New analyses for income and occupational social class

We have now repeated all our main analyses using a method identical to that described in the full paper1 with additional adjustments for income and occupational social class. Occupational social class was determined using the Registrar General’s classification and was grouped as I (professional), II (managerial/technical), III (skilled) non-manual, III (skilled) manual, IV (semiskilled manual) and V (unskilled manual). Income was converted to equalised annual household income that is adjusted for the number of persons in the household. Due to missing income data, approximately 35%–40% of the full sample1 was excluded (all-cause mortality analyses, n=52 031; CVD mortality analyses, n=48 965). Despite this substantial loss of participants, characteristics (table 1 of the original paper, data not shown) were very similar to the full sample.1 Similarly to previous published data,9 we observed a direct socioeconomic gradient for participation across all sport groupings, for example, football/rugby participation decreased from 5.7% in occupational class I and 4.2% in class II to 2.4% in class IV and 1.4% in class V. Equivalent participation figures for running were 10.5%, 8.0%, 3.2% and 1.7%; for swimming 20.2%, 18.2%, 14.4%, 9.3% and 6.6% and for racquet sports 8.2%, 5.9%, 1.7% and 0.9%. Results of the expanded prospective analyses are presented in table 1. Compared with the analysis with adjustment for education only (model 1), there were no or minimal changes when income and occupational social class were taken into account (model 2) across all sports. Since the inclusion of highly correlated socioconomic covariates may bias mortality risk estimates, we also carried out separate analyses with income only and social class only as covariates but there were no appreciable differences compared with the data presented in table 1 of the original paper (data not shown). We also repeated analyses using less conventional markers of socioeconomic status as covariates such as economic activity, number of cars and number of bedrooms but none of these influenced the results (data not shown).

We urge readers that the analyses described in table 1 solely serve the purpose of comparing the estimates between model 1 and model 2 (prior to vs after adjustment for income and social class). The most robust and accurate estimates describing the associations between each sport type and mortality are those presented in the main study1 that used the maximal possible sample size and largest number of events.

In conclusion, the data presented here suggest that despite the direct socioeconomic gradient in participation, the type-specific beneficial associations between sports and all-cause and CVD mortality are not explained by socioeconomic confounding. Socioeconomic factors do not seem to explain the lack of associations between football/rugby and running and mortality either. Instead, these null associations could be due to the younger baseline age of the participating groups, and due to participants perhaps retiring from these activities as they get older, and losing the health enhancing effects.

The key messages of our study remains that (a) we need to develop and promote health-enhancing sport programmes to reach more people from all ages across all socioeconomic groups and (b) when sports participation is not a possibility to support opportunities to adopt new active choices both in everyday life and during recreation.