嗨堆栈溢出社区，
我是 R 的新手(9 个月)，这是我第一个关于 reprex 的堆栈溢出问题，非常感谢任何帮助。尽管我对基础 R 解决方案持开放态度，但我主要使用 tidyverse。
问题:
我有大约 21,000 行症状数据，每天有 >10 个变量。我希望能够通过使用规则来定义发作的开始和结束来对疾病的“恶化”(在这种情况下为肺部疾病的胸部感染)进行分类，以便我以后可以计算发作的持续时间，发作的类型(这取决于症状的组合)和接受的治疗。与任何涉及患者的数据集一样，存在缺失值。如果缺少不到 2 天的数据，我会从最近一天算起。
以下代码是一个简化的虚构示例，仅涉及一种症状。
恶化规则: 恶化开始 = 症状恶化 2 天 (>= 3) 缓解恶化 = 5 天正常呼吸 (<=2)
理想情况下，我希望能够确定恶化正在发生的所有日子。
数据如下:
#load packages
library(tidyverse)
#load data
id <- "A"
day <- c(1:50)
symptom <- c(2,2,2,2,2,2,2,2,2,2,2,3,2,2,2,2,NA,NA,NA,2,2,2,3,3,3,4,4,3,3,2,3,2,2,3,3,2,2,2,2,2,2,3,2,2,2,2,2,3,2,2)
df <- data.frame(id,day,symptom)
#Data Dictionary
#Symptom: 1 = Better than usual, 2 = Normal/usual, 3 = Worse than usual, 4 = Much worse than usual
我尝试过的:
我试图通过结合使用 lag() 和 lead() 以及条件语句 case_when() 和 ifelse() 来解决这个问题。
df %>%
mutate_at(vars("symptom"), #used for more variables within vars() argument
.funs = list(lead1 = ~ lead(., n = 1),
lead2 = ~ lead(., n = 2),
lead3 = ~ lead(., n = 3),
lead4 = ~ lead(., n = 4),
lead5 = ~ lead(., n = 5),
lag1 = ~ lag(., n = 1),
lag2 = ~ lag(., n = 2),
lag3 = ~ lag(., n = 3))) %>%
mutate(start = case_when(symptom <= 2 ~ 0,
symptom >= 3 ~
ifelse(symptom >= lag2 & symptom <= lag1,1,0)),
end = case_when(symptom >=3 ~
ifelse(lead1 <=2 &
lead2 <=2 &
lead3 <=2 &
lead4 <=2 &
lead5 <=2,1,0)))
我的主要问题是复杂性。随着我构建更多的症状和规则，我必须引用其中包含 ifelse()/case_when() 语句的不同变量。我确信我的问题有更优雅的解决方案。
另一个问题是，在“恶化”期间，exacerbation_start 变量应该只在开始时使用，而不是在发作期间使用。与 exacerbation_end 类似，它仅适用于已经发生恶化的情况。我曾尝试使用 ifelse() 语句来指代病情恶化的时间，但无法让它发挥作用并遵守我想要的规则。
我想要的输出是:
id day symptom start end exacerbation
1 A 1 2 0 0 0
2 A 2 2 0 0 0
3 A 3 2 0 0 0
4 A 4 2 0 0 0
5 A 5 2 0 0 0
6 A 6 2 0 0 0
7 A 7 2 0 0 0
8 A 8 2 0 0 0
9 A 9 2 0 0 0
10 A 10 2 0 0 0
11 A 11 2 0 0 0
12 A 12 3 0 0 0
13 A 13 2 0 0 0
14 A 14 2 0 0 0
15 A 15 2 0 0 0
16 A 16 2 0 0 0
17 A 17 NA 0 0 0
18 A 18 NA 0 0 0
19 A 19 NA 0 0 0
20 A 20 2 0 0 0
21 A 21 2 0 0 0
22 A 22 2 0 0 0
23 A 23 3 0 0 0
24 A 24 3 1 0 1
25 A 25 3 0 0 1
26 A 26 4 0 0 1
27 A 27 4 0 0 1
28 A 28 3 0 0 1
29 A 29 3 0 0 1
30 A 30 2 0 0 1
31 A 31 3 0 0 1
32 A 32 2 0 0 1
33 A 33 2 0 0 1
34 A 34 3 0 0 1
35 A 35 3 0 1 1
36 A 36 2 0 0 0
37 A 37 2 0 0 0
38 A 38 2 0 0 0
39 A 39 2 0 0 0
40 A 40 2 0 0 0
41 A 41 2 0 0 0
42 A 42 3 0 0 0
43 A 43 2 0 0 0
44 A 44 2 0 0 0
45 A 45 2 0 0 0
46 A 46 2 0 0 0
47 A 47 2 0 0 0
48 A 48 3 0 0 0
49 A 49 2 0 0 0
50 A 50 2 0 0 0
期待您的回复!
