round((z_a * sqrt(2 * pbar * (1 - pbar)) + z_b * sqrt(p0 * (1 - p0) + q1 * (1 - q1)))^2 / (q1 - p0)^2, 6))
}
for (f in c(1, 0.75, 0.5))
  cat(sprintf("%3.0f%% of observed effect: n per arm = %s (80%% power), %s (90%% power)\n", 100 * f,
              format(n_per_arm(0.80, f), big.mark = ","), format(n_per_arm(0.90, f), big.mark = ",")))

# Check against R's built-in function (unrounded n per arm)
cat("power.prop.test:", round(power.prop.test(p1 = p0, p2 = p1, power = 0.80)$n, 1), "\n")

# Simulation: how often does the z-test reject at the 80%-power sample size?
set.seed(2856893)
n    <- n_per_arm(0.80)
reps <- 20000
x0 <- rbinom(reps, n, p0); x1 <- rbinom(reps, n, p1)
ph <- (x0 + x1) / (2 * n)
z  <- ((x1 - x0) / n) / sqrt(ph * (1 - ph) * 2 / n)
cat(sprintf("Simulated power at n = %s per arm: %.3f (%d replicates)\n", format(n, big.mark = ","), mean(abs(z) > z_a, na.rm = TRUE), reps))

Output from running the code

100% of observed effect: n per arm = 131 (80% power), 174 (90% power)
 75% of observed effect: n per arm = 236 (80% power), 315 (90% power)
 50% of observed effect: n per arm = 538 (80% power), 719 (90% power)
power.prop.test: 130 
Simulated power at n = 131 per arm: 0.805 (20000 replicates)

The simulation draws many pairs of event counts at the planned sample size and counts how often the two-proportion z-test rejects the null. It should land close to the target power; at very small required sample sizes it can fall below it, because the normal approximation is rough there.

13. Analyses Reported by the Sponsor

The registry’s own primary endpoint results are Kaplan–Meier rates with 84% confidence intervals: 67.2% for the T790M-test-guided gefitinib sequence and 53.5% for the progression-guided gefitinib sequence. The osimertinib-first group has a 51.1% reference estimate. The sponsor results are the appropriate posted summaries for the time-to-event endpoint as defined by the registry.

The recalculated comparisons instead use the posted event counts and denominators to form unadjusted proportion comparisons. Their estimates and intervals therefore need not match the Kaplan–Meier results, because they use a different calculation and do not account for event timing or censoring in the same way. The registry record does not flag a discrepancy between the sponsor analysis and the recalculations. Even so, the recalculations should be read as supplementary, unadjusted summaries rather than replacements for the sponsor’s endpoint analysis.

The sponsor’s reported analyses list Kaplan–Meier estimates and 84% confidence intervals for the two gefitinib-first groups, without p-values. The record does not report sponsor-provided direct comparative tests for those estimates. The unadjusted p-values of 0.097 and 0.746 belong to the recalculated comparisons, not to the sponsor’s Kaplan–Meier analyses.

14. Multiplicity, Missing Data and Analysis Populations

The registry record describes a primary endpoint and lists eight secondary outcomes, including PFS while receiving osimertinib, the proportion receiving osimertinib based on positive cfDNA T790M status, time to progression on osimertinib, overall response rate, treatment duration, overall survival, brain PFS, and PFS-2. It also lists two analyses. The available information does not explain how the primary and secondary outcomes were prioritized for statistical testing.

The registry record does not report a multiplicity adjustment, a testing hierarchy, or a family-wise error-control strategy. When several outcomes or comparisons are assessed, multiplicity matters because the chance of at least one false-positive finding can increase if each is tested without an appropriate plan. The record does not establish whether such procedures were planned elsewhere, so their use cannot be assumed.

The primary endpoint analysis is defined for a per-protocol population: participants who started study treatment and satisfied all eligibility criteria after medical review. The registry record does not report a full account of exclusions from that population, the number of participants with missing endpoint data, or the reasons for missingness. It also does not report sensitivity analyses using alternative analysis populations, such as an intention-to-treat population.

For a time-to-event endpoint, censoring is part of the analysis because some participants may not have an observed event by the end of their available follow-up. The registry provides Kaplan–Meier estimates, but further details about follow-up completeness and censoring are not reported in the registry record. The handling of missing covariates, if any, is also not reported in the registry record.

These omissions limit assessment of the analysis process and robustness, particularly for comparisons among multiple treatment strategies. They do not establish that the analyses were conducted incorrectly; they mean that the registry record does not provide enough detail to evaluate those aspects fully.

15. Safety Results

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

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 5% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: covid-19 (3.9%), acute respiratory insufficiency (3.8%), dyspnea (3.8%), bilateral renal infarcts (2.0%), deterioration of health status (2.0%).

The serious adverse event summaries report the number at risk and the number affected in each group. In the osimertinib-until-progression group, 53 participants were at risk and 9 were affected. In the T790M-test-guided gefitinib sequence, 52 were at risk and 6 were affected. In the progression-guided gefitinib sequence, 51 were at risk and 13 were affected. These are counts, not rates adjusted for follow-up time.

The record also reports other adverse events: 52 of 53 at risk in the osimertinib-first group, 51 of 52 in the T790M-test-guided group, and 50 of 51 in the progression-guided group. Death counts were 19, 20, and 18, respectively. The registry record does not report in these summaries whether deaths were attributed to treatment, nor do the counts alone establish cause or comparative risk.

The most frequently listed serious terms include COVID-19 at 3.9%, acute respiratory insufficiency at 3.8%, dyspnea at 3.8%, bilateral renal infarcts at 2.0%, deterioration of health status at 2.0%, and loss of consciousness at 2.0%. These terms are safety descriptions, not evidence that a particular treatment caused an event. The record does not report further detail about event timing, severity grading, or treatment attribution.

