The target population of the Whitehall study was 19,019 non-industrial London-based male government employees aged from 40 to 69 years at screening between September 1967 and January 1970 (response 74%).(21)

Socioeconomic position was based on civil service employment grade. To obtain sufficient numbers for each socioeconomic group and to simplify data interpretation, we dichotomised the level of employment grade into high (administrative, professional or executive) or low (clerical, messenger and other unskilled manual grades). Employment grade was not comparable to the rest of the sample for 886 men from the Diplomatic Service and the British Council so these individuals were excluded from all analyses. A single blood pressure reading was taken from the left arm by a trained observer using a London School of Hygiene sphygmomanometer.(22) Systolic blood pressure was recorded at the first appearance of the Korotkoff sounds. A capillary sample of blood was taken from the ear lobe after an overnight fast. Participants then drank a 50-g anhydrous dextrose preparation. Plasma cholesterol concentration was measured using the standard Technicon method N24a.(21) Two hours later a second blood sample was taken. Post-load blood glucose concentration was measured by the ferricyanide reduction method on an autoanalyser (Technicon method N-9a). Those with self-reported diabetes did not undertake the glucose tolerance test and those with a post-load blood glucose ≥11.1 mmol/L (≥200 mg/dl) were categorised as non-insulin dependent diabetics (non-IDDM). Diabetic participants on insulin treatment were classified as insulin dependent diabetics (IDDM). Smoking status was self-reported.

Fifteen-year mortality data were obtained from the National Health Services (NHS) Central Registry using the NHS identification number assigned to each UK citizen. As these records do not cover deaths among people who emigrated from the UK and died abroad, participants flagged as embarkees were censored at the date of embarkation. CHD death was indicated by codes 410–414 (International Classification of Diseases (ICD) 8 and 9) and codes I20-I25 (ICD-10).

We excluded from the analysis men who had any missing values for the risk factors of interest - systolic blood pressure (n = 6), cholesterol (n = 709), smoking status (n = 8), blood glucose/diabetes (n = 135). Difficulties in retrieving a sufficient sample of blood using the capillary method from an incision in the earlobe may have contributed to the higher rate of missing values for cholesterol (3.7%), but this missingness was non-differential by socioeconomic position (p=0.38). We also excluded 16 men who had died but for whom the cause of death was unknown. In all, 17,186 men were included in the analysis.

Follow-up after baseline was limited to 15 years to minimise the possible dilution of effects associated with extended mortality surveillance and to cover a period when statins had yet to become a commonly prescribed treatment for high cholesterol.(21, 23) In analyses of baseline characteristics and mortality, the prevalence of risk factors and 15-year CHD mortality rates between men in the high and low employment grade were adjusted for age (using 5 year age groups) by the direct standardisation method and compared using the Mantel-Haenszel test. For continuous measures, linear regression was used to calculate age-adjusted least squares means, and differences between the grades were tested using the dichotomous grade term.

The study sponsors had no role in study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

During the 15-year follow-up, 1262 men died from CHD. As expected, low employment grade and all of the risk factors were associated with elevated mortality (Table 2). Including best practice interventions in the model resulted in an overall decrease of 57% in the risk of CHD mortality at age 55. Lowering cholesterol alone was associated with a 33% reduction in CHD death. Corresponding reductions for the other risk factors were 23% for smoking, 14% for high blood pressure, but only 2% for intervening in diabetes risk.

The primordial prevention model suggested that removing all risk factors from the study population reduces CHD mortality by 73%. Risk reductions associated with the risk factors individually were 32% for high blood pressure, 33% for high cholesterol, 3% for diabetes and 37% for smoking.

Fifteen-year risk of CHD death at age 55 is 11.0 per 100 for men in the low employment grades and 7.5 for those in the high grades, a between-group risk difference of 3.5 (table 3). Including best practice interventions in the model reduced this risk difference by 43% for smoking, 31% for cholesterol, 13% for blood pressure, and 6% for diabetes risk. All these interventions together reduced the 15-year CHD mortality to 4.4 per 100 for men in the low grades and 3.3 for men in the high grades. Thus, the estimated post-intervention risk difference between high and low employment grade groups is 1.1, a reduction in absolute risk disparity of 69%.

