[R] Extracting sentences with combinations of target words/terms from cancer patient text medical records

Robert McGehee rmcgehee at walleyetrading.net
Thu Jul 13 22:23:36 CEST 2017 

Hi Paul,
No need to collapse the information into a single text string, gregexpr() can take a vector of strings (sentences in your case). You can split your sentences up, number them how you want, then search for your pattern either via regex or via these extra packages you use which probably use the PCRE regex library anyway. However, as this is basically what you did, I'm not sure why you're not happy with your existing approach.



-----Original Message-----
From: Paul Miller [mailto:pjmiller_57 at yahoo.com] 
Sent: Thursday, July 13, 2017 3:01 PM
To: Robert McGehee <rmcgehee at walleyetrading.net>
Cc: r-help at r-project.org
Subject: Re: [R] Extracting sentences with combinations of target words/terms from cancer patient text medical records

Hi Robert,

Thank you for your reply. An attempt to solve this via a regular expression query is particularly helpful. Unfortunately, I don't have much time to play around with this just now. Ultimately though, I think I would like to implement a solution something along the lines of what you have done. I have a book on regular expressions that I am now starting to read. In the meantime, the code I'm using is a good way to assess the feasibility of some ideas I'd like to implement.

The advantage of your approach I think is that it makes fewer passes through the data. That should make it a lot faster and more efficient than what I've done. I'm currently working with a little more than 2.5 million text records and I think that number will only rise. So efficiency really should matter. 

I've pasted the latest version of my sample code below. This shows how I'd like to add the result of the text search as a column in a data frame. It also shows how I'd like to append the sentence number to each identified sentence. The single colon that appears where there is no match is not by design. It's something that I need to tidy.

My sense is that if I used your regular expression as written, I'd lose the information about the sentence number when I added the result as a column in my data frame. Presumably, I'd need to collapse the information into a single text string, and then the numbering would be lost. If you were going to get the sentence numbers as well, without making several passes through the data like my code does, how would you go about it?

Thanks,

Paul


library(tidyverse)
library(stringr)
library(lubridate)
 
sentence_match <- function(x){
  sentence_extract <- str_extract_all(x, boundary("sentence"), simplify = TRUE)
  sentence_number <- intersect(str_which(sentence_extract, "breast"), str_which(sentence_extract, "metastatic|stage IV"))
  sentence_match <- str_c(sentence_number, ": ", sentence_extract[sentence_number], collapse = "")
  sentence_match
}
 
sampletxt <-
  structure(
    list(
      PTNO = c(1, 2, 2, 2),
      DATE = structure(c(16436, 16436, 16832, 16845), class = "Date"),
      TYPE = c("Progress note", "CAT scan", "Progress note", "Progress note"),
      TVAR = c(
        "This sentence contains the word metastatic. This sentence contains the term stage IV.",
        "This sentence contains no target words. This sentence also contains no target words.",
        "This sentence contains the word metastatic and the word breast. This sentence contains no target words.",
        "This sentence contains the words breast and the term metastatic. This 
sentence contains the word breast and the term stage IV."
      )
    ),
    .Names = c("PTNO", "DATE", "TYPE", "TVAR"),
    class = c("tbl_df",
              "tbl", "data.frame"),
    row.names = c(NA,-4L)
  )
 
sampletxt$EXTRACTED <- sapply(sampletxt$TVAR, sentence_match)
sampletxt$EXTRACTED
 
> sampletxt$EXTRACTED
[1] ": 
"                                                                                                                                  
[2] ": 
"                                                                                                                                  
[3] "1: This sentence 
contains the word metastatic and the word breast. 
"                                                                 
[4] "1: This sentence contains the words breast and the term metastatic. 2: This sentence contains the word breast and the term stage IV."

________________________________
From: Robert McGehee <rmcgehee at walleyetrading.net>
To: Paul Miller <pjmiller_57 at yahoo.com>; Bert Gunter <bgunter.4567 at gmail.com> 
Cc: "r-help at r-project.org" <r-help at r-project.org>
Sent: Wednesday, July 12, 2017 12:47 PM
Subject: RE: [R] Extracting sentences with combinations of target words/terms from cancer patient text medical records



Hi Paul,
Sounds like you have your answer, but for fun I thought I'd try solving your problem using only a regular expression query and base R. I believe this works:

> txt <- "Patient had stage IV breast cancer. Nothing matches this sentence. Metastatic and breast match this sentence. French bike champion takes stage IV victory in Tour de France."

> pattern <- "([^.?!]*(?=[^.?!]*\\bbreast\\b)(?=[^.?!]*\\b(metastatic|stage IV)\\b)(?=[\\s.?!])[^.?!]*[.?!])"

