在组内分离非重叠间隔并在 R 中计数

Question

使用 R，我拥有按 DNA 菌株（病原体的）、住院诊所和重叠的住院时间分组的住院患者数据，以确定是否可能传播。

我需要对重叠的组进行顺序编号。这看起来很简单，但有两个问题：

1. 我在 SO 或其他地方找到的所有内容都在讨论组内的行编号。我需要一个组中的每一行都具有相同的数字，并且需要对组本身进行计数。
2. 使用%>% group_by(strain, clinic) %>% 最初看起来很简单的任何方法都可以实现，但这并没有考虑到不重叠的时间间隔

我已经尝试了几种方法和搜索，然后最终放弃并在此处发布（我的任何尝试都不值得在此处发布事件以浪费您的时间。）以下代码是我拥有的数据的示例（have）和我想要的数据 (want)。注意菌株B，所有患者都在Clinic_1，但由于时间间隔分开，分为两组。

任何建议将不胜感激。

have <- data.frame(id=c("K01","K02","K03","K04","K05","K06","K07","K08","K09"),
                   strain=c(rep("A",4),rep("B",5)),
                   clinic=c(rep("Clinic_1",2),rep("Clinic_2",2),rep("Clinic_1",5)),
                   datein=as.Date(c("2020/01/01","2020/01/03","2020/02/03","2020/02/09","2020/02/18","2020/02/20","2020/02/21","2020/03/06","2020/03/18")),
                   dateout=as.Date(c("2020/01/05","2020/01/16","2020/02/09","2020/02/19","2020/02/27","2020/02/23","2020/02/22","2020/03/21","2020/03/22"))
                   )

want <- data.frame(have,overlap_number=c(1,1,2,2,3,3,3,4,4))

#How the final data would look
> View(want)
   id strain   clinic     datein    dateout overlap_number
1 K01      A Clinic_1 2020-01-01 2020-01-05              1
2 K02      A Clinic_1 2020-01-03 2020-01-16              1
3 K03      A Clinic_2 2020-02-03 2020-02-09              2
4 K04      A Clinic_2 2020-02-09 2020-02-19              2
5 K05      B Clinic_1 2020-02-18 2020-02-27              3
6 K06      B Clinic_1 2020-02-20 2020-02-23              3
7 K07      B Clinic_1 2020-02-21 2020-02-22              3
8 K08      B Clinic_1 2020-03-06 2020-03-21              4
9 K09      B Clinic_1 2020-03-18 2020-03-22              4

基于 Akrun 评论的替代数据集，K07 的日期略有变化：

have2 <- data.frame(id=c("K01","K02","K03","K04","K05","K06","K07","K08","K09"),
                   strain=c(rep("A",4),rep("B",5)),
                   clinic=c(rep("Clinic_1",2),rep("Clinic_2",2),rep("Clinic_1",5)),
                   datein=as.Date(c("2020/01/01","2020/01/03","2020/02/03","2020/02/09","2020/02/18","2020/02/20","2020/02/25","2020/03/06","2020/03/18")),
                   dateout=as.Date(c("2020/01/05","2020/01/16","2020/02/09","2020/02/19","2020/02/27","2020/02/23","2020/02/29","2020/03/21","2020/03/22"))
                   )

#Output:

#> have2 %>% 
#+   mutate(overlap_number = rleid(strain, clinic, 
#+                                 cumsum(datein > lag(dateout, default = #first(dateout)))))
#   id strain   clinic     datein    dateout overlap_number
#1 K01      A Clinic_1 2020-01-01 2020-01-05              1
#2 K02      A Clinic_1 2020-01-03 2020-01-16              1
#3 K03      A Clinic_2 2020-02-03 2020-02-09              2
#4 K04      A Clinic_2 2020-02-09 2020-02-19              2
#5 K05      B Clinic_1 2020-02-18 2020-02-27              3
#6 K06      B Clinic_1 2020-02-20 2020-02-23              3
#7 K07      B Clinic_1 2020-02-25 2020-02-29              4 ## treats this as single, should be 3
#8 K08      B Clinic_1 2020-03-06 2020-03-21              5 ## should be 4
#9 K09      B Clinic_1 2020-03-18 2020-03-22              5 ## should be 4

Answer 1

使用data.table的选项：

setkey(setDT(have), clinic, strain, datein, dateout)
have[, g := cumsum(c(0L, (shift(datein, -1L) > cummax(as.integer(dateout)))[-.N])), 
  .(clinic, strain)][, 
    g := rleid(clinic, strain, g)]

还有：

have[, g02 := cumsum(datein > shift(cummax(as.integer(dateout)), fill=dateout[1L])),
    .(clinic, strain)][,
        g2 := rleid(clinic, strain, g02)]

输出：

    id strain   clinic     datein    dateout g g2
1: K01      A Clinic_1 2020-01-01 2020-01-05 1  1
2: K02      A Clinic_1 2020-01-03 2020-01-16 1  1
3: K05      B Clinic_1 2020-02-18 2020-02-27 2  2
4: K06      B Clinic_1 2020-02-20 2020-02-23 2  2
5: K07      B Clinic_1 2020-02-21 2020-02-22 2  2
6: K08      B Clinic_1 2020-03-06 2020-03-21 3  3
7: K09      B Clinic_1 2020-03-18 2020-03-22 3  3
8: K03      A Clinic_2 2020-02-03 2020-02-09 4  4
9: K04      A Clinic_2 2020-02-09 2020-02-19 4  4

cummax 的想法来自 David Aurenburg 帖子：How to flatten / merge overlapping time periods

Answer 2

我们可以使用rleid

library(dplyr)
library(data.table)
have %>% 
  mutate(overlap_number = rleid(strain, clinic, 
       cumsum(datein > lag(dateout, default = first(dateout)))))
#    id strain   clinic     datein    dateout overlap_number
#1 K01      A Clinic_1 2020-01-01 2020-01-05              1
#2 K02      A Clinic_1 2020-01-03 2020-01-16              1
#3 K03      A Clinic_2 2020-02-03 2020-02-09              2
#4 K04      A Clinic_2 2020-02-09 2020-02-19              2
#5 K05      B Clinic_1 2020-02-18 2020-02-27              3
#6 K06      B Clinic_1 2020-02-20 2020-02-23              3
#7 K07      B Clinic_1 2020-02-21 2020-02-22              3
#8 K08      B Clinic_1 2020-03-06 2020-03-21              4
#9 K09      B Clinic_1 2020-03-18 2020-03-22              4