In dominicmagirr/modestWLRT: Functions for implementing modestly-weighted logrank tests.

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Installation

You can use devtools::install_github() to get the package from GitHub:

install.packages("devtools")
library(devtools)
install_github("dominicmagirr/modestWLRT")

Load packages

library(dplyr)
library(ggplot2)
library(modestWLRT)

library(dplyr)
library(ggplot2)
devtools::load_all("..")

Simulate example data set

You can use the function delayed_effect_sim to simulate an example data set from a 2-arm RCT. Survival times on the control arm are exponentially distributed with median med_c. Survival times on the experimental arm follow a 2-piece exponential distibution: from time zero up to time delay the event rate is rate_e_1; thereafter the event rate is rate_e_2. Patient recruitment times follow a simple power distribution:

pr(recruited before t) = (t / rec_period)^rec_power, for t in (0, rec_period).

Data cut-off happens at time max_cal_t, and any patients still alive have their survival time censored.

example_data = delayed_effect_sim(n_c = 10,
                                  n_e = 10,
                                  rec_period = 12,
                                  rec_power = 1,
                                  med_c = 15,
                                  rate_e_1 = log(2) / 15,
                                  rate_e_2 = 0.03,
                                  delay = 6,
                                  max_cal_t  = 36)

example_data

Risk table

The function get_risk_table takes a data frame produced from delayed_effect_sim (or a data frame of the same form) and turns it into a risk table. This tells you how many patients were at risk / had an event / censored on each arm, at each event time.

example_risk_table = get_risk_table(example_data)

example_risk_table

Calculate weights

From the risk table, you can calculate the scores / weights from a modestWLRT. The argument delay is used to specify how long the scores are kept constant. See the paper

http://arxiv.org/abs/1807.11097

for details.

modest_weights = add_weights(example_risk_table, 
                             method = "fixed_c", 
                             delay = 12, 
                             plot_weights = TRUE)

modest_weights$risk_table

modest_weights$p

Test statistics

Given the risk table with the corresponding weights, it is simple to calculate the standardized weighted logrank statistic. Larger values of Z correspond to longer survival times on the experimental arm.

get_zs(modest_weights)

dominicmagirr/modestWLRT documentation built on Sept. 16, 2020, 2:43 p.m.