In relative terms, the ratio of 15-year CHD mortality risk between the high and low employment grade groups is 1.46 and the corresponding post-intervention risk ratio is 1.32. The overall 30% reduction in the risk ratio is almost entirely attributable to the effect of smoking cessation which accounted for 28%. As employment grade differences in the prevalence of other coronary risk factors were small, best practice intervention reducing these risk factors has little effect on the relative risk associated with low socioeconomic position.

Interventions that reduced all the outstanding non-optimal levels of risk to optimal levels and completely eliminated non-IDDM and smoking (i.e. all men considered to be never smokers) further reduced the population means: systolic blood pressure 115.0 mm Hg, cholesterol 3.1 mmol/L, post-load glucose among non-diabetics 4.0 mmol/L, and no exposure to smoking. The 15-year CHD mortality risk under these circumstances is 2.2 per 100 in the high grade men and 2.7 in the low grade men. This represents an overall reduction in absolute risk disparity of 86%. For relative risk, the reduction in socioeconomic disparities is 53%, with smoking and blood pressure the most important contributing factors.

Numerous population-based cohort studies have examined the contribution of the classic risk factors to relative socioeconomic inequalities ignoring the contribution to absolute differences in CHD between the socioeconomic groups. Our quantification suggests that using interventions shown to affect the greatest reduction in CHD in randomised controlled trials (best practice interventions) reduces the relative difference in CHD between high and low socioeconomic groups only by 30%. Previous studies using adjustment-based approaches suggest a similar reduction of 15% to 40% in relative risk.(5, 7) The adjustment-based method differs from ours in that it approximates what would happen after standardization to the risk factor distribution in the total study population irrespective of whether this is possible.

Smoking cessation for all smokers made the greatest contribution to both relative and absolute difference in CHD between high and low socioeconomic groups. Reductions in other risk factors had only a modest effect on relative inequalities in CHD. Although smoking cessation for all smokers is the ultimate aim of all health policy, this may seem a very optimistic goal at a population level. Re-calculating the overall best-practice intervention effect with smoking prevalence only halved suggests a 55% (rather than 69%) drop in mortality differences between high and low grades and a 12% (not 30%) reduction in the relative risk. In this scenario, total cholesterol reduction becomes the factor that generates the greatest reduction in absolute socioeconomic disparities in CHD mortality.

The relative approach to socioeconomic inequalities is only sensitive to reductions in risk factors that are socially patterned. As cholesterol levels were not higher in low socioeconomic groups, reduction in cholesterol did not reduce relative socioeconomic inequalities in CHD mortality despite resulting in substantial reduction in the CHD mortality difference between the socioeconomic groups. This starkly illustrates that the determinants of relative socio-economic inequalities may be different to the determinants of absolute differences.(5)

The proportion of people who had optimal risk factor levels was very small; <2% in our cohort. Other studies, with less stringent criteria, found 15% in a Finnish sample(5) and 5% in a multiracial study of men in the US Atherosclerosis Risk in Communities study.(9) Getz et al calculated that in Norway, one of the healthiest nations in the world, about 86% of men at age 40 would be classified as high risk and none at low risk using the 2003 European guidelines criteria.(25)

Although there are currently no evidence-based strategies or tools to completely eliminate the classic coronary risk factors from the total population, this may not necessarily be the case in the future. In addition to best-practice prevention, interventions that target upstream risk factors, such as the psychosocial and material circumstances in which people live and work, are suggested to be effective in reducing relative inequalities.(26–28) Lifetime risk factor prevention starting early in life is often seen as the key to eliminating most CHD and reducing absolute differences between socioeconomic groups.(29) The two prevention strategies are not mutually exclusive. For example, attempts to reduce smoking levels in populations through fiscal policy will both prevent initiation of smoking (and ultimately may essentially eliminate smoking) and reduce absolute differences in smoking levels between social class groups (even if relative differences increase).

