Search code examples
rweb-servicesasp.net-web-apircurlhttr

Using R to call a Web Service: Send data and get the result table back in R

http://snomedct.t3as.org/ This is a web service that will analyse English clinical text, and report any concepts that can be detected.
For e.g.- I have headache. It will identify headache as a Symptom.

Now what I would like to do is send the sentence to the web service through R, and get the table back from the web page to R for further analysis purpose.

Solution

  • If we take their example curl command-line:

    curl -s --request POST \
        -H "Content-Type: application/x-www-form-urlencoded" \
        --data-urlencode "The patient had a stroke." \
        http://snomedct.t3as.org/snomed-coder-web/rest/v1.0/snomedctCodes

    that can be translated to httr pretty easily.

    The -s means "silent" (no progress meter or error messages) so we don't really have to translate that.

    Any -H means to add a header to the request. This particular Content-Type header can be handled better with the encode parameter to httr::POST.

    The --data-urlencode parameter says to URL encode that string and put it in the body of the request.

    Finally, the URL is the resource to call.

    library(httr)

result <- POST("http://snomedct.t3as.org/snomed-coder-web/rest/v1.0/snomedctCodes", 
               body="The patient had a stroke.",
               encode="form")

    Since you don't do this regularly, you can wrap the POST call with with_verbose() to see what's going on (look that up in the httr docs).

    There are a ton of nuances that one should technically do after this (like check the HTTP status code with stop_for_status(), warn_for_status() or even just status_code(), but for simplicity let's assume the call works (this one is their example so it does work and returns a 200 HTTP status code which is A Good Thing).

    By default, that web service is returning JSON, so we need to convert it to an R object. While httr does built-in parsing, I like to use the jsonlite package to process the result:

    dat <- jsonlite::fromJSON(content(result, as="text"),  flatten=TRUE)

    The fromJSON function takes a few parameters that are intended to help shape JSON into a reasonable R data structure (many APIs return horrible JSON and/or XML). This API would fit into the "horrible" category. The data in dat is pretty gnarly and further decoding of it would be a separate SO question.