Risk models for cardiovascular disease could be misleading, finds study
Current risk prediction models for cardiovascular disease may be misleading at the individual level because of large variations in practice data, a study has found.
Analysis of records from 3.6 million patients across 392 GP practices found considerable variability in the quality of data available.
The study, by a team at University of Manchester, showed that for a QRISK3 predicted score of 10%, an individual patient could have a risk of between 7.2% and 13.7%, depending which practice they came from.
Variation between practices were so great that a ‘substantive number of patients’ would fall into a different treatment category after this was taken into account, the researchers said.
Writing in Scientific Reports, the University of Manchester team concluded that risk prediction models based on routinely collected health data perform well for populations but ‘with great uncertainty for individuals’.
Study leader Professor Tjeerd van Staa said: ‘We looked at a large number of general practices and we say a large variation in the incidence of cardiovascular disease and when we looked further that wasn’t taken into account by the QRISK model.’
The variation could be to do with how practices record data as well as other lifestyle or generic factors that are not taken into account by the model, he explained.
‘The model gives you an estimate but our findings suggest you need to take that with a pinch of salt because there could be wide variation around that number.’
QRISK alone should not form the basis of a decision around treatment because that could lead to both under or overtreatment, he added, and it is questionable whether they should be used ‘without clinical interpretation’.
There has been concern that use of risk scores may overmedicalise large proportions of the population.
Dr Samuel Finnikin, a GP in Sutton Coldfield and research fellow at the University of Birmingham, said the paper raised some valid points about the application of risk scores.
He said: ‘Risk estimates are the starting point for a conversation, not the be all and end all. But it is important that we don’t discount them due to their imperfections.
‘If we didn’t have risk calculators we would be relying on a much less accurate and reliable way of estimating risk - just using clinical judgement alone.’
He added: ‘It’s not the information that is a problem - but how we use it. We must combine the risk estimate with our clinical skills.’
Professor Azeem Majeed, head of primary care and public health at Imperial College London, agreed that risk scores need to be used alongside additional clinical interpretation.
‘Ideally, a risk score should be presented with its measure of uncertainty (using methods such as 95% confidence intervals). However, I am not aware of any clinical system that is able to do this.
‘Guidelines bodies such as NICE treat risk scores as "fixed". However, statistically, there is not much difference between a QRISK score of 9.9% (below threshold for statin treatment) and 10.1% (above threshold for statin treatment),’ he added.