9.2 Statistical Methods

9.2.1 Justification of sample size

9.2.1.1 Objective response rate

The primary endpoint of the trial is the objective tumor
response rate. Patients will be ~~randon1ized ~~randomized between
250 mg and 500 mg daily dose levels of ZDI839 and stratified by ethnicity as
Japanese versus non-Japanese.
The trial is sized
~~sired~~~~designed????~~~~ ~~to
independently evaluate the tumor response rate in the four strata defined by
ZDI839 dose and ethnicity.

Within each strata, the goal is to have 90% power for a
2-sided 5% significance level test ~~so~~ that
the response rate is greater than 5% when the true response rate is 20%. This
requires a total of 45 patients evaluable for response per strata~~strata~~~~stratum~~~~.~~
It will be concluded that the response rate within a ~~strata ~~strata~~stratum~~ is greater than 5% if there are at
least 6 responses in 45 patients (13.3% observed rate, 95% exact confidence
interval 5.1% to 26.8%).

Assuming 10% of patients are not evaluable for response, a
total of 100 Japanese patients and 100 non-Japanese patients ~~win ~~will be randomized
to obtain 45 patients evaluable for response in each of the 4 strata.

With the 2 ethnicity strata combined, the trial will have 90
patients evaluable for response per ZDI839 dose level. Assuming an overall
response rate of 20%
for each dose level, a 95% exact confidence interval for the overall response
rate can be calculated for each dose level independently with a width of +/-
10% (Fleiss,~~F1ciss,~~~~???~~~~ ~~Joseph
L 1981).

__Study IL0039__

9.2 Statistical methods

9.2.1 Justification of sample size

9.2.1.1 Objective response rate and symptom improvement rate

Patients will be randomized equally between 250 and 500 mg
ZD1839. The two co-primary endpoints are the objective tumor response rate and
symptom improvement rate, ~~ ~~which will be analyzed on an
intent-to-treat basis. The trial is sized to independently evaluate the two
co-primary endpoints for each ZD1839 dose. For both the objective tumor
response rate and symptom improvement rate, the goal is to have a power of .90
for a 1-sided .0125 significance level test that the rate is £5 % when the true rate is 15%. This requires that 200
patients be randomized to obtain 100 per dose of ZD 1839 (Fleiss 1981). This
design achieves a 1-sided .0115 significance level and power of .901. The two
doses of ZD 1839 will be compared with respect to the two co-primary endpoints
with Fisher's exact lest.

9.2.2 Primary endpoints

· Objective tumor response (complete + partial response} using Southwest Oncology Group (SWOG) modified UICCIWHO criteria

· Symptom improvement rate as measured by the Lung Cancer Subscale (LCS) of the FACT-L

9.2.3 Secondary endpoints

· Disease control rate (complete + partial response + stable disease}

· Progression-free survival, overall survival

· Frequency and severity of adverse events (AEs)

· Changes in QOL using the FACT -L including time to worsening

· Trough concentrations of ZDI839

9.2.4 Exploratory endpoint

· .EGFR expression

9.2.5 Definitions of trial populations used in the analysis

Two populations, intent-to-treat, per-protocol, are defined in Table 4 below:

9.2.6 Assessment of efficacy

The primary analysis of the overall best objective tumor response rate will be based on the investigator's assessment. Data from the REC's assessment will be used to corroborate the analysis of objective tumor response.

The primary analysis population for the overall best objective tumor response rate and symptom improvement rate per LCS will be the intent-to-treat population. To assess population sensitivity, the tumor response rate will also be analyzed in the per-protocol population, and the symptom improvement rate will also be analyzed in the per-protocol population.

For each dose of ZD1839, Hochberg's procedure will be used to
maintain an overall 1~~I ~~sided
.025 significance level for the two co-primary endpoints. The I-sided
significance level for each endpoint will be calculated as the probability of
the observed number or greater events (objective responses or symptom
improvements) given the sample size assuming a true event rate of 5%. If the
larger of the two significance levels is ~~oS .~~???≤ .025, then it
can be concluded that the event rate for both co-primary endpoints is > 5%.
Otherwise, if the smaller of the two significance levels is £ .0125,
then it can be concluded that the event rate for the corresponding endpoint
alone is > 5%. Given exactly 100 patients per dose, 11 events are required
to conclude that the event rate for a dose is > 5% at both a I-sided .025
and .0125 significance level (11.0% observed rate, 95% c.i. 5.6% to 18.8%,
97.5% c.i.s 5.1% to 20.0%).

The response rate and symptom improvement rate will be estimated separately by dose. Exact 95% confidence intervals (c.i.) will be calculated for each rate.

The two doses ofZDI839 will be compared with respect to the two co-primary endpoints with Fisher's exact test. Multivariate logistic regression models will be used to further explore a significant difference.

Overall survival, progression-free survival, and time to worsening of symptoms per LCS will be analyzed in the intent-to-treat and per-protocol populations.

Kaplan-Meier plots will be calculated by dose (Kaplan and Meier 1958) Non-parametric 95% c.i..s will be calculated for the median event times (Brookmeyer and Crowley 1982). .The two doses will be compared with respect to these endpoints with an unadjusted logrank test at a 1-sided .025 significance level.

Multivariate proportional hazard models will be used to further explore significant levels.

Disease control will be based on objective tumor assessments and will include those patients responding complete response and partial response (CR and PR), plus those patients with stable disease, confirmed and sustained for at least 4 weeks.