We have developed a new cardiovascular risk score, ASSIGN, to mitigate potential unfairness in the Framingham and similar risk scores when applied across different social groups in the same population. Nevertheless, with fewer variables than the ASSIGN score, the overall performance of the Framingham score, tested in the population from which ASSIGN was derived, is very similar. However, table 2 shows that like all other cardiovascular scores, performance in both scores is poor in terms of classical ideals of sensitivity and specificity, resulting in many false positives and false negatives. It might be anticipated that the addition of the extra, individually significant, risk factors to the ASSIGN score would improve overall prediction, but it is a common observation that there are rapidly diminishing returns in adding new factors to cardiovascular risk scores after the first small number. Cardiovascular risk scores are imperfect and resist attempts to perfect them, but they are needed for prioritising allocation of preventive treatment fairly to those at highest risk.
The Framingham score might therefore be preferred to ASSIGN, following the principle of parsimony of risk factors. However, the justification for ASSIGN is not greater discrimination, yet to be shown in other populations, but greater fairness. Its added complexity would be hidden in software for computerised data management in primary care. This would contain look‐up tables deriving the deprivation score from the postcode of residence of the individuals concerned, as well as consulting or requesting risk factor values, including cigarette consumption and family history. Because we have used continuous variables where possible, this score is more appropriate to electronic calculation than to the colour charts which use risk factor categories. The readout is the same: 10‐year risk of cardiovascular disease in the disease free, although again a continuous score rather than a category.
Our earlier analyses in the SHHEC population,2
like others, suggested that the Framingham coronary risk score overestimated risk. Based on observed/expected event rate ratios, it failed to compensate for social deprivation across SIMD fifths of the population, overestimating risk least in the socially deprived. These findings are repeated here for cardiovascular disease (table 2) with similar results and ratios but larger numbers of events. Social deprivation or socioeconomic status is not only a powerful determinant of coronary and cardiovascular risk but also of chances of reaching hospital alive in a coronary event.19
For these analyses we used mean Framingham score in the population group concerned to calculate “expected” event rates. After deriving the ASSIGN score, calibrated to 10‐year risk in the SHHEC population, we found that it too appeared to overestimate mean risk when it should not have done. We realised then that this paradox was explained by the skewed distribution of risk factor scores. Although the Framingham score does read too high in our population, and it does read higher than ASSIGN, the degree of overestimation is exaggerated using means (see fig 1 and table 2). What happens at the chosen high cutpoints is what matters in practice. This paradox was previously missed both by us and by others. It needs further exploration and discussion elsewhere.
Table 2 shows, using the Framingham score, an undesirable and significant social gradient both in observed/expected ratios, and in the event rate for unanticipated cardiovascular events by SIMD fifth, when using a convenient but artificial criterion for high risk in men and women of the top 20% of risk. These gradients are abolished by the ASSIGN score which redistributes high risk, and potential preventive treatment, towards the most deprived. In its own parent population it has therefore succeeded in its primary objective of social equity. But it needs testing elsewhere.
Apart from social deprivation, the ASSIGN score incorporates a quantitative measure of cigarette smoking where in Framingham it is yes or no.4,15
Attempts to characterise ex‐smokers were less successful. We recommend classifying them as smokers for the first year and then as non‐smokers.
Our third difference from Framingham is the use of family history. The original survey question was about heart disease in parents or siblings below the age of 60. Because younger people may not have parents or siblings who have reached 60, the question has been modified for future use to include premature stroke, and a positive history in several relatives such as uncles, aunts or cousins (see appendix 2). Apart from other advantages of incorporating family history, it may help with ethnic susceptibility. The SHHEC cohort was insufficiently heterogeneous to study risk in ethnic subgroups, but a positive family history was common. Susceptible groups, such as South Asians, could identify their risk through their family's medical history. A non‐threatening question, it avoids labelling people where there could be sensitivity or confusion on the part of the questioner or the questioned. We suspect family history replaced some of the risk associated with social deprivation since they are associated.
We were unable to find a simple adjustment to Framingham scores for social deprivation to make them similar to those from ASSIGN. They read higher on average, and dose of cigarettes and family history complicate a simple one‐factor change.
Our comparisons have given us considerable respect for the Framingham score whose coefficients for classic risk factors appear robust.20,21
Whether ASSIGN's marginally better discrimination and its coefficients for deprivation and family history apply elsewhere awaits further testing. Further comparisons will evaluate ASSIGN against Framingham with different cutpoints, age and sex and social distributions, both in other historical populations with equivalent follow‐up data, but also in modern populations such as the Scottish Health Survey22
to assess potential workload and economic consequences of its adoption. It needs installing into computerised databases for pilot testing in primary care.
Whether or not it performs marginally better than the Framingham score overall, ASSIGN addresses the issues of social deprivation and family history. Through greater fairness to disadvantaged, high‐risk, minority groups in the population, it should appeal to clinicians and to those responsible for health service strategy.