Piecewise transition models with random effects for unequally spaced longitudinal measurements

Abstract

In this paper, we consider the analysis of unequally spaced longitudinal data using transition regression models with random effects. Diffusion as well as stabilization processes will be discussed, but our main focus will be on the latter. The initial conditions problem, which usually arises in transition models with random effects, is addressed. The usefulness of the proposed model is assessed on a large database of longitudinal haemoglobin values collected from blood donations by a Dutch private organization.

Prediction of hemoglobin in blood donors using a latent class mixed‐effects transition model

“…Furthermore, our results support a donation interval longer than 56 days for both sexes, which has also been recommended in previous literature . The US Food and Drug Administration (FDA) is currently considering revising this interval to better protect donors .…”

“…p d i t is equal to H b i , t − 1 if the last visit was a donation. For a stationary process, i.e., | γ |<1, the correlation between two subsequent measurements can be expressed as …”

“…This will help to improve the planning of the donors' visits to the blood bank. Predicting Hb value has been studied previously by Rikhtegaran et al , and Nasserinejad et al , . In the latter study, the authors applied mixed‐effect models and transition models separately to predict the Hb value.…”

Prevalence and determinants of declining versus stable hemoglobin levels in whole blood donors

“…These prediction models were proposed as mixed effects models, transition models, or a combination of these two approaches for predicting Hb values in blood donors. 18,25 The current findings suggest that describing the total donor population using a single trajectory oversimplifies the complex growth patterns of this population. Instead, a growth mixture modeling approach, which accounts for different subgroups of donors, seems to be an appropriate method for capturing differences in Hb trajectories between donors.…”

“…The outcome in the linear mixed model is the Hb level. The predictors are age 15 at the first visit, season 16,17 of the visit (a binary covariate, i.e., the cold season includes fall and winter and the warm season includes spring and summer), a linear and quadratic effect of the time since the previous donation, 18 and the number of donations in the past two years. 5 Male and female donors have different Hb profiles; therefore, the data for men and women are analyzed separately.…”