Scottish Health Survey - topic report: The Glasgow Effect

Appendix 4: Heart attack model development

Many models are discussed in this section; all odds ratios given are for the sample with complete data on all variables contained in any of the 'final' models, allowing them to be directly compared. This sample size was 11,201. Participants aged 16 to 54 were combined into one category due to the very low prevalence of heart attacks in young people.

In the initial logistic regression model (Model 1) containing age, sex and residence, residents of Greater Glasgow and Clyde had an odds ratio of 1.46 of having had a heart attack, meaning their odds of having had a heart attack were 46% higher than the rest of Scotland. This model had a McFadden's pseudo R ² of 0.159.

When SIMD was added to the model (Model 2) all four variables were significant predictors of having had a doctor-diagnosed heart attack, with the odds ratio for Greater Glasgow and Clyde reduced slightly to 1.35 by the addition of SIMD, showing that around a quarter of the additional odds of having had a heart attack was explained by SIMD. However there was still a 35% increased risk for residents of Greater Glasgow and Clyde compared to the rest of Scotland. McFadden's pseudo R ² for this model was 0.165.

The third model included age, sex, residence, SIMD and the socio-economic variables. After backward selection had been performed the variables which remained in the model (Model 3) were age, sex, residence, economic activity, educational qualifications, household tenure and marital status. The odds ratio for residence in Greater Glasgow and Clyde remained very similar at 1.36, implying that after adjusting for age, sex and SIMD, adjusting for other socio-economic variables did not alter the excess risk found in residents of Greater Glasgow and Clyde. This model had a McFadden's pseudo R ² of 0.193.

The next model included age, sex, residence, SIMD, socio-economic and behavioural variables. After backward selection had been performed the variables which remained in the model (Model 4) were age, sex, residence, NS-SEC, household tenure, marital status, smoking status, abstaining from alcohol, drinking over the recommended weekly alcohol limit and physical activity level. The odds ratio for residence in Greater Glasgow and Clyde remained similar at 1.37, implying that this effect was independent of the behavioural variables added to the model. This model had a McFadden's pseudo R ² of 0.218, showing this model was a better fit than the previous model.

When all the biological variables were added the sample size was reduced to 974, with just one participant having suffered a heart attack. This was clearly not sufficient for analysis. There were also not a sufficient number of heart attacks for analysis when the nurse variables excluding the blood analytes were included in the model; therefore BMI was added to the socio-economic and behavioural variables, with the full sample and full sample weights used.

The variables which remained in the model after backward selection (Model 5) were age, sex, residence, household tenure, marital status, smoking status, abstaining from alcohol, drinking over the recommended weekly alcohol limit, physical activity level and BMI. The odds ratio for residents of Greater Glasgow and Clyde increased slightly to 1.44, implying that the BMI distribution in Greater Glasgow and Clyde would predict a lower rate of heart attack than in the rest of Scotland. This makes sense, as Greater Glasgow and Clyde had a lower prevalence of overweight than the rest of Scotland, and the odds ratio of having had a heart attack for being overweight compared to being a healthy weight was 1.42. This model had a McFadden's pseudo R ² of 0.224, showing a slight improvement over the previous model.

Parental NS-SEC, social mobility, single parenthood and being a stepparent were added for the final model (Model 6), and after backward selection had been carried out the model contained stepparent in addition to the variables in the previous model. The odds ratios for all variables remain virtually unchanged, with an odds ratio 1.44 for Greater Glasgow and Clyde. This model had a slightly higher McFadden's pseudo R ² of 0.229, which indicated that Model 6 was the best-fitting model.