因果推断与生存分析

```{r}
# Install required packages if not already installed

# Load necessary libraries
library(ggplot2)

# Create a dataset similar to the image
data <- data.frame(
  Patient = factor(c("Patient 1", "Patient 2", "Patient 3", "Patient 4", "Patient 5", "Patient 6"),
    levels = rev(c("Patient 1", "Patient 2", "Patient 3", "Patient 4", "Patient 5", "Patient 6"))
  ),
  Entry = c(2000, 2000, 2002, 2002, 2003, 2001), # Year of entry
  Exit = c(2008, 2005, 2007, 2007, 2004, 2005), # Year of follow-up end
  Event = c(0, 1, 0, 0, 1, 1) # 1 = Event occurred, 0 = Censored
)

# Define the accrual and follow-up periods
accrual_start <- 2000
accrual_end <- 2002
followup_end <- 2008

# Create the plot
ggplot(data, aes(y = Patient)) +
  # Add follow-up lines
  geom_segment(aes(x = Entry, xend = Exit, y = Patient, yend = Patient), color = "black") +
  # Add entry points
  geom_point(aes(x = Entry, y = Patient), shape = 16, size = 3) +
  # Add censoring (open circles) and event markers (crosses)
  geom_point(aes(x = Exit, y = Patient, shape = as.factor(Event)), size = 3) +
  scale_shape_manual(values = c(1, 4), labels = c("Censored", "Event")) + # 1 = Open circle, 4 = Cross
  # Add vertical lines for accrual and follow-up periods
  geom_vline(xintercept = c(accrual_start, accrual_end, followup_end), linetype = "dashed", color = "red") +
  # Labels and themes
  labs(x = "Year of entry – calendar time", y = "", shape = "Outcome") +
  theme_minimal() +
  theme(axis.text.y = element_text(size = 12), legend.position = "top")
```

NHEFS¹

变量名	解释
death	是否在1992.12前死亡(1:是,0:否)
time	存活时间,NA:Administrative censoring
qsmk	处理变量,是否停止吸烟(1:是,0:否)
age	年龄
smokeyrs	烟龄

介绍

一般来说,因果推断问题中我们都是在估计treatment对outcome的effect,只不过在生存分析领域,outcome是实验开始到事件发生的时间长度,举例来说,我们对停止吸烟于寿命的因果效应感兴趣(当然,已经有数不清的实验数据表明两者有强相关性,但正如Fisher那个经典的质疑,可能存在某种遗传或环境因素，既导致人们更容易吸烟，又增加了患肺癌的风险).

当然,虽然名为生存分析,事件并不局限于死亡,其实”failure time analysis”这个名字我以为更恰当些,我们关注的是实验开始到发生事件的时间,事件可以是婚姻,癌症,感染甚至找到工作的时间.简化起见,本文只分析固定的treatment,vary的情况日后再说.

hazards 与 risks

Footnotes

1. 一项持续数十年的跟踪实验,数据来源