Authors' replyBMJ 1996; 313 doi: https://doi.org/10.1136/bmj.313.7063.1006b (Published 19 October 1996) Cite this as: BMJ 1996;313:1006
- Richard J Davenport, Senior registrar,
- Martin S Dennis, Senior lecturer,
- Charles P Warlow, Professor of neurology
- University of Edinburgh, Department of Clinical Neurosciences, Western General Hospital, Edinburgh EH4 2XU
EDITOR,—All of the authors of the letters about our paper seem to agree with its central tenet. David Barer and colleagues and Steve Kendrick and Marion Bain suggest that our regression models may have overcorrected the data, and we discussed this possibility in our paper. While multiple logistic regression modelling is far from perfect, we thought that attempting to correct for case mix in this way was a more sensible approach than simply ignoring the problem. As discussed, we were reassured that we obtained almost identical results when we applied other prognostic models, which had been derived from and validated on different stroke datasets.1 We had developed these considerably simpler models (which rely on five or six variables, such as age and systolic blood pressure on admission) because we realised that collecting the amount of data included in our study is beyond the resources of most hospitals, at least in routine clinical practice. We are currently testing the applicability of these models in several hospitals (as referred to by Kendrick and Bain). We agree with the suggestion that the structure and process of care may be more appropriate measures of quality of care, although there are still considerable problems associated with this approach.2 3
We have conducted a similar study to that of Barer and colleagues, comparing the accuracy of routine hospital coding statistics with our stroke register.4 Of 566 patients registered as having a stroke, 84 (15%) were not given a code (International Classification of Diseases, ninth revision) for stroke; although better than Barer and colleagues' experience, this is far from perfect.
Finally, while we agree with Kendrick and Bain that the purpose of the Scottish outcome indicators is to raise meaningful questions, we are aware that they have been interpreted by many people as comparative league tables of hospitals' performance, and the purpose of our paper was to show the dangers of using the data as a direct measure of quality of care. As an example of this problem, the BMJ recently published a news item which suggested that the clinical outcomes initiative could identify hospitals which “seemed to perform badly.”5 We were interested to note that this report also raised doubts about whether the outcomes initiative would continue.