Mind the Gap: Addressing Missing Person Time When Estimating Outcome Incidence in Longitudinal Data.

Area of Science:

• Biostatistics
• Epidemiology
• Longitudinal Data Analysis

Background:

• Longitudinal studies frequently encounter missing data due to missed visits or participant dropout.
• These data gaps complicate the accurate estimation of outcome incidence.
• Understanding and addressing these gaps is vital for reliable research findings.

Purpose of the Study:

• To explore the fundamentals of data gaps in longitudinal datasets.
• To compare different methods for handling missing data when estimating outcome incidence.
• To evaluate the bias and precision of various approaches under different missingness mechanisms.

Main Methods:

• A simulation study involving 1000 individuals over 10 visits was conducted.
• Four data-generating mechanisms were used, varying the relationship between missingness and the outcome.
• Crude, inverse probability of censoring weights (IPCW), inverse probability of observation weights (IPOW), and multiple imputation (MI) methods were compared for estimating outcome risk and rate.

Main Results:

• Crude methods showed bias, especially when participants returned after a gap, even with independent missingness.
• IPCW and MI demonstrated relative unbiasedness across scenarios and outcome types; MI offered greater precision.
• IPOW methods were biased for permanent outcomes and less precise than IPCW or MI.

Conclusions:

• Crude methods that permit participants to return after data gaps are not recommended due to potential bias.
• These biased results can occur even when missingness is independent of the outcome.
• Researchers should opt for censoring or multiple imputation to accurately handle data gaps in longitudinal studies.