编辑
我又添加了 50 行数据来模拟多次恶化以及右删失和 NA 的问题。我还包括了第二个参与者“B”，看看这是否是问题的原因。
id <- c("A","A","A","A","A","A","A","A","A","A","A","A","A","A","A","A",
"A","A","A","A","A","A","A","A","A","A","A","A","A","A","A","A",
"A","A","A","A","A","A","A","A","A","A","A","A","A","A","A","A","A","A",
"B","B","B","B","B","B","B","B","B","B","B","B","B","B","B","B",
"B","B","B","B","B","B","B","B","B","B","B","B","B","B","B","B",
"B","B","B","B","B","B","B","B","B","B","B","B","B","B","B","B","B","B")
day <- c(1:50,1:50)
symptom <- c(2,3,3,3,3,2,2,2,2,2,2,3,2,2,2,2,NA,NA,NA,2,2,2,3,3,3,4,4,3,3,2,3,2,2,3,3,2,2,2,2,2,2,3,2,2,2,2,2,3,2,2, 2,2,2,2,2,2,3,2,3,3,2,3,2,3,2,2,2,2,2,2,3,3,3,3,NA,NA,NA,2,2,2,3,2,2,2,2,2,3,2,2,3,NA,NA,NA,3,3,3,3,3,3,2)
df <- data.frame(id,day,symptom)
id day symptom start end exacerbation censor
1 A 1 2 0 0 0 0
2 A 2 3 1 0 1 0
3 A 3 3 0 0 1 0
4 A 4 3 0 0 1 0
5 A 5 3 0 1 1 0
6 A 6 2 0 0 0 0
7 A 7 2 0 0 0 0
8 A 8 2 0 0 0 0
9 A 9 2 0 0 0 0
10 A 10 2 0 0 0 0
11 A 11 2 0 0 0 0
12 A 12 3 0 0 0 0
13 A 13 2 0 0 0 0
14 A 14 2 0 0 0 0
15 A 15 2 0 0 0 0
16 A 16 2 0 0 0 0
17 A 17 NA 0 0 0 0
18 A 18 NA 0 0 0 0
19 A 19 NA 0 0 0 0
20 A 20 2 0 0 0 0
21 A 21 2 0 0 0 0
22 A 22 2 0 0 0 0
23 A 23 3 1 0 1 0
24 A 24 3 0 0 1 0
25 A 25 3 0 0 1 0
26 A 26 4 0 0 1 0
27 A 27 4 0 0 1 0
28 A 28 3 0 0 1 0
29 A 29 3 0 0 1 0
30 A 30 2 0 0 1 0
31 A 31 3 0 0 1 0
32 A 32 2 0 0 1 0
33 A 33 2 0 0 1 0
34 A 34 3 0 0 1 0
35 A 35 3 0 0 1 0
36 A 36 2 0 0 1 0
37 A 37 2 0 0 1 0
38 A 38 2 0 0 1 0
39 A 39 2 0 0 1 0
40 A 40 2 0 0 1 0
41 A 41 2 0 1 1 0
42 A 42 3 0 0 0 0
43 A 43 2 0 0 0 0
44 A 44 2 0 0 0 0
45 A 45 2 0 0 0 0
46 A 46 2 0 0 0 0
47 A 47 2 0 0 0 0
48 A 48 3 0 0 0 0
49 A 49 2 0 0 0 0
50 A 50 2 0 0 0 0
51 B 1 2 0 0 0 0
52 B 2 2 0 0 0 0
53 B 3 2 0 0 0 0
54 B 4 2 0 0 0 0
55 B 5 2 0 0 0 0
56 B 6 2 0 0 0 0
57 B 7 3 0 0 0 0
58 B 8 2 0 0 0 0
59 B 9 3 0 0 0 0
60 B 10 3 1 0 1 0
61 B 11 2 0 0 1 0
62 B 12 3 0 0 1 0
63 B 13 2 0 0 1 0
64 B 14 3 0 0 1 0
65 B 15 2 0 0 1 0
66 B 16 2 0 0 1 0
67 B 17 2 0 0 1 0
68 B 18 2 0 0 1 0
69 B 19 2 0 1 1 0
70 B 20 2 0 0 0 0
