Objective: To investigate the impact of left truncation on commonly used missing data approaches and explore this impact in a Parkinson’s disease dementia study.
Background: Missing data often arises in epidemiologic studies when covariates are expensive or invasive to collect. Ignoring observations with missing values can produce biased and inefficient results. One strategy to overcome these limitations involves predicting missing values by modeling the missing covariate distribution. Delayed enrollment of subjects into a time-to-event study (left truncation) produces a sample in which larger event times are overrepresented. The corresponding covariate sample is similarly biased. Research is needed to understand how bias in the sample covariate distribution affects the performance of general missing data approaches.
Method: We examine inverse probability weighting (IPW), augmented inverse probability weighting (AIPW), and multiple imputation (MI) in estimating the hazard ratio with the Cox model under a variety of truncation scenarios. Interest surrounds AIPW and MI, which involve predicting missing values and as such, may be susceptible to selection bias in the covariate sample. These methods are also used to assess the impact of cerebrospinal fluid (CSF) Aβ42 on conversion to dementia in Parkinson’s disease.
Results: When there is substantial truncation, the sample covariate distribution differs markedly from the population distribution, leading to bias in MI. The performance of AIPW depends on the validity of additional assumptions. While subjects with abnormal CSF Aβ42 values convert to dementia earlier than subjects with normal CSF Aβ42, the estimated effect changes depending on the missing data method used.
Conclusion: Left truncation can produce inaccurate predicted values for missing data. Thus, careful consideration of the amount of truncation and the assumptions of each method is necessary to obtain valid results in left truncated samples with missing data.
To cite this abstract in AMA style:H. Richardson, S. Xie. Methods for addressing missing covariate data in left truncated samples with application to Parkinson’s disease research [abstract]. Mov Disord. 2021; 36 (suppl 1). https://www.mdsabstracts.org/abstract/methods-for-addressing-missing-covariate-data-in-left-truncated-samples-with-application-to-parkinsons-disease-research/. Accessed November 29, 2023.
« Back to MDS Virtual Congress 2021
MDS Abstracts - https://www.mdsabstracts.org/abstract/methods-for-addressing-missing-covariate-data-in-left-truncated-samples-with-application-to-parkinsons-disease-research/