Re: Author response regarding the Letter to the Editor about the article ‘Temporal trends in the accuracy of hospital diagnostic coding for identifying acute stroke: A population-based study’. European Stroke Journal 2019 Oct 14. DOI: 10.1177/2396987319881017

We thank Dr Weber and colleagues for their interest in our paper.

We fully agree, and acknowledged in our paper, that coding accuracy is likely to differ between healthcare systems and the sensitivity of coding in identifying all incident stroke is expected to be higher in countries with higher hospitalisation rates, such as in Germany. ¹ In countries with good quality coding data, there is no doubt that nationwide analyses based on administrative coding can inform clinical practice. ² Nevertheless, it is important to recognise that changes of coding accuracy over time, even in countries with relatively good coding accuracy to begin with, may still introduce biases especially for studies looking at apparent trends in stroke incidence or severity. There can also be serious shortcomings in using routine coding data to identify strokes that occur during a hospital admission for another reason, which can lead, for example, to substantial underestimation of operative/procedural risks. ³

We also agree that ICD-10 code I64 (‘stroke, not specified as haemorrhage or infarction’) should be rarely used in studies focusing on ischaemic stroke, although I64 has been widely used in previous studies using ‘Big Data’ approaches. As rightly pointed out by Dr Weber and colleagues and also reported in our paper, brain imaging rates have increased substantially over the last couple of decades resulting in a significant decrease of the use of I64 in patients with definite strokes in more recent years (61.9% in 2002–2005 vs. 15.3% in 2014–2017). ¹ Therefore, completely excluding patients with otherwise unknown strokes (i.e. I64) could overestimate the incidence change over time.

Whilst it is important to acknowledge differences in coding practices between countries, it is also crucial to recognise that, in the current stroke literature using ‘Big Data’ approaches, the validity and generalisability of some studies can be undermined by the frequent lack of reporting details on how administrative coding data were acquired, searched and applied. Ideally, researchers publishing such work should also include an up-to-date validation of the validity of their coding data against a clinical gold standard. Joint efforts between the stroke physicians and the ‘Big Data’ community are still required to improve coding accuracy as well as to optimise reporting standards across countries.