R-对一组进行动态迭代，以便差异大于26.5并且时间戳小于<48

如何解决R-对一组进行动态迭代，以便差异大于26.5并且时间戳小于<48

我正在尝试在R Studio中编写脚本，目的是在患者发展为急性肾损伤时进行捕获，如果是这样，将在第一时间发生。急性肾脏损伤的定义如下：

在48小时内肌酐的增加（实验室测试的名称）> = 26.5 umol / L（单位）

请参见下面的示例表，其数据结构如下：

PatientKey             Creatinine            Timestamp                 Mycommentsforthisquestion
1                      70                    2020-04-03 14:10:10
1                      90                    2020-04-03 17:11:10
1                      98                    2020-04-03 19:10:10       FirsT TIME OF AKI,i.e. > 26.5 
                                                                       increase and less than 48 hours
1                      100                   2020-04-03 22:10:10       NOT RELEVANT,AKI ALREADY

2                      140                   2019-08-01 00:00:00       ONLY ONE VALUE,IGnorE SINCE NO 
                                                                       DIFFERENCE CAN BE CALculaTED

3                      120                   2017-01-06 00:00:05       CAME TO HOSPITAL WITH HIGH VALUE
3                      70                    2017-01-06 10:00:05       DECREASES MORE THAN 26.5 DONT 
                                                                       COUNT --> NO AKI                 
3                      80                    2017-01-08 10:00:05

4                      70                    2020-01-08 22:00:05
4                      60                    2020-01-09 22:00:05       NOTE IT IS NOT ALWAYS THE FirsT 
                                                                       TEST THAT SERVES AS THE COMPARISON 
                                                                       VALUE
4                      90                    2020-01-10 02:00:05       90 - 60 > 26.5 --> AKI
4                      110                   2020-01-10 06:00:05       NOT RELEVANT,AKI ALREADY 

5                      50                    2020-01-12 06:00:05
5                      70                    2020-01-13 08:00:05 
5                      80                    2020-01-14 22:00:05       NO AKI,DIFFERENCE > 26.5 BUT MORE 
                                                                       THAN 48 HOURS BETWEEN TESTS

我想要的输出如下：

PatientKey    CreatinineLOW    CreatinineHIGH   TimestampLOW          TimestampHIGH 
1             70               98               2020-04-03 14:10:10   2020-04-03 19:10:10
4             60               90               2020-01-09 22:00:05   2020-01-10 02:00:05

请注意，只有患者1出现了AKI，因此输出中仅应包括该患者的数据。

在R中完成这项工作是否可行？我试图使用dplyr / tidyverse包并做类似的事情（MYDATA是数据帧的名称）：

datalist = list()

for (m in MYDATA$PatientKey %>% unique()) {
  x = filter(MYDATA,PatientKey == m) %>% pull(PatientKey)
  table <- MYDATA %>% filter(PatientKey == m) %>% arrange(Timestamp)
  for (i in 1:length(x)) {
     table %>% 
     mutate(indx_creat = Creatinine[1],new_creatinine = Creatinine - indx_creat,indx_time = Timestamp[i],new_time = as.numeric(difftime(Timestamp,indx_time,units = "hours"))) %>%
  filter(new_creatinine >= 26.5 & new_time <= -48) -> r
 if (nrow(r) == 0) {
   table <- table[-1,]
 } else if (nrow(r) > 0 {
   datalist[[i]] <- r
   break
  }
 }
}

summary.table = do.call(rbind,datalist)
summary.table <- summary_table %>% group_by(PatientKey) %>% slice(1)

但是，这没有用！任何人都对如何完成工作有任何想法？对于临床社区来说，使用程序轻松检测急性肾脏损伤将非常有用！

解决方法

让我们分两个步骤进行操作，使其更通用。

第1步：每次观察发现任何关键增长
首先，我们将创建一列嵌套的数据框，每个数据框包含 any 行，这些行显示了Creatinin的严重增加：

library(dplyr)
library(purrr)
library(tidyr)
library(lubridate)

# loading OPs data
df <- structure(list(PatientKey = c(1L,1L,2L,3L,4L,5L,5L),Creatinine = c(70L,90L,98L,100L,140L,120L,70L,80L,60L,110L,50L,80L),Timestamp = c("2020-04-03 14:10:10","2020-04-03 17:11:10","2020-04-03 19:10:10","2020-04-03 22:10:10","2019-08-01 00:00:00","2017-01-06 00:00:05","2017-01-06 10:00:05","2017-01-08 10:00:05","2020-01-08 22:00:05","2020-01-09 22:00:05","2020-01-10 02:00:05","2020-01-10 06:00:05","2020-01-12 06:00:05","2020-01-13 08:00:05","2020-01-14 22:00:05")),row.names = c(NA,-15L),class = "data.frame")

# find any critical increase,per observation row
df2 <- df %>%
  mutate(
    Timestamp = as_datetime(Timestamp)
  ) %>%
  group_by(PatientKey) %>%
  do(
    mutate(.,critical_increase = map2(
        Creatinine,Timestamp,.f = function(c,t,data) {
          data %>%
            filter(
              Creatinine - c >= 26.5,Timestamp > t,as.numeric(Timestamp - t,units = "hours") <= 48
            ) %>%
            select(Timestamp,Creatinine)
        },data = .
      )
    )
  ) %>%
  ungroup()

上面的代码映射到每行的“肌酸酐”和“时间戳”，并抛出相应患者数据的分组子集，并对其进行过滤以进行关键性增加。过滤后的df存储在“ critical_increase”列中。

步骤2：将数据转换为预期格式
现在，要获得您描述的格式，我们将取消嵌套先前计算过的数据框，将列重命名为您想要的数据列，并通过“ TimestampHIGH”选择每个患者的第一行：

# unnest,rename & select first critical increase
df_final <- df2 %>%
  unnest(critical_increase,names_sep = ".") %>%
  transmute(
    PatientKey = PatientKey,CreatinineLOW = Creatinine,CreatinineHIGH = critical_increase.Creatinine,TimestampLOW = Timestamp,TimestampHIGH = critical_increase.Timestamp
  ) %>%
  group_by(PatientKey) %>%
  arrange(TimestampHIGH,desc(TimestampLOW)) %>%
  filter(row_number() == 1) %>%
  ungroup()

您可以轻松进行变化，例如arrange(desc(CreatinineHIGH - CreatinineLOW))，以发现48小时内观察到的最大增加量。

结果
上面给出了预期的结果：

> df_final
# A tibble: 2 x 5
  PatientKey CreatinineLOW CreatinineHIGH TimestampLOW        TimestampHIGH      
       <int>         <int>          <int> <dttm>              <dttm>             
1          4            60             90 2020-01-09 22:00:05 2020-01-10 02:00:05
2          1            70             98 2020-04-03 14:10:10 2020-04-03 19:10:10