Abstraction tool developed for our UCC trial

Abstraction is the tedious process of obtaining data from medical records. We need key information about presenting complaints, observations, streaming descisions and timings of investigations, treatments to compare with SortED® records. Inspection of the UCC patient’s records showed inconsistent use of fields so that automated abstraction was not likely to produce accurate results. Most of the data we needed to recover was embedded in plain text fields where each ‘streamer’ GP or the nurse/GP responsible for the later UCC consultation simply typed in their observations.

 

It is important to recognise that manual abstraction can be a higly error prone process. A thesis dissertation by M. Nahm ‘Data accuracy in medical record abstraction.’ reported error rates of 70 to 5018 errors per 10,000 fields for manual abstraction.  Analysis of error rates for re-abstraction in our ED trial showed us that transcription of date/time fields and retyping should be avoided, so we increased the use of pasted data, with built in cross-checking for time stamps.  We also made one-click data entry for most variables of interest.

 

This strategy worked well and re-abstraction tests of 221 records and 10 fields of different types found only 2/2210 errors (a normalised rate of 9.05 per 10,000 fields).

 

Gillie Francis – Jul 2017