The Studies

In the Whitehall study, 18,403 men aged 40–64 years were examined between 1967 and 1969 [12]. Of this total, 15,395 of the men were aged 45–64 years (the age group recruited in the Renfrew & Paisley cohort). The Collaborative cohort of men and women was recruited from 27 workplaces in Glasgow, Grangemouth, and Clydebank (in central Scotland) between 1970 and 1973 [13]. Response rates were available for the workplaces from which 87% of the sample was recruited. For these sites 70% of those invited completed the questionnaire and attended for examination. The achieved sample included 6,022 men, of whom 4,021 were aged 45–64 years. The Renfrew & Paisley general population study was carried out between 1972 and 1976. The sampling frame was residents of the towns of Renfrew & Paisley (in central Scotland) aged 45–64 years, and a 78% response was achieved. Full details of the study methodology have been reported previously [14]. A total of 7,049 men were included in the study. Due to their geographical locations, there were 26 men who took part in both the Collaborative and Renfrew & Paisley studies. To ensure they were only included once in the combined analysis, their records from the Renfrew & Paisley study were not used, leaving 7,023 men.

The questionnaire and examination procedures were similar in all three studies as they had been developed from an earlier occupational health survey carried out in the west of Scotland between 1964 and 1967 and there was close collaboration between the cohort founders Victor Hawthorne and Geoffrey Rose [15]. Cohort participants completed a questionnaire to collect data on demographics, occupation and smoking habit. In the Scottish cohorts, occupation was coded and assigned to social class using the Registrar General Classification. In the Whitehall study, civil service employment grade was categorised as administrative, professional or executive, clerical, and “other grades” (men in ‘messenger’ and other unskilled manual jobs) and matched to the Registrar General Classification of social class [16]. Smoking was categorised according to cigarette, pipe or cigar use (as “never smoker”, “ex-cigarette smoker”, “current cigarette smoker” or “current pipe or cigar smoker”). In addition, the number of cigarettes smoked per day was recorded and controlled for in the smoking adjusted analyses. The Rose angina questionnaire [17] and the Medical Research Council respiratory questionnaire [18] were also included. As part of these questionnaires persistent phlegm was defined as: usually bringing up phlegm from the chest first thing in the morning on most days for three months during winter each year. “Infective phlegm” was defined as: usually bringing up phlegm from the chest first thing in the morning in winter and having had a period of increased cough and phlegm lasting for three weeks or more in the previous three years. Breathlessness was defined as a positive response to the question: “Do you get short of breath walking with people of your own age on level ground?” and bronchitis as having persistent or “infective” phlegm and being breathless. Angina was considered present if chest pain or discomfort when walking uphill or hurrying was cited in the sternum or the left chest and arm; caused the subject to stop or slow down; went away when the subject stopped or slowed down; and went away in 10 minutes or less.

The examination measurements included height, weight, blood pressure, lung function, a six lead electrocardiogram and plasma cholesterol concentration in all three cohorts. Heights were measured with shoes on in the Whitehall study, so a deduction of 2.54 cm (1 inch) in height was made for each participant to improve comparability with the other cohorts (a sensitivity analyses was also conducted using a deduction of 1.27 cm, or ½ inch). In the Whitehall study, the cholesterol measures taken later in the baseline data collection period were systematically lower than those at earlier times and may have been adversely affected by a change in the concentration of the laboratory standard over time. The cholesterol values were therefore adjusted to their predicted values as if they were taken at the start of the baseline data collection period (which had the effect of increasing the mean cholesterol in the Whitehall cohort from 5.12 to 5.69).

The electrocardiogram was coded according to the Minnesota system [19] and was regarded as positive for ischaemia if Q/QS items (codes 1.1–3), ST/T items (codes 4.1-4 or 5.1-3), or left bundle branch block (code 7.1) were present. Forced expiratory volume in 1 second (FEV₁) was also recorded. The FEV ratio was calculated for each individual as the ratio of measured FEV₁ divided by predicted FEV calculated from the subjects’ age and height based on an equation derived in those men free of respiratory symptoms in the Renfrew & Paisley study.

Records were traced and flagged at the National Health Service Central Registry. Death certificates coded according to the eighth, ninth or tenth revision of the International Classification of Diseases (ICD), were obtained. Mortality was classified as being due to coronary heart disease (ICD8/9: 410–414; ICD10: I20–I25), stroke (ICD8/9: 430–438; ICD10: I60-I69), respiratory disease (ICD8/9: 460–519; ICD10: J00–J99), lung cancer (ICD8/9: 162; ICD10: C33–C34), accidents and suicide (ICD8/9: E800–E999; ICD10: S00-Y98) or alcohol-related deaths (ICD8/9: 141, 143–6, 148–9, 150, 155, 161, 291, 303, 571 or E800–E999; ICD10: C01–C06, C10, C13–C15, C22, C32, F10, K70, K74.6 or S00-Y98). Mortality follow-up was until 31^st December 2008.

