Scottish Health Survey - topic report: The Glasgow Effect

Topic report in the Scottish Health Survey series investigating the existence of

This document is part of a collection

1. Introduction and Methodology

1.1 Introduction

The link between socio-economic circumstances and health is well known, and has been widely investigated, with deprivation found to be a key factor for a variety of health outcomes. One such health outcome is mortality. Scotland has the highest mortality rate in western Europe among the working age population, and has done since the late 1970s 1.

Carstairs and Morris 2 analysed data from 1980 - 1982 investigating whether social class and deprivation could explain the excess mortality experienced by Scotland compared to England and Wales. They found that standardising for social class had little effect, whereas standardising for relative affluence and deprivation greatly reduced the difference. However the impact of deprivation on the difference in mortality between Scotland and England and Wales has been found to have reduced since 1981; using census data from 1981, 1991 and 2001, Hanlon et al 3 found that whilst in 1981 deprivation explained over 60% of the excess mortality found in Scotland, in 1991 and 2001 deprivation explained less than half of the excess mortality. The excess mortality increased from 4.7% in 1981 to 8.2% in 2001 after adjusting for age, sex and deprivation. The largest excesses have been found in the most deprived areas of Scotland.

Work published earlier this year compared the health outcomes in Glasgow with those of almost identically deprived cities Liverpool and Manchester 4, and found that premature deaths in Glasgow were over 30% higher, with the excess mortality found across men and women, all ages except the very young, and both deprived and non-deprived neighbourhoods. Approximately half of the excess premature deaths were found to be directly related to alcohol and drugs.

Other recent work has investigated whether the mortality excess relates to country of residence or country of birth, as it is known that those born in Scotland who live in England and Wales have a higher mortality rate than those born in England and Wales 5, and those born in England and Wales but living in Scotland have a lower mortality rate than those born in Scotland 6. Popham et al 7 therefore compared mortality by country of birth and country of residence, with the effect of country of residence attenuated by country of birth, but not the other way round.

This recent work has shown that there exist factors beyond deprivation which influence the excess mortality rate found in Glasgow. Many hypotheses have been suggested, including societal breakdown leading to self-destructive behaviours and adverse childhood experiences and the Glasgow population's response to them 4.

Much of the work investigating the 'Glasgow Effect' has focused on mortality as an outcome; it is also of interest to know whether there is a 'Glasgow Effect' for other health outcomes and health behaviours, which themselves influence mortality. A report written for the Glasgow Centre for Population Health in 2006 8 examined the levels of many health behaviours in Glasgow City and Greater Glasgow compared to the rest of Scotland, and found many examples of worse health behaviours, including alcohol consumption, diet and smoking, as well as worse health outcomes, such as higher prevalence of limiting long-term illness. A piece of work carried out by the Glasgow Centre for Population Health in 2008 9 compared health indicators in Greater Glasgow with those in areas across Europe. It found that Greater Glasgow had the worst levels for a number of health behaviours and health outcomes, including binge drinking, excess weekly alcohol consumption, self-assessed general health and psychological morbidity.

Using data from the 1995, 1998 and 2003 Scottish Health Surveys, Gray 10 investigated the impact of living in Glasgow City, Greater Glasgow and West Central Scotland on a range of health-related factors, covering mental health, physical health and health behaviours, and the extent to which adjustment for socio-economic conditions explained any effects. The socio-economic conditions adjusted for contained both area-level and individual-level deprivation, using the Carstairs measure of area-level deprivation, social class, educational qualifications and economic activity. The study found that the levels of binge-drinking and alcohol consumption in men were higher than in the rest of Scotland, even after adjusting for Glasgow's socio-economic profile, as were the levels of psychological distress for both men and women. However adjusting for socio-economic conditions accounted for many of the worse health behaviours and outcomes in Glasgow, implying that improving Glasgow's health is strongly linked to addressing the socio-economic conditions in Glasgow. More detailed conclusions from Gray's report are discussed at the end of each section, alongside the results from the analyses carried out in this study.