> regmatches(txt, gregexpr(pattern, txt, perl=TRUE, ignore.case=TRUE))[[1]]
[1] "Patient had stage IV breast cancer."        
[2] " Metastatic and breast match this sentence."

Cheers, Robert

-----Original Message-----
From: R-help [mailto:r-help-bounces at r-project.org] On Behalf Of Paul Miller via R-help
Sent: Wednesday, July 12, 2017 8:49 AM
To: Bert Gunter <bgunter.4567 at gmail.com>
Cc: R-help <r-help at r-project.org>
Subject: Re: [R] Extracting sentences with combinations of target words/terms from cancer patient text medical records

Hi Bert,

Thanks for your reply. It appears that I didn't replace the variable name "sampletxt" with the argument "x" in my function. I've corrected that and now my code seems to be working fine.

Paul

________________________________
From: Bert Gunter <bgunter.4567 at gmail.com>

Cc: R-help <r-help at r-project.org>
Sent: Tuesday, July 11, 2017 2:00 PM
Subject: Re: [R] Extracting sentences with combinations of target words/terms from cancer patient text medical records



Have you looked at the CRAN Natural Language Processing Task View? If not, why not? If so, why were the resources described there inadequate?

Bert


On Jul 11, 2017 10:49 AM, "Paul Miller via R-help" <r-help at r-project.org> wrote:

Hello All,
>
>I need some help figuring out how to extract combinations of target words/terms from cancer patient text medical records. I've provided some sample data and code below to illustrate what I'm trying to do. At the moment, I'm trying to extract sentences that contain the word "breast" plus either "metastatic" or "stage IV".
>
>It's been some time since I used R and I feel a bit rusty. I wrote a function called "sentence_match" that seemed to work well when applied to a single piece of text. You can see that by running the section titled
>
>"Working code". I thought that it might be possible easily to apply my function to a data set (tibble or df) but that doesn't seem to be the case. My unsuccessful attempt to do this appears in the section titled "Non-working code".
>
>If someone could help me get my code up and running, that would be greatly appreciated. I'm using a lot of functions from Hadley Wickham's packages, but that's not particularly necessary. Although I have only a few entries in my sample data, my actual data are pretty large. Currently, I'm working with over a million records. Some records contain only a single sentence, but many have several paragraphs. One concern I had was that, even if I could get my code working, it would be too inefficient to handle that volume of data.
>
>Thanks,
>
>Paul
>
>
>library(tidyverse)
>library(stringr)
>library(lubridate)
>
>sentence_match <- function(x){
>  sentence_extract <- str_extract_all(sampletxt, boundary("sentence"), simplify = TRUE)
>  sentence_number <- intersect(str_which(sentence_ extract, "breast"), str_which(sentence_extract, "metastatic|stage IV"))
>  sentence_match <- str_c(sentence_number, ": ", sentence_extract[sentence_ number], collapse = "")
>  sentence_match
>}
>
>#### Working code ####
>
>sampletxt <- "This sentence contains the word metastatic and the word breast. This sentence contains no target words."
>
>sentence_match(sampletxt)
>
>#### Non-working code ####
>
>sampletxt <-
>  structure(
>    list(
>      PTNO = c(1, 2, 2, 2),
>      DATE = structure(c(16436, 16436, 16832, 16845), class = "Date"),
>      TYPE = c("Progress note", "CAT scan", "Progress note", "Progress note"),
>      TVAR = c(
>        "This sentence contains the word metastatic. This sentence contains the term stage IV.",
>        "This sentence contains no target words. This sentence also contains no target words.",
>        "This sentence contains the word metastatic and the word breast. This sentence contains no target words.",
>        "This sentence contains the words breast and the term metastatic. This
>sentence contains the word breast and the term stage IV."
>      )
>    ),
>    .Names = c("PTNO", "DATE", "TYPE", "TVAR"),
>    class = c("tbl_df",
>              "tbl", "data.frame"),
>    row.names = c(NA,-4L)
>  )
>
>sampletxt2 <- group_by_at(sampletxt, vars(PTNO, DATE, TYPE))
>sampletxt2 <-
>  sampletxt2 %>%
>  mutate(
>    EXTRACTED = sentence_match(TVAR)
>  )
>
>______________________________ ________________
>R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
>https://stat.ethz.ch/mailman/ listinfo/r-help
>PLEASE do read the posting guide http://www.R-project.org/ posting-guide.html
>and provide commented, minimal, self-contained, reproducible code.

>

______________________________________________
R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

More information about the R-help mailing list