在 R 中创建条件总和（基于日期）作为数据框的新列

Question

我正在尝试在 R 中进行一些特征工程。假设我有以下数据框：

events = data.frame(patient = c("A","A","A","A","B","B","B"), 
                    date = as.Date(c("2017-12-15", "2018-01-09", "2018-01-31", "2018-02-05", 
                                     "2017-12-12", "2017-12-12", "2018-02-01")), 
                    type = c("AnE","Inpatient","Inpatient","Inpatient","AnE","AnE",
                             "Inpatient"))`

我现在想添加一个列，其中包含过去 30 天内来自同一患者的 "Inpatient" 事件的总和。

有没有直接的方法（不涉及 for 循环）？

Answer 1

根据您的数据集，我会创建一些句柄变量和运行 data.table 方法。

首先我添加了患者上次月经的日期。然后，我根据患者和距当前日期早于 30 天的上一期日期计算 "Inpatient" 在数据集中出现的次数。

library(data.table)
events = data.table(patient = c("A","A","A","A","B","B","B"), 
                    date = as.Date(c("2017-12-15", "2018-01-09", "2018-01-31", "2018-02-05", 
                                     "2017-12-12", "2017-12-12", "2018-02-01")), 
                    type = c("AnE","Inpatient","Inpatient","Inpatient","AnE","AnE",
                             "Inpatient"))
events = events[order(date), .SD, by = patient]
events[, date_t1 := lag(date), by = patient]
events[, timesInpatient := cumsum(type=="Inpatient"), by = .(patient, date_t1 > date - 30)]

结果是这样的

   patient       date      type      date1 timesInpatient
1:       B 2017-12-12       AnE       <NA>              0
2:       B 2017-12-12       AnE 2017-12-12              0
3:       B 2018-02-01 Inpatient 2017-12-12              1
4:       A 2017-12-15       AnE       <NA>              0
5:       A 2018-01-09 Inpatient 2017-12-15              1
6:       A 2018-01-31 Inpatient 2018-01-09              2
7:       A 2018-02-05 Inpatient 2018-01-31              3

Answer 2

这可能不如 data.table 方法简洁，但您可以使用 lubridate 包中的 span 和 %within%。

这是它们如何工作的示例：

# creating a span object and a vector of dates
span <- lubridate::interval("2018-01-01", "2018-01-30")
dates <- as.Date(c("2018-01-01", "2018-01-30", "2018-01-03", "2018-02-01"))
dates %within% span
[1]  TRUE  TRUE  TRUE FALSE
# adding a vector indicating inpatient visits
inpatient_visit <- c(TRUE, FALSE, TRUE, FALSE)
# counting dates are both fall within the span and are inpatient visits
sum(dates %within% span & visit)
[1] 2

然后您可以使用 split-apply-combine approach（使用 split 和 purrr:map_df）并为数据集中的每个患者重复此计数过程：

library(dplyr)
library(lubridate)

events = data.frame(patient = c("A","A","A","A","B","B","B"), 
                    date = as.Date(c("2017-12-15", "2018-01-09", "2018-01-31", "2018-02-05", 
                                     "2017-12-12", "2017-12-12", "2018-02-01")), 
                    type = c("AnE","Inpatient","Inpatient","Inpatient","AnE","AnE",
                             "Inpatient"))

count_visits <- function(df) {
  res <- map(df$span, ~ sum(df$date %within% .x & df$inpatient))
  df$count <- res
  return(df)
}

events <- events %>%
  mutate(inpatient = type == "Inpatient",
         span = interval(date - days(30), date)) %>%
  split(.$patient) %>%
  map_df(count_visits) %>%
  select(-inpatient, -span) %>%
  arrange(date)

events
  patient       date      type count
1       B 2017-12-12       AnE     0
2       B 2017-12-12       AnE     0
3       A 2017-12-15       AnE     0
4       A 2018-01-09 Inpatient     1
5       A 2018-01-31 Inpatient     2
6       B 2018-02-01 Inpatient     1
7       A 2018-02-05 Inpatient     3

在 R 中创建条件总和（基于日期）作为数据框的新列

Creating conditional sums (based on dates) as new column of data frame in R

r

dplyr

tidyverse