从特定列中选择值并跳过 R 中的 NA 值

Question

我正在处理癌症登记数据。在以下数据示例 (ex_data) 中，变量 id 和 diagnosis_yr 分别代表癌症诊断时的 ID 和年份。 x_2005 到 x_2010 列和 y_2005 到 y_2010 列分别代表每年（2005 年到 2010 年）的 x 和 y 状态）。在我的实际工作数据中，我有很多年（2005-2020）的专栏。我想通过排除 NA 从最早可用年份、最新可用年份和诊断年份（即“wanted”中的 x_earliest, y_latest,x_at_diagnosis,y_at_diagnosis 变量）中提取 x 和 y 值.例如，对于 id 1，我想通过跳过 NA 从最早的一年中提取 x 值，从最近一年中提取 y 值。对于诊断年份的 x 和 y 值，如果诊断年份有 NA，我想跳过 NA 并提取前一年的可用数据。如何实现在 R 中获取想要的变量？

library(tidyverse)

#example data
ex_data <- tribble(
~id,~diagnosis_yr,~x_2005,~x_2006,~x_2007,~x_2008,~x_2009,~x_2010,~y_2005,~y_2006,~y_2007,~y_2008,~y_2009,~y_2010,
1,  2007,   NA, NA, 1,  2,  2,  3,  "a",    "b",    "c",    "d",    "e",    NA, 
2,  2008,   1,  3,  1,  NA, 1,  2,   NA,    "b",    "b",    "e",    "d", "d",
3,  2010,   NA, 2,  2,  2,  3,  NA, "a",    "b",    "c",     NA,     NA,    NA,
4,  2009, 1,    3,  1,  NA, 1,  2,   NA,     NA,     NA,     NA,     NA,    NA,
5,  2005, NA,   1,  1,  2,  2,  3,  "a",    "b",    "c",    "d",    "e",    "e"
)

#wanted variables
wanted <- tribble(
  ~id,~diagnosis_yr,~x_earliest,~y_latest,~x_at_diagnosis,~y_at_diagnosis,
  1,    2007,   1,  "e",    1,  "c",
  2,    2008,   1,  "d",    1,  "e",
  3,    2010,   2,  "c",    3,  "c",
  4,  2009, 1,   NA,  1,  NA,
  5,  2005, 1,  "e", NA,  "a"
)

Answer 1

我不完全确定，如果这是正确的：

library(dplyr)
library(tidyr)

ex_data %>% 
  pivot_longer(-c(id, diagnosis_yr), 
               names_to = c(".value", "year"),
               names_pattern = "(.*)_(\\d+)") %>% 
  group_by(id) %>% 
  mutate(x_earliest     = first(na.omit(x)),
         x_at_diagnosis = last(na.omit(x[diagnosis_yr >= year])),
         y_latest       = last(na.omit(y)),
         y_at_diagnosis = last(na.omit(y[diagnosis_yr >= year]))) %>% 
  select(id, diagnosis_yr, x_earliest, y_latest, x_at_diagnosis, y_at_diagnosis) %>% 
  distinct() %>% 
  ungroup()

返回

# A tibble: 3 x 6
     id diagnosis_yr x_earliest y_latest x_at_diagnosis y_at_diagnosis
  <dbl>        <dbl>      <dbl> <chr>             <dbl> <chr>         
1     1         2007          1 e                     1 c             
2     2         2008          1 d                     1 e             
3     3         2010          2 c                     3 c    

Answer 2

策略：

1. 拆分x和y数据帧，最后加入，x和y的逻辑是一样的：

2. 使用coalesce 以正确的顺序您可以获得x_earliest 和y_latest。

3. 用之前的值填充 NA：这是 Anoushiravan Fill missing values with previous values by row using dplyr 的代码这是使用 library(zoo)

4. mutate 一个新列，其值取决于年份（来自 akrun 的代码，感谢大师）。

library(zoo)
library(tidyverse)

x <- ex_data %>% 
    select(id, diagnosis_yr, starts_with("x_")) %>% 
    mutate(x_earliest= coalesce(x_2005, x_2006, x_2007, x_2008, x_2009, x_2010 )) %>% 
    mutate(pmap_df(., ~ na.locf(c(...)[-1]))) %>% # fill NA with value before
    rowwise() %>% 
    mutate(x_at_diagnosis = get(str_c('x_', diagnosis_yr)))


y <- ex_data %>% 
    select(id, diagnosis_yr, starts_with("y_")) %>% 
    mutate(y_latest = coalesce(y_2010, y_2009, y_2008, y_2007, y_2006, y_2005)) %>% 
    mutate(pmap_df(., ~ na.locf(c(...)[-1]))) %>% 
    rowwise() %>% 
    mutate(y_at_diagnosis = get(str_c('y_', diagnosis_yr))) %>% 
    type.convert(as.is=TRUE)

left_join(x, y, by=c("id", "diagnosis_yr")) %>% 
    select(id, diagnosis_yr, x_earliest, y_latest, x_at_diagnosis, y_at_diagnosis)

输出：

 id diagnosis_yr x_earliest y_latest x_at_diagnosis y_at_diagnosis
  <dbl>        <dbl>      <dbl> <chr>             <dbl> <chr>         
1     1         2007          1 e                     1 c             
2     2         2008          1 d                     1 e             
3     3         2010          2 c                     3 c        

Answer 3

在@Martin 和@TarJae 的建议代码和策略的帮助下，我想分享以下代码（Martin 和 TarJae 建议代码的组合）来解决我的问题（编辑版）。

library (zoo)
library(dplyr)
library(tidyverse) 

ex_data %>% 
  pivot_longer(-c(id, diagnosis_yr), 
               names_to = c(".value", "year"),
               names_pattern = "(.*)_(\\d+)") %>% 
  group_by(id) %>% 
  mutate(x_earliest     = first(na.locf(x,fromLast=T,na.rm = F)),
         x_at_diagnosis = last(na.locf(x[diagnosis_yr >= year],na.rm = F)), #na.rm=F is to keep as it is if there is no replacement 
         y_latest       = last(na.locf(y,fromLast=F, na.rm =F)), 
         y_at_diagnosis = last(na.locf(y[diagnosis_yr >= year],na.rm=F))) %>% 
  dplyr::select(id, diagnosis_yr, x_earliest, y_latest, x_at_diagnosis, y_at_diagnosis) %>% 
  distinct() %>% 
  ungroup()

输出

id       diagnosis_yr x_earliest y_latest  x_at_diagnosis  y_at_diagnosis
  <dbl>        <dbl>      <dbl>  <chr>             <dbl> <chr>         
     1         2007          1    e                    1   c             
     2         2008          1    d                    1   e             
     3         2010          2    c                    3   c             
     4         2009          1    NA                   1   NA            
     5         2005          1    e                   NA   a