Emergency Department Data Set – Improving Data Quality

Improving EDDS data quality is a key piece of work which contributes towards all of AWISS objectives, but specifically fulfils AWISS’s first long term objective, ‘to continuously strive to improve the quality and completeness of injury surveillance data’.

AWISS are involved in several sub-projects which are working towards this goal including:

  1. Cross-mapping datasets in SAIL (e.g. linking EDDS to PEDW) to identify the proportion of injuries potentially missing from EDDS
  2. Improving the quality of historic EDDS data by importing ED data directly from local ED systems into SAIL (e.g. skipping the conversion to EDDS format)
  3. Developing simpler and more concise data entry systems in ED departments, to improve the quality and completeness of data collected by ED receptionists.
  4. Supporting the development of a standardised dataset for use in hospitals across Wales
  5. Enhanced analysis of free text fields to populate missing fields and validate current records.

In relation to the 1st point, AWISS staff are currently exploring the potential to link EDDS data to inpatient and outpatient data, to improve injury incidence estimates. At present, inpatient and outpatient data are of much higher quality than the data in EDDS.  Thus, we have explored the potential to use inpatient and outpatient data to estimate the proportion of injuries in ED (e.g. where data is missing / unspecified in EDDS). For example, recent analyses demonstrated that whilst 334,007 ED cases (Welsh residents) were recorded as having an injury diagnosis in 2013, a further 171,984 cases had a code suggestive of an injury (e.g. attendance recorded as an accident, assault, self-harm, undetermined, or withheld intent), but with missing or null diagnostic codes, indicating poor data collection at source or poor mapping of local to national codes. The individual linkage system enabled ED attendances to be mapped to hospital inpatient data, identifying 32,324 emergency admissions on the same or next day, of which 8,509 had a primary injury diagnosis on discharge coding (26.3%).  Thus this suggests that in 2013, the ED system may currently be underestimating injuries by an estimated 45,273 cases (13.6% of the total).

Regarding  point 2, AWISS staff are currently in discussions with Morriston hospital, to see whether it would be possible to import historic ED records directly into SAIL via SAIL’s split file anonymisation approach (rather than via hospital IT departments and the EDDS).  Currently AWISS rely on EDDS data which has been converted from local hospital systems.  By bringing ED records directly into SAIL, AWISS analysts will be able to investigate whether cross-mapping issues may have occurred in the past.

In relation to point 3 above, AWISS staff devised and developed a new injury surveillance dataset as part of the EU funded JAMIE project (now funded by BRIDGE-Health – discussed in another section). The simplified JAMIE minimum level dataset (MDS) was designed to make the data collection process within EDs more simplistic and less time consuming, thus easing the data entry burden on receptionists and clinicians, and improving the completeness of the data records collected.  MDS contains only the most useful codes for the most important injury prevention variables. Whilst this means the dataset will not meet all the needs for detailed information on all permutations of intent/activity/mechanism and location, it will provide high level data to allow enumeration of injuries in the home, home and leisure (combined), during work, and due to road traffic, falls, sports or burns/scalds, and resulting from unintentional injuries, self-harm or assaults (reflecting the main focus of prevention strategies across the world).

In relation to point 4, despite lengthy discussions with ED information managers across Wales, it was not possible to insert the exact original MDS specifications into the new all-Wales ED system.  However, the new ED dataset will still allow the JAMIE MDS fields to be populated for Wales, and we are working closely alongside the project leads who are keen to ensure the new ED system meets the requirements of the MDS.

Finally, with regards to point 5, AWISS staff are working on an enhanced injury surveillance project in collaboration with Morriston Hospital and Clinithink (partially funded by the JAMIE/BRIDGE-Health project).  We are attempting to utilise Clinithink’s algorithms (which are designed to convert unstructured clinical narrative into rich structured data) to improve the data quality/completeness of EDDS, by extracting information on the how, where and why an injury occurred from the narrative ‘presenting complaint’ field in EDDS. If it is possible to extract accurate and valid information from the presenting complaint field, it will then be possible to populate missing values in categorical fields in EDDS, as well as validate completed categorical fields.  Current analyses includes exploration of free text fields in the Morriston ED dataset (2014) to identify falls and injuries in the home which have been recorded incorrectly in the derived coded fields.  Initial results have shown 13819 fall related injuries reported in the presenting complaint free text field compared to 7327 fall injuries coded in the ‘mechanism of injury’ field.  Thus it appears approximately 50% of fall related ED attendances are coded incorrectly in the Morriston data set.