Statistical Analysis

We excluded from the three studies a total of 1,775 (6.8%) men with missing values for any of the covariates and those who were not followed up for mortality. This resulted in an analytic sample of 24,653 men (13,884 Whitehall, 3,956 Collaborative and 6,813 Renfrew & Paisley). The prevalence of baseline characteristics in the three studies was adjusted for age (5-year age groups) using direct standardisation with the combined population as the standard. Differences in prevalence between the studies were tested for significance using the Mantel-Haenzsel test. For continuous variables, least-squares means were used to present the age-adjusted means and the significance of the study group variable was used to test for heterogeneity.

Mortality rates, by follow-up period and overall, were calculated using person years at risk and were standardised for age at entry as above. Hazard ratios and 95% confidence intervals for mortality in the Scottish cohorts compared to the Whitehall study were computed using Cox’s proportional hazards regression model with follow up time as the time scale. Initially, models were adjusted for age and each of the potential explanatory factors separately. Subsequently, multiply adjusted models controlled for all factors both including and excluding social class. An analysis stratified by social class was also performed adjusting for all other factors. Adjustment for BMI in these models was achieved by including both linear and quadratic BMI terms [20]. All other continuous measures were controlled for by including a single linear term. To adjust for smoking we used three indicator variables to compare ex-smokers, pipe/cigar smokers and current smokers with never smokers (with one indicator variable for each comparison) and fitted a term that adjusted for the numbers of cigarettes smoked per day in the current smokers. Proportional hazard models were also fitted within each of the social class groups to examine the magnitude of the hazard ratios and to assess the effect of controlling for the explanatory factors within each group. Further models were fitted to compare the hazard ratios and the effect of the adjustments by stratifying the follow-up into periods 0–9 years, 10–19 years and ≥20 years.

Strengths and Weaknesses of the Study

This study provides a unique insight into the higher mortality in Scotland and the ‘Scottish Effect’ by using very similar cohort study data from Scotland and England to adjust for baseline differences in characteristics known to influence health outcomes. It therefore provides a complementary analysis to the existing ecological analyses [7], [8] and studies using individual mortality data with a shorter follow-up period [10]. It uses individual-level data and includes several objective biological measures in addition to self-reported behaviours and morbidity.

Two of the cohorts were drawn from workplaces and so may the results not be applicable to the general population. There are very large baseline differences between the Whitehall cohort and the Scottish cohorts relating to the much higher proportion of individuals in social class I and II in Whitehall. Although this is adjusted for in the model, it is possible that there remains confounding due to the very different types of occupations recruited in the studies (such as the higher proportion of manual workers, and the potential for direct occupational exposures [22], in the Collaborative study). We have therefore presented analyses stratified by broad social class categories to reduce the potential for residual confounding. However, there may be other unmeasured aspects of social class and deprivation which are not captured in this measure (for example, area-based deprivation [23]), and the process of combining social class categories in order to provide sufficient precision of hazard ratio estimates, may have under-differentiated the social differences and left residual confounding.

The slightly later period of recruitment in the Scottish cohorts, against a background of population health improvements as was seen in the UK at this time, may have resulted in an underestimate of the mortality difference between the cohorts since a cohort recruited slightly earlier would have been expected to have had higher mortality.

The sensitivity analysis of the height correction in the Whitehall cohort, to allow for measurement with shoes on, did not make a large difference to the adjusted hazard ratios and so is unlikely to endanger the validity of the analysis. The sensitivity analyses which stratified by period of follow-up showed similar patterns of results within each period suggesting that the decline in the hazard ratios between the Scottish and Whitehall cohorts over time is unlikely to have affected the overall results. The limited sample sizes in the analyses stratified by social class make the estimates of the excess imprecise and make it difficult to be certain that the observed higher mortality rates in the Scottish cohorts are not due to chance.

Plasma cholesterol levels are a potentially important explanatory factor in the cohort comparisons because they are, on average, lower in the Whitehall cohort. Unfortunately, the cholesterol measures in the Whitehall study demonstrated a downward trend over time between the first samples taken and later samples, suggesting that there may be a systematic error (towards an underestimate) in the measures related to a change in the laboratory standard. We have therefore applied a correction factor based on the predicted cholesterol level at the start of the analysis period, but this may not have fully eliminated the bias in favour of lower cholesterol in the Whitehall cohort. This issue was discussed in earlier publications of data from this cohort [24], and the possibility of some small residual bias is reinforced by the finding in the British Regional Heart Health study of no difference in plasma cholesterol levels between the Scottish sample towns and English sample towns [25]. Overall, given the small impact of cholesterol on the hazard ratios, it is unlikely that any remaining bias would have a large impact on the overall results.

