标记持续观察并创建入学率

Question

我有一些大型注册数据集，我正在尝试创建两件事：

1. 我想标记每个不间断的每月观察（ final_df1 ），
2. 我想创建一个不间断跨度的数据集（< 我觉得{lubritation}和{data.table}之间

例如：

library(tidyverse)
library(lubridate)
library(magrittr)
df<-tibble(id=c(rep("X",10),rep("Y",20)),
           date=c(ymd("20120101")%m+%months(c(1:5,7:11)),ymd("20120401")%m+%months(c(1:10,12:17,19:22))))

final_df1 <- df %>% mutate(cont_enroll=c(rep(1,5),rep(0,5),rep(1,10),rep(0,10)))

final_df2 <- tibble(id=c(rep("X",2),rep("Y",3)),
                  span_start=c(ymd("20120101")%m+%months(1),
                               ymd("20120101")%m+%months(7),
                               ymd("20120401")%m+%months(1),
                               ymd("20120101")%m+%months(12),
                               ymd("20120101")%m+%months(19)),
                  span_end=c(ymd("20120101")%m+%months(5),
                             ymd("20120101")%m+%months(11),
                             ymd("20120101")%m+%months(10),
                             ymd("20120101")%m+%months(17),
                             ymd("20120101")%m+%months(22))
                  )

必须有一种简单的方法来执行此操作，但我觉得我正在绘制空白。有技巧吗？

Answer 1

由“ ID”分组，创建一个 Interval ，其先前的“日期”（ lag ）和当前的“日期”，除以月份，检查它是否小于2，然后采取累积最小值（ Cummin ）。创建“ find_df_new'之后，然后我们按'id'和'cont_enroll'列的运行长度ID进行分组，总结 first 和>最后“日期”的值分别创建“ span_start”和“ span_end”

library(dplyr)
library(lubridate)
library(data.table)
final_df_new <- df %>%
   group_by(id)  %>%
   mutate(cont_enroll2 = cummin(interval(lag(date, default = first(date)), 
       date) /months(1) < 2))  %>%
   ungroup
final_df_new %>% 
  group_by(id, grp = rleid(cont_enroll2))  %>%
  summarise(span_start = first(date), span_end = last(date), .groups = 'drop')

Answer 2

我认为您可以用 ivs 软件包来解决这个问题。您的日期似乎确实代表了1个月的间隔，而IVS软件包专门用于处理此类数据。

我们可以用 final_df2 用 iv_groups（）来返回合并所有重叠间隔后保留的非重叠间隔。

然后，每个组的第一行 final_df2 表示第一个连续间隔，因此您只需要检查每个范围是否在该间隔内，即可确定它是否是不间断设置的一部分以获取 final_df1 。

请注意，我的 final_df2 看起来与您的有可能在编码方式上有错误吗？

library(dplyr)
library(lubridate)
library(ivs)

df <- tibble(
  id = c(
    rep("X", 10), 
    rep("Y", 20)
  ),
  date = c(
    ymd("20120101") %m+% months(c(1:5,7:11)), 
    ymd("20120401") %m+% months(c(1:10,12:17,19:22))
  )
)
df
#> # A tibble: 30 × 2
#>    id    date      
#>    <chr> <date>    
#>  1 X     2012-02-01
#>  2 X     2012-03-01
#>  3 X     2012-04-01
#>  4 X     2012-05-01
#>  5 X     2012-06-01
#>  6 X     2012-08-01
#>  7 X     2012-09-01
#>  8 X     2012-10-01
#>  9 X     2012-11-01
#> 10 X     2012-12-01
#> # … with 20 more rows

df <- df %>%
  mutate(start = date, end = date + months(1), .keep = "unused") %>%
  mutate(range = iv(start, end), .keep = "unused")

df
#> # A tibble: 30 × 2
#>    id                       range
#>    <chr>               <iv<date>>
#>  1 X     [2012-02-01, 2012-03-01)
#>  2 X     [2012-03-01, 2012-04-01)
#>  3 X     [2012-04-01, 2012-05-01)
#>  4 X     [2012-05-01, 2012-06-01)
#>  5 X     [2012-06-01, 2012-07-01)
#>  6 X     [2012-08-01, 2012-09-01)
#>  7 X     [2012-09-01, 2012-10-01)
#>  8 X     [2012-10-01, 2012-11-01)
#>  9 X     [2012-11-01, 2012-12-01)
#> 10 X     [2012-12-01, 2013-01-01)
#> # … with 20 more rows

# `iv_groups()` returns the groups that remain after merging all overlapping ranges.
# It gives you `final_df2`.
continuous <- df %>%
  group_by(id) %>%
  summarise(range = iv_groups(range), .groups = "drop")

continuous
#> # A tibble: 5 × 2
#>   id                       range
#>   <chr>               <iv<date>>
#> 1 X     [2012-02-01, 2012-07-01)
#> 2 X     [2012-08-01, 2013-01-01)
#> 3 Y     [2012-05-01, 2013-03-01)
#> 4 Y     [2013-04-01, 2013-10-01)
#> 5 Y     [2013-11-01, 2014-03-01)

# The first continuous range per id
first_continuous <- continuous %>%
  group_by(id) %>%
  slice(1) %>%
  ungroup() %>%
  rename(range_continuous = range)

first_continuous
#> # A tibble: 2 × 2
#>   id            range_continuous
#>   <chr>               <iv<date>>
#> 1 X     [2012-02-01, 2012-07-01)
#> 2 Y     [2012-05-01, 2013-03-01)

# Join the first continuous range df back onto the original df and see if
# the current `range` falls within the first continuous range or not.
# This gives you `final_df1`.
left_join(df, first_continuous, by = "id") %>%
  mutate(continuous = iv_pairwise_overlaps(range, range_continuous, type = "within"))
#> # A tibble: 30 × 4
#>    id                       range         range_continuous continuous
#>    <chr>               <iv<date>>               <iv<date>> <lgl>     
#>  1 X     [2012-02-01, 2012-03-01) [2012-02-01, 2012-07-01) TRUE      
#>  2 X     [2012-03-01, 2012-04-01) [2012-02-01, 2012-07-01) TRUE      
#>  3 X     [2012-04-01, 2012-05-01) [2012-02-01, 2012-07-01) TRUE      
#>  4 X     [2012-05-01, 2012-06-01) [2012-02-01, 2012-07-01) TRUE      
#>  5 X     [2012-06-01, 2012-07-01) [2012-02-01, 2012-07-01) TRUE      
#>  6 X     [2012-08-01, 2012-09-01) [2012-02-01, 2012-07-01) FALSE     
#>  7 X     [2012-09-01, 2012-10-01) [2012-02-01, 2012-07-01) FALSE     
#>  8 X     [2012-10-01, 2012-11-01) [2012-02-01, 2012-07-01) FALSE     
#>  9 X     [2012-11-01, 2012-12-01) [2012-02-01, 2012-07-01) FALSE     
#> 10 X     [2012-12-01, 2013-01-01) [2012-02-01, 2012-07-01) FALSE     
#> # … with 20 more rows

^由