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

The primary endpoint results describe the estimated proportion alive and free of a RECIST 1.1 PFS event at 18 months while receiving osimertinib. The largest posted Kaplan–Meier estimate among the strategies was 67.2% for the gefitinib strategy guided by positive plasma T790M testing. Its 84% confidence interval was 56.4% to 75.9%. The progression-guided gefitinib strategy had an estimate of 53.5%, and the osimertinib-first group had a 51.1% reference estimate.

The estimates do not, by themselves, show that the strategies differ reliably. The recalculated unadjusted comparisons have intervals that include zero, with p-values of 0.097 and 0.746. Those results are uncertain and are not proof that the strategies are equivalent. They also do not replace the sponsor’s Kaplan–Meier endpoint summaries, because the recalculations use unadjusted counts rather than the time-to-event method.

The endpoint is limited to PFS while receiving osimertinib under the trial’s sequence-specific definitions. It does not directly answer every question raised by APPLE, including the overall value of liquid biopsy as a monitoring approach, treatment duration, overall survival, or the mechanisms of acquired resistance. Those are distinct objectives or outcomes, and numerical results for them are not reported in the registry record.

These data should be interpreted as trial results, not as individual treatment guidance. The primary analysis is per protocol, the analysis groups are modest in size, and information about missing data, censoring details, multiplicity, and the original design assumptions is not reported in the registry record. The findings are therefore best understood as estimates with material uncertainty and limited information for making broad comparative claims.

17. Why This Trial Matters Statistically

APPLE raises statistical questions that are central to trials of treatment sequencing and biomarker-guided care. Its strategies differ not only in drug order but also in what triggers the switch to osimertinib: a positive plasma T790M result or progression under RECIST 1.1. That makes the endpoint’s time origin and event definition important. The registry explicitly distinguishes the two gefitinib-first groups from the osimertinib-first reference group.

The trial also illustrates the difference between a time-to-event estimate and a simple binary comparison. Kaplan–Meier estimates account for event timing and censored follow-up under their assumptions, whereas an unadjusted count-based comparison treats the posted counts as proportions. Those methods can yield summaries that are not directly interchangeable, even when they concern related clinical questions.

Confidence intervals help keep attention on uncertainty rather than a single point estimate. In APPLE, the posted endpoint intervals are broad, and the unadjusted comparison intervals include zero. This supports cautious interpretation: the numerical pattern is informative, but it does not settle the comparative question. A p-value alone cannot express the range of plausible effect sizes or the clinical relevance of those effects.

Finally, the record highlights why transparent reporting of analysis populations, missingness, censoring, multiplicity, and design assumptions matters. Those details help readers judge the scope and reliability of estimates. Where they are not reported in the registry record, conclusions should remain bounded by the results and methods that are documented.

18. 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.
• Participant counts were reconstructed from posted percentages and arm sizes, so small rounding differences are possible.

The primary analysis uses a per-protocol population defined by treatment initiation and medical review of eligibility. This focuses the estimate on participants who started treatment and met eligibility criteria, but the registry record does not report how many otherwise enrolled participants were excluded or why. The relationship between this analysis population and the randomized population cannot be fully assessed from the available record.

The sponsor’s endpoint estimates are Kaplan–Meier rates, while the recalculated comparisons are based on unadjusted counts. The latter do not account for the timing of events or censoring in the way the Kaplan–Meier method does. They are useful as supplementary summaries but should not be treated as direct substitutes for the sponsor’s time-to-event analyses.

Further details needed to evaluate the endpoint analysis are not reported in the registry record, including censoring patterns, follow-up completeness, missing endpoint data, and sensitivity analyses. Details of the original sample-size justification, randomization procedures, and any adjustment for multiplicity are also not reported in the registry record. The planning illustrations based on observed effects cannot fill those gaps.

The primary endpoint definitions differ by treatment sequence: the gefitinib-first strategies assess progression or death after switching to osimertinib, while the osimertinib-first result is measured from baseline and presented as a reference. This should be kept in mind when comparing the posted estimates. The registry record also lists secondary outcomes but does not report their numerical results in the information used for this account.

Safety summaries provide counts and selected serious-event terms, but the registry record does not report further details about event attribution, timing, or severity grading in these summaries. Counts alone cannot establish whether events were caused by treatment or support a complete comparison of safety across strategies.

19. Related Tutorials

20. Related Statistical Calculators

21. Sources

• ClinicalTrials.gov: Osimertinib Treatment on EGFR T790M Plasma Positive NSCLC Patients (APPLE) — NCT02856893.
• Registry record last updated 2026-06-02; status last verified 2026-05.
• Results first posted 2025-10-31.
• 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.

22. Record Summary

APPLE (NCT02856893) was a completed, randomized, open-label, parallel-group phase II trial sponsored by the European Organisation for Research and Treatment of Cancer - EORTC. The registry lists enrollment of 156, a start date of 2017-10-10, and primary completion on 2022-09-20. The trial studied treatment strategies involving osimertinib and gefitinib in EGFR-mutant advanced NSCLC.

The primary endpoint was the PFS rate at 18 months while receiving osimertinib, assessed formally in the two gefitinib-first strategies and reported as a reference for the osimertinib-first strategy. The posted Kaplan–Meier estimates were 67.2% for the T790M-test-guided sequence and 53.5% for the progression-guided sequence, with 84% confidence intervals. The osimertinib-first reference estimate was 51.1%. The unadjusted recalculations are separate from those sponsor endpoint estimates, and both comparisons have intervals that include zero.