Time to Event Summary

Time to Event Analysis with Kaplan-Meier Estimates

Setup

See Prerequisites for installation instructions.

library(arframe)
library(pharmaverseadam)
library(dplyr, warn.conflicts = FALSE)
library(tidyr)
library(survival)

adtte <- pharmaverseadam::adtte_onco |>
  blank_to_na() |>
  filter(PARAMCD == "OS", ARM %in% c("Xanomeline High Dose", "Placebo"))

arm_levels <- c("Xanomeline High Dose", "Placebo")
arm_n <- adtte |>
  count(ARM) |> pull(n, name = ARM)
arm_n <- arm_n[arm_levels]

Data Preparation

n_pct <- function(n, N) sprintf("%d (%.1f)", n, n / N * 100)

# ── Events and Censoring ──
events_by_arm <- adtte |>
  group_by(ARM) |>
  summarise(
    n_events   = sum(CNSR == 0),
    n_censored = sum(CNSR == 1),
    N          = n(),
    .groups    = "drop"
  )

events_row <- tibble(
  category   = "",
  stat_label = "Events (n (%))"
) |> bind_cols(
  events_by_arm |>
    mutate(value = n_pct(n_events, N)) |>
    select(ARM, value) |>
    pivot_wider(names_from = ARM, values_from = value)
)

censored_row <- tibble(
  category   = "",
  stat_label = "Censored (n (%))"
) |> bind_cols(
  events_by_arm |>
    mutate(value = n_pct(n_censored, N)) |>
    select(ARM, value) |>
    pivot_wider(names_from = ARM, values_from = value)
)

# ── Kaplan-Meier estimates ──
km_fit <- survfit(Surv(AVAL, 1 - CNSR) ~ ARM, data = adtte)
km_summary <- summary(km_fit)

# Extract median and quartiles per arm
km_stats <- lapply(arm_levels, function(a) {
  arm_data <- adtte |> filter(ARM == a)
  fit_arm  <- survfit(Surv(AVAL, 1 - CNSR) ~ 1, data = arm_data)
  q_fit    <- quantile(fit_arm, probs = c(0.25, 0.50, 0.75))

  tibble(
    ARM = a,
    q25    = q_fit$quantile[1],
    median = q_fit$quantile[2],
    q75    = q_fit$quantile[3]
  )
}) |> bind_rows()

fmt_ci <- function(est, lower, upper) {
  if (is.na(est)) return("NE")
  sprintf("%.1f", est)
}

km_rows <- bind_rows(
  tibble(category = "Time to Event (months)", stat_label = "Median") |>
    bind_cols(
      km_stats |>
        mutate(value = vapply(median, function(x)
          if (is.na(x)) "NE" else sprintf("%.1f", x), character(1))) |>
        select(ARM, value) |>
        pivot_wider(names_from = ARM, values_from = value)
    ),
  tibble(category = "Time to Event (months)", stat_label = "25th Percentile") |>
    bind_cols(
      km_stats |>
        mutate(value = vapply(q25, function(x)
          if (is.na(x)) "NE" else sprintf("%.1f", x), character(1))) |>
        select(ARM, value) |>
        pivot_wider(names_from = ARM, values_from = value)
    ),
  tibble(category = "Time to Event (months)", stat_label = "75th Percentile") |>
    bind_cols(
      km_stats |>
        mutate(value = vapply(q75, function(x)
          if (is.na(x)) "NE" else sprintf("%.1f", x), character(1))) |>
        select(ARM, value) |>
        pivot_wider(names_from = ARM, values_from = value)
    )
)

# ── Cox PH: Hazard Ratio ──
cox_fit <- coxph(Surv(AVAL, 1 - CNSR) ~ ARM, data = adtte)
cox_sum <- summary(cox_fit)
hr      <- cox_sum$conf.int[1, 1]
hr_lo   <- cox_sum$conf.int[1, 3]
hr_hi   <- cox_sum$conf.int[1, 4]

# ── Log-rank test ──
lr_test <- survdiff(Surv(AVAL, 1 - CNSR) ~ ARM, data = adtte)
lr_pval <- 1 - pchisq(lr_test$chisq, df = 1)

hr_row <- tibble(
  category   = "Comparison",
  stat_label = "Hazard Ratio (95% CI)",
  !!arm_levels[1] := sprintf("%.3f (%.3f, %.3f)", hr, hr_lo, hr_hi),
  !!arm_levels[2] := ""
)

pval_row <- tibble(
  category   = "Comparison",
  stat_label = "Log-Rank p-value",
  !!arm_levels[1] := if (lr_pval < 0.001) "<0.001" else sprintf("%.3f", lr_pval),
  !!arm_levels[2] := ""
)

# ── Combine ──
tte_wide <- bind_rows(events_row, censored_row, km_rows, hr_row, pval_row) |>
  mutate(across(all_of(arm_levels), ~ replace_na(.x, "")))

arframe Pipeline

tte_wide |>
  fr_table() |>
  fr_titles(
    "Table 14.2.1",
    "Time to Event — Overall Survival",
    "ITT Population"
  ) |>
  fr_cols(
    category   = fr_col(visible = FALSE),
    stat_label = fr_col("Parameter", width = 2.5),
    !!!setNames(
      lapply(arm_levels, function(a) fr_col(a, align = "decimal")),
      arm_levels
    ),
    .n = arm_n
  ) |>
  fr_header(bold = TRUE, align = "center") |>
  fr_rows(
    group_by    = list(cols = "category", label = "stat_label"),
    blank_after = "category",
    group_style = list(bold = TRUE)
  ) |>
  fr_styles(
    fr_row_style(rows = which(tte_wide$stat_label %in%
      c("Hazard Ratio (95% CI)", "Log-Rank p-value")), bold = TRUE)
  ) |>
  fr_footnotes(
    "NE = Not Estimable (fewer than required events reached).",
    "Hazard Ratio: Drug X vs Placebo from Cox proportional hazards model.",
    "p-value from unstratified log-rank test.",
    "CDISCPILOT01 Oncology Extension — Overall Survival."
  )

Rendered Table

Table 14.2.1

Time to Event — Overall Survival

ITT Population

Parameter	Placebo (N=86)	Xanomeline High Dose (N=84)
Events (n (%))	2 ( 2.3)	0 ( 0.0)
Censored (n (%))	84 (97.7)	84 (100.0)
Time to Event (months)
Median
25th Percentile
75th Percentile
Comparison
Hazard Ratio (95% CI)
Log-Rank p-value		0.244

NE = Not Estimable (fewer than required events reached).

Hazard Ratio: Drug X vs Placebo from Cox proportional hazards model.

p-value from unstratified log-rank test.

CDISCPILOT01 Oncology Extension — Overall Survival.

/opt/quarto/share/rmd/rmd.R 01APR2026 09:54:27