71 B 21 3 1 0 1 0
72 B 22 3 0 0 1 0
73 B 23 3 0 0 1 0
74 B 24 3 0 0 1 0
75 B 25 NA 0 0 0 1
76 B 26 NA 0 0 0 1
77 B 27 NA 0 0 0 1
78 B 28 2 0 0 0 1
79 B 29 2 0 0 0 1
80 B 30 2 0 0 0 1
81 B 31 3 0 0 0 1
82 B 32 2 0 0 0 1
83 B 33 2 0 0 0 1
84 B 34 2 0 0 0 1
85 B 35 2 0 0 0 1
86 B 36 2 0 0 0 1
87 B 37 3 0 0 0 0
88 B 38 2 0 0 0 0
89 B 39 2 0 0 0 0
90 B 40 3 0 0 0 0
91 B 41 NA 0 0 0 0
92 B 42 NA 0 0 0 0
93 B 43 NA 0 0 0 0
94 B 44 3 1 0 1 0
95 B 45 3 0 0 1 0
96 B 46 3 0 0 1 0
97 B 47 3 0 0 1 0
98 B 48 3 0 0 1 0
99 B 49 3 0 0 1 0
100 B 50 2 0 0 1 0
>
最佳答案
这是一种尝试以更优雅和可扩展的方式编写您的算法:
首先，在使用 case_when 之前，您不必计算 lead 和 lag 调用。值得注意的是，我发现明确编写 case_when 的 TRUE 选项是一种很好的做法。这是一些代码。
df2=df %>%
mutate(
exacerbation_start = case_when(
is.na(symptom) ~ NA_real_,
symptom <= 2 ~ 0,
symptom >= 3 & symptom >= lag(symptom, n=2) & symptom <= lag(symptom, n=1) ~ 1,
TRUE ~ 0
),
exacerbation_end = case_when(
symptom >=3 ~ ifelse(lead(symptom, n=1) <=2 &
lead(symptom, n=2) <=2 & lead(symptom, n=3) <=2 &
lead(symptom, n=4) <=2 & lead(symptom, n=5) <=2,
1,0),
TRUE ~ NA_real_
)
)
all.equal(df1,df2) #TRUE
或者，如果您的算法对所有症状都相同，您可能需要使用自定义函数:
get_exacerbation_start = function(x){
case_when(
is.na(x) ~ NA_real_,
x <= 2 ~ 0,
x >= 3 & x >= lag(x, n=2) & x <= lag(x, n=1) ~ 1,
TRUE ~ 0
)
}
get_exacerbation_end = function(x){
case_when(
x >=3 ~ ifelse(x >=3 & lead(x, n=1) <=2 &
lead(x, n=2) <=2 & lead(x, n=3) <=2 &
lead(x, n=4) <=2 & lead(x, n=5) <=2,
1,0),
TRUE ~ NA_real_
)
}
df3=df %>%
mutate(
exacerbation_start = get_exacerbation_start(symptom),
exacerbation_end = get_exacerbation_end(symptom)
)
all.equal(df1,df3) #also TRUE
对于某些 mutate_at 调用，后一种方法可能会更加强大。
编辑:在看到您的编辑后，这里是获取恶化期的尝试。在我看来，代码非常丑陋，我不确定 row_number 是否应该以这种方式使用。
df_final=df %>%
transmute(
id,day,symptom,
start = get_exacerbation_start(symptom),
end = get_exacerbation_end(symptom),
exacerbation = row_number()>=which(start==1)[1] & row_number()<=which(end==1)[1]
)
关于r - 如何使用 lag/lead 和 ifelse/case_when(或其他解决方案)处理 R 中的纵向症状数据？，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/60168370/