Abstract High levels of inflammatory and stress-related biomarkers have been linked with several health conditions in older adults. Living in socioeconomic disadvantage may affect the levels of the biomarkers, however, previous findings are not consistent. Previous studies have used complete case analysis and ignored the high proportion of missing biomarker data in biosocial surveys. Longitudinal studies examining ageing populations are susceptible to attrition and non-random dropout and ignoring missing data can produce biased estimates due to selection processes and loss of precision. This thesis investigated socioeconomic differences in inflammatory biomarker C-reactive protein and stress-related biomarkers cortisol and cortisone after compensating for missing data. The English Longitudinal Study of Ageing (ELSA) was used for the analyses. Complete case analyses were compared with methods considering random missingness: Inverse Probability Weighting, Full Information Maximum Likelihood, and Multiple Imputation, and non-random missingness: Diggle-Kenward and Pattern-Mixture approaches. Differences between the least and most disadvantaged categories of education, wealth, and social class in C-reactive protein and cortisol and cortisone levels existed after adjusting for covariates. C-reactive protein levels were higher in the inverse probability weighting and multiple imputation models compared to complete case models in cross-sectional analysis. In longitudinal analysis, the C-reactive protein levels were higher in the Diggle-Kenward model compared to the other models considering random and non-random missingness. Socioeconomic differences in cortisol and cortisone levels were greater in the inverse probability weighting and multiple imputation models compared to the complete case models. The conclusions drawn suggest that living in socioeconomic disadvantage was a significant predictor of higher levels of inflammatory and stress-related biomarkers and that complete case analyses may underestimate the socioeconomic differences in biomarkers compared to missing data approaches. This study demonstrates the importance of compensating for missingness in longitudinal biosocial studies for statistical inference.
- social inequalities
- missing data