Guide 10: outpatients

Script needed for this guide

Introduction

So far, from HES, we have exclusively used the admitted patient care data. The use of these data has a long pedigree and they contain rich diagnostic and procedure information, recorded on a mandatory basis. Outpatient data have been relatively underexplored. Even though the outpatient dataset has columns that could contain diagnoses, these data are missing in almost all instances (in fact, code R69 – unknown and unspecified causes of morbidity is recorded). Procedure information is slightly more complete, but possibly of limited use (e.g., the most common procedure is OPCS-4 code X62.1 – assessment by uniprofessional team). As a result, the outpatient data contain limited clinical or diagnostic information that is useful for phenotyping.

There are, however, situations where outpatient data may be useful. One is where the interest is in use of hospitals generally, such as in the study by Mc Grath-Lone et al on hospital use before and during the COVID-19 pandemic. Additionally, the outpatient dataset does contain information on the treatment specialty. It is therefore possible to identify patients using a particular service. If the referral acceptance criteria of such services are sensitive and specific enough, you may be able to use outpatient data for the purpose of identifying patients with such conditions. In this example, we therefore extract outpatient data in order to identify patients in our cohorts who attended specialist chronic pain clinics.

Data extraction

In 11_outpatient_data.r we (yet again) adopt the same approach as in previous guides to data extraction: identify relevant tables and loop a query through them. In fact, we let the SQL query do all the heavy lifting (Code Snippet 1). You will see that we extract the appointment date, whether it was a first attendance, whether it was attended and the treatment specialty (TRETPSEF). We also only select appointments with services with treatment specialty 191 or 241, both of which, as you will see in the HES data dictionary, represent pain services.

Code Snippet 1 (R - 11_outpatient_data.r)

temp <- data.table(
  sqlQuery(
    dbhandle, paste0(
      "SELECT TOKEN_PERSON_ID, APPTDATE, FIRSTATT, ATTENDED, TRETSPEF ",
      "FROM ", table_name, " ",
      "WHERE TRETSPEF IN ('191', '241')")
  )
)
---

Recall again that HES are organised in financial years. We therefore extract more years than we need and then subset. There is nothing much else to this script. We end up with a dataset that contains appointments with pain services.

Some outpatient quirks

Some things to bear in mind about outpatient data:

  • Each row represents a planned appointment. Appointments can be attended (on-time or late), cancelled or not attended (such appointments are often referred to as “Did Not Attend” [DNA] or, in paediatrics, more appropriately, “Was Not Brought”). You can discern all of these in the data.
  • An appointment can be a first or a follow-up. This is coded explicitly, though beware of possible coding error and left-censoring (i.e., a patient’s first appointment with a service may have been before the data began).
  • A patient could in theory have a “first” appointment with the same service more than once as they can be discharged and then re-referred.
  • There are two waiting time variables. WAITING relates to actual waiting days from the date the referral was received to the first planned appointment (whether attended or not). The other, WAITDAYS, refers to the waiting time according to the referral-to-treatment pathway, which includes clock stops. You may wish to calculate your own waiting time, for example using the referral received date (REQDATE) and the date of the first attended appointment.

Coming up…

In Guide 11, we finally return to our spine and combine all the data so far extracted into it. With this, we shall have an analysis-ready dataset.