Statistics from Altmetric.com
SSV model is a useful research tool
In their paper (see pp 401),1 Counsell et al compare a “six simple variable” (SSV) model, which they developed for predicting outcome after stroke,2 with two other simple models and with clinical predictions of stroke outcome, and find that their model is at least as good as the other evaluated predictive systems.
Predictive models of stroke outcome may be useful in epidemiological studies and clinical trials of stroke to stratify cohorts by baseline severity, and they could also guide patient management. A multitude of predictive models already exists,3 and one might argue that yet another model is unlikely to add anything new. However, the SSV model is attractive for several reasons: the predictor variables are all based on history and examination and are very easy to collect, and the model has been extensively validated on population and hospital based cohorts, and a clinical trial population.2,4 In this paper, the authors show that the SSV model performs as well or better than two pre-existing models. While this is an important finding, the authors were unable to compare it with any of the other existing models, and we do not know if these might have performed better.
Of more interest is the comparison of the SSV model with clinical judgement. A prognostic model would only be useful in clinical practice if it performed at least as well as a clinician,5 and it therefore must be evaluated against clinical judgement. This is the first study to do so, and the authors find little difference between clinical judgement and the statistical model’s performance, although they suggest that the model may perform better when compared with the judgement of less experienced clinicians. Even if further studies proved this to be true, the model is still limited to predicting whether a patient will be alive and independent at 1 year. This is only a rather crude outcome measure, because the model cannot take into account type of disability, social circumstances, and other factors, which should play an important part in the physician’s prediction of the overall outcome for an individual patient. Even if accurate, the model’s clinical usefulness will therefore be limited.
Furthermore, it is important to realise that using prognostic models in clinical practice may already represent a clinical intervention5—for example, if only patients with a poor prognosis according to the model were given a specific kind of treatment. Before such models are used in clinical practice, they should therefore not only be shown to be accurate, but also their usefulness should be demonstrated in clinical trials.
The authors quite rightly advise against the use of their model as a guide towards clinical management at present. However, the SSV model is undoubtedly a useful research tool. Epidemiological studies and clinical trials would be much more comparable if they used a single model to stratify according to prognosis or baseline severity rather than each using their individual model. Because of its ease of use, and because of its extensive validation, the SSV model here seems to be a step in the right direction.
SSV model is a useful research tool
Declaration of interest: none
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.