We were limited by a lack of comparative data on alcohol intake, physical activity and diet in the cohorts. It may not therefore be surprising that alcohol-related deaths and mortality due to accidents and suicide were not explicable across the cohort studies given that the baseline risk factors captured in these cohorts were designed to explore cardiovascular disease aetiology. However, given that the fully adjusted model accounts for all of the excess all-cause mortality in the Scottish cohorts, this does not seem to have been important for all-cause mortality.

There were no further consistent waves of data collection across all three cohorts to allow changes in occupation, socioeconomic circumstances or risk factors over time to be accounted for. Given that changes are likely to be socially patterned (for example, smoking cessation and medical treatment for hypertension are both likely to have been more common amongst social classes I & II, and there was a marked rise in alcohol-related mortality in deprived areas during the 1990s [26] and a large increase in relative poverty in Scotland during the 1980s and 1990s [27]), this could have underestimated the impact of these risk factors.

A generalisation of the findings of this study to the populations of Scotland and England & Wales has to be cautious since the Whitehall and Collaborative studies are occupational cohorts and unlikely to be representative of the population as a whole (because of the ‘healthy-worker’ effect). Furthermore, we did not have comparable data for women to carry out any analyses for females.

Comparison with Previous Studies

A recent synthesis of the various hypotheses to explain the higher mortality in Scotland noted that there are two distinct time phases in the pattern [28], [29]. The earlier phase from around 1950 to 1980 saw higher mortality rates in Scotland as compared to the rest of western Europe largely due to deaths from cardiovascular disease, stroke, respiratory disease and cancer in middle-aged men and women. Prior to this, Scotland had a similar life expectancy to most other western European countries [1], [3] but with marked inequalities in mortality within the nation. From 1980 onwards the pattern of mortality changed. Causes of death which had thus far been relatively uncommon in Scotland such as liver cirrhosis, illicit drug-related deaths, suicide and violent deaths began to increase amongst a younger group of men and women, in addition to a continuing trend of higher deaths due to cardiovascular disease, stroke, respiratory disease and cancer in middle-aged men and women [2]. Indeed, premature male mortality increased from 1981 (and even more markedly from 1991) in some persistently deprived areas in Glasgow [11].

The cohorts included in this study were recruited in their middle age during the late 1960s and early 1970s. By 1981 (the first year in which we have been able to measure the ‘Scottish Effect’ using Carstairs deprivation index data), the surviving (and therefore healthier) participants in the Whitehall study were aged 52–78 years, in the Collaborative study were aged 53–75 years, and in the Renfrew & Paisley study were aged 50–73 years. Therefore, the Scottish cohorts are more likely to be representative of the populations affected by the earlier divergence in Scottish mortality (between 1950 and 1980) and of the continuing higher mortality amongst elderly men from cardiovascular disease, stroke and cancer than of the divergence in premature mortality occurring after 1980 which affected a younger age cohort. This therefore suggests that, not withstanding the difficulties in generalising from the occupational nature of two of the three cohorts, socio-economic circumstances may be the most important explanatory factor for the higher mortality during this earlier period of mortality divergence between 1950 and 1980, and amongst elderly men from the 1980s onwards.

Implications of the Study

This study suggests that socio-economic status is the most important explanation for the higher mortality amongst Scottish middle aged men during the 1970s and amongst late middle aged and elderly men in the 1980s and 1990s. This means that the differences in observed male mortality are health inequalities as they arise from the social circumstances in which people live. This paper supports the large existing body of literature which suggests that a redistribution of income and power between socioeconomic groups and geographical areas is likely to be an effective means of reducing health inequalities [30], [31]. Comparisons of more recent and younger cohorts may be able to generate better insights into the more recent divergence in premature mortality between Scotland and England and in determining the causes of the ‘Scottish Effect’ which emerged from 1981 onwards.

Conclusions

The higher all-cause male mortality in two Scottish cohorts compared to an English cohort all of which recruited participants in the late 1960s and early 1970s was accounted for mostly by differences in social class, and almost entirely by social class and cardiovascular risk factors, but not for some specific causes (stroke, alcohol-related causes, accidents and suicide). This provides further support for policy which redistributes income and power as a means to reducing health inequalities between social classes and geographical areas. The use of cohort studies which recruited younger groups from the 1980s onwards in Scotland and England may be able to shed further light on the causes of the divergent premature mortality trends during this later period.