My Library

University LibraryCatalogue

For faster,
Use Lean
Get it now
Don't show me again
Limit search to items available for borrowing or consultation
Result Page: Previous Next
Can't find that book? Try BONUS+

Check Minerva for Digitised Copy

Search Discovery

Search CARM Centre Catalogue

Search Trove

Add record to RefWorks

Author McGuinness, Myra

Title Quantifying the risk factors for age-related macular degeneration in the presence of survival bias

Published 2018


Location Call No. Status
Physical description 1 online resource
Thesis notes Thesis (PhD thesis)-- Melbourne School of Population and Global Health 2018
Summary Background:Age-related macular degeneration (AMD) is the most common cause of severe irreversible visual impairment among adults living in developed nations. The number of people living with AMD is projected to increase considerably over the coming decades, thus there is a real need to identify modifiable risk factors to delay its onset and progression. Longitudinal studies have been conducted to track exposures and AMD status over many years. However, those at risk of AMD also face the competing risk of death and participant selection for analyses becomes conditional on survival. Under these conditions, traditional statistical methods may produce biased results. Aims: The principal aims of this thesis are: to investigate the plausibility of shared survival-AMD risk-factors by examining the association between AMD and mortality, to examine the performance of a marginal structural model (MSM) and a sensitivity analysis approach when estimating the causal effect of an exposure on an outcome in the presence of survival bias and loss to follow-up, and to extend the sensitivity approach for analysis of a time-varying exposure. In addition, this thesis aims to explore the effect of survival bias when investigating selected risk factors for AMD, namely dietary iron intake, smoking and Mediterranean diet. Methods: Illustrative examples have been drawn from the Melbourne Collaborative Cohort Study (MCCS), a large community-based study conducted between 1990 and 2007. During that time three study waves were completed with assessment of AMD status at the final wave. Analysis of the data from the MCCS and a meta-analysis are presented to assess the association between AMD and mortality. The performances of a MSM (with inverse probability weights for exposure, survival and having non-missing data) and a sensitivity analysis approach for survival bias were assessed via the generation and analysis of simulated data. The sensitivity analysis approach was then extended to allow for the analysis of a time-varying exposure. The performance of this extended approach was compared to that of a MSM in a second simulation study. These statistical methods were then applied to data from the MCCS to examine the association between the selected risk factors and late AMD. Results: Late AMD was found to be associated with increased all-cause and cardiovascular mortality after adjusting for known confounders. MSMs produced biased estimates in the presence of simulated unmeasured survival-outcome confounders. The sensitivity analyses captured the true magnitude of effect within their bounds; however, these bounds were wider than what would be considered clinically useful. The analysis of the effect of smoking on AMD was found to be highly susceptible to survival bias; unlike the exposures of dietary iron and Mediterranean diet, for which the association with mortality was not as strong. Conclusion: The evidence from this thesis supports the theory of shared survival-AMD risk factors which are likely to be unmeasured in large cohort studies such as the MCCS. Therefore, survival bias will be present when the risk factor under investigation is also associated with survival. Sensitivity analyses for survival bias should be considered when investigating AMD risk factors in studies which have been impacted by attrition due to death.
Subject age-related macular degeneration biostatistics epidemiology causal inference survival bias