Individual results in our paper should be viewed within the context of our overall analysis plan. In the whole sample (aged 30-75 years), people with SMI were more likely to have a raised CHD risk score for their age and gender on univariate analysis. We explored ‘effect modification’ or ‘interaction’ by age using conventional statistical methods to determine whether interaction was important in predicting CHD risk and found that SMI does predict excess CHD risk even after adjustment, but the heterogeneity of results at different ages cannot be ignored. Several results were only significant in those under 65 (e.g. Table 2) and we illustrated this age interaction in two figures.

Dr Gilleard quotes one insignificant adjusted odds ratio which includes people of all ages. In fact, this odds ratio demonstrates that the findings are less striking when differences between age-groups are ignored.

We agree that in the overall sample some dichotomous results lost significance after adjustment for unemployment, although most odds ratios still did not approach unity. However, SMI still predicted CHD risk after adjustment when the age interaction was included in a statistical model. Furthermore, continuous lipid and risk score variables were also predicted by SMI in those under 60 even after adjustment for unemployment (Table 2).

The clinical importance of the interaction is that excess cardiovascular risk is demonstrable in younger people with SMI. Consistent with this we have recently found that excess mortality from cardiovascular disease is also more pronounced in younger people with SMI (Reference Osborn, Levy and NazarethOsborn et al, 2006).

Contrary to Dr Gilleard's assertion, Tables 2 and 3 and Fig. 2 show that CHD risk in those under 60 is not simply reducible to smoking. Risk also relates to differences in cholesterol ratios, diabetes and hypertension, as we stressed in the conclusions of our abstract.