1.2 Aims

The overall aim of this work was to investigate whether residence in Glasgow was independently associated with poorer health outcomes and worse health behaviours compared to the rest of Scotland, after controlling for socio-economic, behavioural, biological and other health-related risk factors. The supplementary research questions are:

1. To what extent do socio-economic factors explain differences in health-related outcomes?

Previous analyses have examined the role of area based deprivation in explaining poor health related outcomes in Glasgow 10, however these were based on the Carstairs measure of area-based deprivation at postcode level, which is less spatially sensitive than the Scottish Index of Multiple Deprivation ( SIMD), which is measured at datazone level, with an average population of only 750. These previous analyses only controlled for socio-economic factors, and therefore did not investigate other possible explanations for the remaining effect of residence after adjusting for socio-economic factors. The analyses in this report used data from the 2008 and 2009 Scottish Health Surveys, whereas the previous analyses used data from the 1995, 1998 and 2003 Scottish Health Surveys.

As part of this aim the socio-economic factor which best explained both the health outcomes and the differences between Glasgow and the rest of Scotland was investigated.

2. To what extent are differences in health-related outcomes influenced by 'relationship'-based factors?

An advantage of using the Scottish Health Survey data to investigate the 'Glasgow Effect' is the wealth of data available, including the relationships between members of each household. There were not sufficient foster parents or adoptive parents in the study to examine their effect on the various outcomes; therefore only single parenthood and being a stepparent were examined.

3. Are aspects of 'social mobility' significantly associated with health and health-related outcomes?

One of the current hypotheses relating to the 'Glasgow Effect' is the effect of social mobility 11. Therefore the effect of social mobility on both health behaviours and health-related outcomes were investigated, as well as their impact on explaining any effect of residence in Greater Glasgow and Clyde.

1.3 Methodology

The combined 2008 and 2009 Scottish Health Survey datasets were used to carry out this work. The combined dataset contains information on 18,353 individuals, with 13,996 (76%) aged 16 and over, of whom 7,866 (56%) were female. A subsample from both the 2008 and 2009 surveys were selected for a nurse visit to collect biological measurements, and some of these participants agreed to provide a blood sample. As these subsamples are not representative of those who agreed to take part in the original survey, new weights were developed to allow analysis of these complete subsamples to provide results which are representative of Scotland's population. More information on the sample design and data collection is available in the 2008 and 2009 Scottish Health Survey reports 12, 13.

In order to investigate the 'Glasgow Effect' an area must be identified to be compared with the rest of Scotland. Greater Glasgow and Clyde was chosen as data on Greater Glasgow and Clyde health board is more representative, and more data is available, than if just Glasgow City had been used. This made the results more robust. 3,242 adults in Greater Glasgow and Clyde provided data in the main sample, with 504 in the nurse subsample and 392 in the blood subsample.

An initial logistic regression model was carried out for each outcome of interest with explanatory variables age, sex and residence in Greater Glasgow and Clyde, and a second model added SIMD quintiles. Explanatory variables were then added to this model in groups, so the new model contained the new explanatory variables as well as all the explanatory variables previously entered into the model. Backward selection was carried out after each group of variables had been added, until all the variables in the model were significant at the 5% level. More details on all variables can be found in the Scottish Health Survey 2009 main report 12.

The first group of explanatory variables contained socio-economic risk factors:

  • Income-related benefits (receiving at least one of job seekers allowance, income support or housing benefit)
  • National Statistics Socio-economic Classification ( NS-SEC) (categorised as: managerial and professional occupations, intermediate occupations, small employers and own account workers, lower supervisory and technical occupations and semi-routine occupations, as well as a category for people for whom the NS-SEC is not applicable, such as full-time students)
  • Economic activity (full time education, paid employment/self-employed/government training, looking for/intending to look for work, permanently unable to work, retired, looking after home/family, doing something else)
  • Highest educational qualifications attained ( HNC/D or degree level or higher, Standard Grade or Higher Grade, other school level, none)
  • Housing tenure (owner occupied, private rental and social rental)
  • Marital status (single (never married or in a civil partnership), married/civil partnership and living together, married/civil partnership but separated, divorced/civil partnership legally dissolved, widowed/surviving civil partner).

