Poster The Joint Annual Meeting of the Stroke Society of Australasia (SSA) and Smartstrokes 2023

Stroke Coding Accuracy – a retrospective Pre- and Post-Electronic Medical Record Implementation Study (#130)

Joshua Mahadevan 1 2 , Lizzie Dodd 1 , Tej Chuwan 3 , Michelle Hutchinson 4 , Andrew Moey 3 , Matthew WIllcourt 4 , Julie Morrison 5 , Monique Kilkenny 5 6 , Peter Psaltis 2 , Dominique Cadilhac 5 6 , Amanda Thrift 6 , Timothy Kleinig 1 2
  1. Department of Neurology/Stroke, Royal Adelaide Hospital, Adelaide, SA, Australia
  2. School of Medicine, University of Adelaide, Adelaide, SA, Australia
  3. Department of Neurology, Lyell McEwin Hospital, Salisbury, SA, Australia
  4. Department of Neurology, Flinders Medical Centre, Bedford Park, SA, Australia
  5. Florey Institute of Neuroscience and Mental Health, Heidelberg, VIC, Australia
  6. Stroke and Ageing Research, Department of Medicine, School of Clinical Sciences at Monash Health, Monash University , Clayton, VIC, Australia

Background:

Clinical coding is routinely used both for clinical and epidemiological research, health services planning and validation of data provided to the stroke registry. The Australian Stroke Clinical Registry (AuSCR) uses two methods to perform this surveillance: a primary ICD-10 code, and previously an any-of-first three code method. With wider electronic medical records (EMR) use, these methods, derived pre-EMR, require validation. Since 2017, the Royal Adelaide Hospital (RAH) has maintained a complete neurologist-validated ‘dual-ascertainment’ (prospective and retrospective coding-based) stroke database. Roll-out of RAH EMR (27/11/2019) allowed a coding accuracy comparison pre- and post-implementation.

 

Aims:

To quantify the accuracy of different AuSCR case ascertainment methods pre- and post-EMR implementation.

 

Methods:

Hospital ICD-10 coding data was extracted from pre-EMR (1/1/2017 to 26/11/2019) and post-EMR (27/11/2019 to 31/12/2022) dates. Patients positive for ICD-10 stroke codes in the primary diagnostic and first 3 codes were identified. Using the RAHs neurologist-validated stroke database as the gold standard, we calculated the sensitivity and positive predictive value (PPV) of a positive stroke code, pre- and post EMR initiation. 

 

Results:

Sensitivity of the first three diagnostic code position method improved following EMR implementation from 0.940 (95% CI 0.931 – 0.947) to 0.971 (95% CI 0.965 – 0.976), however the PPV declined (from 0.806 (95% CI 0.793 – 0.788) to 0.776 (95% CI 0.763 – 0.788)). Sensitivity of the primary code-only method also increased (from 0.897 (95% CI 0.886 – 0.907) to 0.935 (95% 0.926 – 0.943)) without significant change in the PPV (0.830 (95% CI 0.818 – 0.842) pre-EMR versus 0.816 (95% CI 0.803 – 0.828) post-EMR).

 

Conclusion:

EMR implementation improved diagnostic stroke coding sensitivity for both methods, however the PPV of the first three diagnostic code method declined.  The utility of these two methods may vary depending on a hospital’s EMR status and the purpose (i.e., maximising sensitivity vs evaluation of AuSCR case ascertainment comprehensiveness).