Our analysis suggests that any prevention programme that completely eliminates the four classic coronary risk factors from the population would reduce CHD deaths by 73% and absolute socioeconomic disparities by 86%. True primordial prevention strategies may have even greater potential than suggested here, as for several factors we only categorise risk according to measures taken in middle-life. We do not have estimates of risk associated with lifetime low levels of circulating cholesterol, which will be lower than those associated with low levels at only one stage of the lifecourse. Recent studies that have examined genetic variants associated with different lifetime levels of cholesterol have shown that reductions in CHD risk obtained in randomised controlled trials of relatively short (less than 10 year) treatment duration, and observational studies initiated when the study members were middle aged, under-estimate the absolute differences in CHD rates associated with longer-term differences in cholesterol levels.(30, 31) The same logic applies to other classic risk factors. In isolated, non-westernised populations with a non-sedentary lifestyle and a low-fat diet free of added salt and refined carbohydrate, such as the Yanomamo Indians, average systolic blood pressure is about 100 mm Hg, total cholesterol 3.1 mmol/L, and there is little evidence of cardiovascular deaths among these people.(19, 32) Thus we will be under-estimating the potential of true primordial prevention in our analyses. A caveat that would influence estimates in the opposite direction, but probably to a lesser extent, is that a small proportion of individuals with marked genetically-influenced elevations in cholesterol, glucose or blood pressure(33–36) would have levels above those used as our cut-offs.

The validity of our estimates depends on the underlying assumptions and the data used. Instead of risk factor categories, we used continuous measures of blood pressure, cholesterol and blood glucose. This may have improved precision. Our analyses of possible risk factor reductions were based on the effects of best practice interventions derived from randomised controlled trials or, where possible, meta-analyses of such trials, that is, the strongest possible evidence.(11–13) This approach may provide a more realistic estimation for policy than those based on risk factor standardisation or comparisons of exposed and unexposed groups. Differential treatment between socioeconomic groups is an unlikely source of bias in this study as treatment for blood pressure was very limited (1.7% in both high and low grades) and effective cholesterol-lowering interventions had yet to become available.(21) Obtaining the prevalence of risk factors and associated excess risks of CHD from the same population (rather than from different populations) is also a strength of this study.

Our study of course is not without its limitations. First, the findings are based on data from a male occupational cohort and it could be argued may not translate to women or to non-working populations. However, differences between the participants and British men aged 45–64 in 2003 were small for total cholesterol (5.1 vs 5.5 mmol/L) and systolic blood pressure (137 vs 135 mm Hg)(15) and there is little evidence of differences in the effects on cardiovascular disease of the studied risk factors between men and women, except for greater diabetes-related risk of fatal CHD in women.(37, 38) Prevalence of smoking was high in this study population compared to more contemporary populations and is likely to overemphasise the role of this behaviour in generating absolute socioeconomic inequalities.

Second, we estimated the benefits from interventions for each risk factor separately while controlling for others. However, we did not take into account the potential interaction effects between different risk reductions because comparable clinical evidence relates to single- rather than multiple-factor interventions. Our predictive models were also unable to capture lag effects or time trends in risk factors and CHD mortality.(39) Further, we estimated the benefits of best practice risk factor reductions in terms of reduced CHD mortality from observational data which are open to confounding. Despite these potential limitations, our estimates of intervention effects did not substantially deviate from those reported for CHD morbidity and mortality in clinical trials: an estimated 33% reduction in CHD from cholesterol lowering in this study versus 20%-36% reductions observed in lipid lowering trials;(40) a 15% reduction for systolic blood pressure lowering in our analysis versus 10%-20% reductions in relative risk of cardiovascular events in blood pressure lowering trials.(41) Although observational data suggest these reductions may be dependent on age, trials provide little evidence of any age-specific effects, supporting our analysis combining all age groups.(40, 41) As we discuss above, with lifetime reductions in these risk factors, greater effects than those predicted in our observational data or seen in short-term randomized controlled trials would be expected.

Finally, we assumed that best practice interventions are implemented equitably and have similar effectiveness in both high and low socioeconomic groups. Obviously, this is a major challenge for policy as there are many barriers to realizing the full potential of interventions in all groups. These include unhealthy environments not conductive to salubrious modes of life, together with limited financial and health care resources. These constraints disproportionately affect low socioeconomic groups.(3)

This study provides an example of a novel way of quantifying socioeconomic inequalities in health. Although the aim of health practitioners and governments is to reduce the classic coronary risk factors to optimal levels and eliminate smoking, population-wide comprehensive screening and best practice interventions remain to be implemented. Our estimations from a working population suggest that such interventions, if implemented equally successfully in high and low socioeconomic groups, would eliminate most of the socioeconomic inequalities in CHD.