The next group covered behavioural risk factors:

  • Smoking status (never/ex-occasional, ex-regular, light, moderate, heavy/don't know how many a day 14)
  • Binge drinking (more than 6 units per day for women, and 8 units for men)
  • Drinking over the recommended weekly alcohol limit (more than 14 units per week for women, and 21 units for men)
  • Abstaining from alcohol consumption
  • Scoring 2 or more on the CAGE questionnaire to identify potential problem drinking 15
  • Level of physical activity (high (30 minutes or more at least 5 days a week), medium (30 minutes or more on 1 to 4 days a week) or low (fewer than 30 minutes of activity a week))
  • Portions of fruit and vegetables consumed per day.

The third group contained biological risk factors:

  • Collected from everyone:
    • BMI (<25 kg/m 2, =25 kg/m 2 and <30 kg/m 2, =30 kg/m 2)
  • Collected from those who had a nurse visit:
    • Waist-hip ratio (high if =0.95 for men, =0.85 for women)
    • Blood pressure (normotensive untreated, normotensive treated, hypertensive untreated, hypertensive treated)
    • Forced expiratory volume in one second
  • Collected from those who had a blood sample taken:
    • Total cholesterol (above or below 5 mmol/l)
    • HDL-cholesterol (above or below 1 mmol/l)
    • Fibrinogen (sex-specific quintiles)
    • C-reactive protein (sex-specific quintiles).

The analyses were adjusted for the complex survey design, and different survey weights were used depending on the variables included in the model. Although BMI was collected from everyone in the main sample, the other biological variables were collected during the nurse visit. As the sample size is reduced for the nurse variables, and reduced again for the blood variables, if all the blood variables dropped out of the model then the model was re-run excluding the blood variables, and based on the nurse weights, thereby enabling a larger sample to be used. If all the nurse variables then dropped out of the model, it was re-run with only BMI added to the model, using the full sample and therefore the full sample weights.

The fourth group were relationship variables and social mobility variables:

  • Single parent
  • Stepparent
  • Parental NS-SEC (the higher of the mother's and father's NS-SEC)
  • Social mobility (indicating whether the participant was upwardly mobile, downwardly mobile or stable by comparing parental NS-SEC and individual NS-SEC. Participants who did not have an NS-SEC category were not assigned a social mobility category.)

Not all predictor variables were added to each model if it was not appropriate; for example BMI and high waist-hip ratio were not added to the models analysing outcomes of being overweight or obese. Any variables which were not included in the modelling are mentioned in the relevant section. All analyses were restricted to participants aged 16 plus.

As not all variables have complete data, the sample size varies depending on which variables remain in the model. For direct comparisons to be made between models using odds ratios and pseudo R-squared values, for the purpose of determining the model which provides the best fit to the data (see Appendix 1), it is important to maintain a constant sample 16. Therefore after the final model was selected using all available data at each stage, the resulting models from adding each group of variables were re-run on data restricted to include participants with full data on all variables included in any of the models, and these are the results reported. Other results are reported as required.

To investigate which socio-economic factor best explained the health outcome and the difference in health outcomes between Glasgow and the rest of Scotland, the final model was run using all available data, and then run containing each of SIMD, NS-SEC, economic activity, household tenure, educational qualifications, receiving income-related benefits and equivalised income in place of all the socio-economic variables (including SIMD, but excluding marital status) which were in the final model. McFadden's pseudo R 2s were compared to find the socio-economic variable which best explained the health outcome, with the highest pseudo R 2 indicating the best model. For the models where a "Glasgow Effect" remained, the odds ratios for residence were compared, with the lowest odds ratio indicating the model which explained the largest proportion of the difference.

Back to top