Consider a hypothetical single-arm phase II trial evaluating the objective response rate (ORR) of a new treatment. The historical ORR is 20%, while 40% is considered a promising response rate. The study uses a binary endpoint (response or no response) and a Beta(0.2, 0.8) prior.
The trial plans one interim analysis after 20 patients, with a maximum sample size of 40. Efficacy and futility decisions are based on the posterior probability that the ORR exceeds the historical rate of 20%:
| Analysis | Decision rule |
|---|---|
| Interim | Stop for efficacy if the posterior probability is > 0.98 |
| Interim | Stop for futility if the posterior probability is <= 0.50 |
| Interim | Otherwise, continue enrollment |
| Final, if continued | Declare success if the posterior probability is > 0.95 |
These sample sizes and thresholds are illustrative. In practice, they should be specified before enrollment and assessed against the trial’s desired type I error rate and power.
Use oc_singlearm_binary() to estimate the operating
characteristics of this two-stage design. We consider a true ORR of 20%
for the null scenario and 40% for the alternative scenario.
true_rates <- c(0.20, 0.40)
oc_results <- lapply(true_rates, function(theta) {
oc_singlearm_binary(
theta_true = theta,
n_current = 20,
N_total = 40,
alpha0 = 0.2,
beta0 = 0.8,
theta0_null = 0.20,
gamma_early = 0.98,
gamma_futility = 0.50,
gamma_final = 0.95,
ntrial = 10000,
seed = 666
)
})
oc_table <- data.frame(
true_ORR = true_rates,
do.call(rbind, lapply(oc_results, unlist)),
row.names = NULL
)
knitr::kable(oc_table, digits = 3)| true_ORR | power_null_post | early_stop_rate_futility | early_stop_rate_efficacy |
|---|---|---|---|
| 0.2 | 0.044 | 0.625 | 0.010 |
| 0.4 | 0.862 | 0.050 | 0.411 |
The power_null_post column gives the overall probability
of declaring the trial successful, including early efficacy stopping. It
estimates the type I error rate when the true ORR is
20%, and power when the true ORR is 40%. The two
remaining columns give the probabilities of stopping early for futility
or efficacy.
The 40% rate is used here to evaluate power, not as an additional requirement for declaring trial success. If the operating characteristics do not meet the design targets, revise the sample sizes or thresholds and evaluate the design again. The function evaluates specified designs; it does not calibrate them automatically.
Suppose 9 of the first 20 patients respond. At the interim analysis, calculate the posterior probability that the ORR exceeds the historical reference rate of 20%:
posterior <- posterior_predictive_singlearm_binary(
x_obs = 9,
n_current = 20,
N_total = 40,
alpha0 = 0.2,
beta0 = 0.8,
theta0_null = 0.20
)
posterior$p_null
#> [1] 0.9922663The posterior distribution is Beta(9.2, 11.8), and the probability of exceeding 20% is approximately 0.992. Apply the prespecified interim stopping rules to the observed data:
decision <- if (posterior$p_null > 0.98) {
"Stop for efficacy"
} else if (posterior$p_null <= 0.50) {
"Stop for futility"
} else {
"Continue to the final analysis"
}
decision
#> [1] "Stop for efficacy"The posterior probability exceeds the efficacy stopping threshold of 0.98, so the trial stops early for efficacy under this design. For trials that continue, final success requires the posterior probability of exceeding 20% to be greater than 0.95.
For illustration, we also assess the probability of meeting the final success criterion if the trial were to continue to 40 evaluable patients, despite meeting the early efficacy stopping rule:
predictive <- posterior_predictive_singlearm_binary(
x_obs = 9,
n_current = 20,
N_total = 40,
alpha0 = 0.2,
beta0 = 0.8,
theta0_null = 0.20,
gamma_final = 0.95,
type = "predictive",
S = 10000,
seed = 666
)
predictive$ppos_null
#> [1] 0.9635Here, ppos_null is the posterior predictive probability
that the final posterior probability will exceed 0.95. This is a
supplementary assessment and does not determine stopping in the current
design. The function defaults to type = "posterior";
setting type = "both" returns posterior summaries and
predictive probabilities together.
The predictive function also reports ppos_clin, the
probability of meeting a separate posterior criterion for a clinically
relevant ORR (theta0_clin, default 0.40). Its final
posterior threshold is gamma_clin_final (default 0.90).
This output can be used as an additional summary but is
not included in the success and stopping rules
evaluated by oc_singlearm_binary().
To adapt the example to another study, specify its historical response rate, prior, interim and maximum sample sizes, and decision thresholds. Reassess operating characteristics before using the design in a trial.