Controlling for selective dropout in longitudinal dementia data: Application to the SveDem registry
Ron Handels, Linus Jönsson, Sara García‐Ptacek, Maria Eriksdotter, Anders Wimo
Abstract
INTRODUCTION: Loss to follow-up in dementia studies is common and related to cognition, which worsens over time. We aimed to (1) describe dropout and missing cognitive data in the Swedish dementia registry, SveDem; (2) identify factors associated with dropout; and (3) estimate propensity scores and use them to adjust for dropout. METHODS: Longitudinal cognitive data were obtained from 53,880 persons from the SveDem national quality dementia registry. Inverse probability of censoring weights (IPCWs) were estimated using a logistic regression model on dropout. RESULTS: The mean annualized rate of change in Mini-Mental State Examination (MMSE) in those with a low MMSE (0 to 10) was likely underestimated in the complete case analysis (+1.5 points/year) versus the IPCW analysis (-0.3 points/year). DISCUSSION: Handling dropout by IPCWs resulted in plausible estimates of cognitive decline. This method is likely of value to adjust for biased dropout in longitudinal cohorts of dementia.