John Maynard Keynes once said: 'The difficulty lies not in the new ideas, but in escaping the old ones, which ramify, for those brought up as most of us have been, into every corner of our minds.' He might well have been describing the diffusion of new technologies into medical practice.
Although medicine is widely regarded as having a scientific basis, the evidence for this is poor. It has been estimated that it takes on average 17 years for 14 per cent of clinical research to translate into clinical practice. Other evidence has shown that only 55 per cent of the care delivered to patients meets the standards within the most up-to-date clinical guidelines.
If a cardiovascular risk prediction equation were developed today from a non-UK population and incorrectly predicted the outcome that it was designed to predict by a margin of 35 per cent, with the result that 1.5 million people might be misclassified, most rational people would no doubt treat that equation with derision.
In this context, the response to the development of QRISK casts useful light not only on the value of the new tool, but also the behavioural sociology of adherence to the previous tool.
The development and publication of QRISK raises significant questions for those committed to the practical implementation of effective strategies for reducing cardiovascular disease.
Framingham risk equations
The Framingham risk equations (there are numerous variants) are now more than 30 years old and the original dataset is more than 50 years old. The risk equation has been in routine use in the UK since at least the late 1990s.
Despite controversies in the approach to risk estimation, particularly concerning the Sheffield Tables, the Framingham risk equations have become the mainstay of risk prediction in the past decade.
Attempts have been made to adapt the equations to extend their prediction from CHD to cardiovascular disease and by adjusting for the prevalence of cardiovascular disease among ethnic groups (ETHRisk), because the original equations were developed from a mostly white cohort of European origin in the US. Understandably, the applicability of such risk equations to the diverse populations of our major cities, such as London, Birmingham, Manchester and Sheffield, is questionable.
QRISK and ASSIGN calculators
The recently published QRISK cardiovascular prediction tool is a novel development (BMJ 2007;335:136-41). QRISK was developed in a UK cohort, so it is a home-grown risk calculator developed to predict events within the UK population. It is a significant methodological development in the field of risk prediction. However, it is not entirely new; Scotland has introduced its own risk calculator, ASSIGN (Heart 2007;93:172-6). This is based on a relatively older cohort study carried out in Scotland.
Another significant feature of QRISK is the incorporation of measures of deprivation, again a feature shared by ASSIGN. Attempts to incorporate a measure of deprivation are particularly significant, given the marked and increasing health inequalities in cardiovascular disease.
Moreover, the underprediction of the Framingham equations in certain socioeconomic groups might result in reduced access for those most in need and with the potential to exacerbate health inequalities. Many patients are already receiving BP treatment, so the inclusion of a category for patients who have hypertension and are receiving treatment is welcome.
In practical terms, using real-time data from live GP databases allows QRISK to be updated at the touch of a button, rather than requiring unwieldy and lengthy cohort studies. QRISK stands out as an exemplar of the rising stock of general practice research that, even today, can generate scepticism and antipathy in many circles.
Ultimately, a risk calculator is only as good as its results. At the individual level, they are all poor and some more so than others. The performance metrics for QRISK in comparison to Framingham indicate an improvement. Framingham now appears to overpredict by a margin of 35 per cent, compared with QRISK. This would mean about 1.5 million patients being misinformed of their true risk, a point of principle worth considering, regardless of the risk thresholds at which patients receive clinical benefit.
More accurate risk assessment tools are needed. None of the current tools is precise enough to justify the term 'risk prediction'. QRISK is promising and given that the authors have addressed the lack of predictive power for total:HDL cholesterol, it requires validation within a further independent dataset, just as ASSIGN seemed promising within its original cohort but appeared less distinguished when applied to the QRISK cohort.
Whether QRISK will demonstrate improved performance in comparison with Framingham should determine the future of risk estimation and the possibility of the demise of the colourful risk charts that have long provided a valuable service in the absence of any alternative.
It is as well to remember that the poor performance of risk tools when applied to individuals and the relatively small clinical benefits of a statin-focused primary prevention strategy must not distract our attention from the 'causes of the causes' that will fuel cardiovascular event rates in the coming decades.
|CARDIOVASCULAR RISK CALCULATORS |
Framingham risk equations
- Dr Rubin Minhas is a GPSI in cardiology in Gillingham, Kent, and Medway PCT CHD lead. The views in this article are those of the author and not of any affiliated organisations.