根据 R 中的其他行选择满足条件的行

Question

I am working in R to identify incident cases of a disease. Each patient has multiple visits over the years (each row of the dataframe is one visit), and to be labeled "incident", a visit has to meet the following criteria:

• The infection test must be positive (Infection == "yes")
• That patient has not already been "positive" for an infection in the last two years

My data looks like this:

I want to create a new variable indicating whether each visit is an incident infection case or not. For example, the output should look like this:

As seen, a patient can be incident more than once. Any time they have a positive infection test and also haven't had another positive infection test in the past two years, they are considered incident.

I can't find an efficient way get this output in R. Can it be done using dplyr? Would appreciate any help on this.

Answer 1

One method is to compute the difference in time between infection events ( event_diff ). Then, incident would be when this difference is greater than 2 years, or difference of 0 (assuming multiple tests are not done on same date). Looking at this now, I suspect there are better alternative solutions to this.

df <- data.frame(
  patient_id = c(1,1,1,1,1,1,2,2,2,2),
  infection = c("no", "yes", "yes", "no", "yes", "yes", "yes", "no", "no", "yes"),
  date = c("2005-02-22", "2005-04-26", "2005-05-06", "2006-05-22", "2007-08-19", "2007-12-15", "2005-10-24", "2005-11-11", "2006-07-12", "2007-12-01")
)

df$date <- as.Date(df$date, "%Y-%m-%d")

library(dplyr)

df %>%
  group_by(patient_id, infection) %>%
  mutate(event_diff = coalesce(date - lag(date), 0)) %>%
  mutate(incident = ifelse(infection == "yes" & (event_diff == 0 | event_diff > (365*2)), "yes", "no"))

   patient_id infection date       event_diff incident
        <dbl> <fct>     <date>     <drtn>     <chr>   
 1          1 no        2005-02-22   0 days   no      
 2          1 yes       2005-04-26   0 days   yes     
 3          1 yes       2005-05-06  10 days   no      
 4          1 no        2006-05-22 454 days   no      
 5          1 yes       2007-08-19 835 days   yes     
 6          1 yes       2007-12-15 118 days   no      
 7          2 yes       2005-10-24   0 days   yes     
 8          2 no        2005-11-11   0 days   no      
 9          2 no        2006-07-12 243 days   no      
10          2 yes       2007-12-01 768 days   yes