Skip to contents

Simulated survival dataset with delayed treatment effect

Source: R/mb_delayed_effect.R
mb_delayed_effect.Rd

Magirr and Burman (2019) considered several scenarios for their modestly weighted logrank test. One of these had a delayed treatment effect with a hazard ratio of 1 for 6 months followed by a hazard ratio of 1/2 thereafter. The scenario enrolled 200 patients uniformly over 12 months and cut data for analysis 36 months after enrollment was opened. This dataset was generated by the sim_pw_surv() function under the above scenario.

Usage

mb_delayed_effect

Format

A data frame with 200 rows and 4 columns:

  • tte: Time to event.

References

Magirr, Dominic, and Carl‐Fredrik Burman. 2019. "Modestly weighted logrank tests." Statistics in Medicine 38 (20): 3782--3790.

Examples

library(survival)

fit <- survfit(Surv(tte, event) ~ treatment, data = mb_delayed_effect)

# Plot survival
plot(fit, lty = 1:2)
legend("topright", legend = c("control", "experimental"), lty = 1:2)


# Set up time, event, number of event dataset for testing
# with arbitrary weights
ten <- mb_delayed_effect |> counting_process(arm = "experimental")
head(ten)
#>   stratum events n_event_tol        tte n_risk_tol n_risk_trt     s  o_minus_e
#> 1     All      1           1 0.07659251        200        100 1.000  0.5000000
#> 2     All      1           0 0.49067015        199         99 0.995 -0.4974874
#> 3     All      1           1 0.65465035        198         99 0.990  0.5000000
#> 4     All      1           0 0.65906384        197         98 0.985 -0.4974619
#> 5     All      1           1 0.81945349        196         98 0.980  0.5000000
#> 6     All      1           0 0.82788909        195         97 0.975 -0.4974359
#>   var_o_minus_e
#> 1     0.2500000
#> 2     0.2499937
#> 3     0.2500000
#> 4     0.2499936
#> 5     0.2500000
#> 6     0.2499934

# MaxCombo with logrank, FH(0,1), FH(1,1)
mb_delayed_effect |>
  maxcombo(rho = c(0, 0, 1), gamma = c(0, 1, 1), return_corr = TRUE)
#> $method
#> [1] "MaxCombo"
#> 
#> $parameter
#> [1] "FH(0, 0) + FH(0, 1) + FH(1, 1)"
#> 
#> $z
#> [1] -2.473248 -2.424018 -2.482653
#> 
#> $corr
#>          v1        v2        v3
#> 1 1.0000000 0.8606625 0.9312916
#> 2 0.8606625 1.0000000 0.9579831
#> 3 0.9312916 0.9579831 1.0000000
#> 
#> $p_value
#> [1] 0.01104234
#> 

# Generate another dataset
ds <- sim_pw_surv(
  n = 200,
  enroll_rate = data.frame(rate = 200 / 12, duration = 12),
  fail_rate = data.frame(
    stratum = c("All", "All", "All"),
    period = c(1, 1, 2),
    treatment = c("control", "experimental", "experimental"),
    duration = c(42, 6, 36),
    rate = c(log(2) / 15, log(2) / 15, log(2) / 15 * 0.6)
  ),
  dropout_rate = data.frame(
    stratum = c("All", "All"),
    period = c(1, 1),
    treatment = c("control", "experimental"),
    duration = c(42, 42),
    rate = c(0, 0)
  )
)
# Cut data at 24 months after final enrollment
mb_delayed_effect_2 <- ds |> cut_data_by_date(max(ds$enroll_time) + 24)