Article Text
Abstract
Background Greater cumulative life-course exposure to socioeconomic disadvantage is associated with higher levels of inflammatory biomarkers, C-Reactive protein and fibrinogen which are associated with increased cardiovascular disease risk. However, in most existing studies, researchers often use complete case data for analyses and ignore the impact of missing data on inference for blood based biomarker data. The English Longitudinal Study of Ageing (ELSA) has considerable missing data but the impact of missingness on inference is seldom examined.
This paper aims primarily to examine whether the levels of adulthood inflammatory biomarkers of C-Reactive protein and fibrinogen can be explained by life course socioeconomic position. We evaluate the typologies of missing data under Missing Completely at Random and Missing at Random mechanisms and methods for compensating for missing data under these mechanisms.
Methods This paper uses cross-sectional data from Wave 2 of ELSA (2004) which includes 9432 men and women aged over 52 living in England. However, only 6000 people had data for C-Reactive protein and fibrinogen. Logistic Regression modelling is implemented to identify predictors of missingness in ELSA health examination and blood collection. We use multiple linear regression modelling for analysing the association between socioeconomic position and C-Reactive protein and fibrinogen after accounting for different mechanisms of non-response using the following methods: complete case analysis (listwise deletion), inverse probability weighting and multiple imputations.
Results Participants who refused to respond to the health examination were more likely to be renters, White, single, living in London or with poor assessed health. Those who refused to give a blood sample were more likely to be older female, living East of England or London, had cancer/cardiovascular disease/stroke and poorer self-assessed health. Complete case analysis showed that people with lower education level [0.14(CI0.04–0.25) and working in lower supervisory position [0.17(CI0.04–0.30)] were more likely to have higher C-Reactive protein levels. While people in the highest wealth quintile were less likely to have higher C-Reactive protein [−0.26(CI-0.36,–0.14)] and fibrinogen levels [−0.02(CI-0.04,–0.004)]. These associations remained similar in inverse probability weighting and multiple imputations although there was some variation in the estimates from the different methods for compensating for missing data.
Conclusion While associations between socioeconomic position and inflammatory markers were similar across different approaches for compensating for missing data, there were differences in the estimated coefficients suggesting that it is important to account for missing biomarker data for statistical inference.