BAyesian Bent-Line Regression model for longitudinal data with an application to the study of cognitive performance trajectories in Wisconsin Registry for Alzheimer's Prevention

Lianlian Du,Rebecca Langhough Koscik,Tobey J Betthauser,Sterling C. Johnson,Bret Larget,Rick Chappell

DOI: https://doi.org/10.48550/arXiv.2211.09915

2022-11-18

Abstract:Preclinical Alzheimer's disease (AD), the earliest stage in the AD continuum, can last fifteen to twenty years, with cognitive decline trajectories nonlinear and heterogeneous between subjects. Characterizing cognitive decline in the preclinical phase of AD is critical for the development of early intervention strategies when disease-modifying therapies may be most effective. In the last decade, there has been an increased interest in the application of change point (CP) models to longitudinal cognitive outcomes. Because patients' change points can vary greatly, it is essential to model this variation. In this paper, we introduce a BAyesian Bent-Line Regression model longitudinal data on cognitive function in middle-aged adults with a high risk of AD. We provide an approach for estimating the fixed (group-level) and random (person-level) CPs, slopes pre- and post-CP, and intercepts at CP for cognition. Our model not only estimates the individual cognitive trajectories but also the distributions of the cognitive bent line curves at each age, enabling researchers and clinicians to estimate subjects' quantiles. Simulation studies show that the estimation and inferential procedures perform reasonably well in finite samples. The practical use is illustrated by an application to a longitudinal cognitive composite in the Wisconsin Registry for Alzheimer's Prevention (WRAP).

Methodology

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the non - linearity of the cognitive decline trajectory and the inter - individual heterogeneity during the pre - clinical stage of Alzheimer's disease (AD). Specifically, the paper introduces a Bayesian Bent - Line Regression model (BaBLR) for analyzing longitudinal cognitive function data in middle - aged adults, especially those at high risk of AD. The model aims to estimate the fixed (group - level) and random (individual - level) change points (CP), the slopes before and after the change points, and the cognitive intercept at the change points. In addition, this model can also estimate the distribution of the cognitive bent - line curve at each age, enabling researchers and clinicians to estimate the quantiles of subjects.

### Main contributions of the paper
1. **Model innovation**: Proposed a Bayesian bent - line regression model with random change points, which can better capture the differences between individuals.
2. **Parameter estimation**: Estimated not only the individual cognitive trajectories but also the distribution of the cognitive bent - line curve at each age.
3. **Application example**: Demonstrated the practical application value of the model by applying it to the dataset of the Wisconsin Registry for Alzheimer's Prevention (WRAP).

### Specific problems
- **Non - linearity of cognitive decline**: During the pre - clinical stage of AD, cognitive decline is non - linear and there are significant differences among different individuals.
- **Individual differences in change points**: The change point of each individual (i.e., the time point when cognitive function begins to decline at an accelerated rate) may be different, and the model needs to be able to capture this individual difference.
- **Early intervention strategies**: By more accurately identifying the critical time points of cognitive decline, provide a basis for early intervention, thereby improving the treatment effect.

### Model structure
- **Observation data**: Let \( y_{ij} \) represent the \( j \) - th consecutive outcome measurement value of the \( i \) - th subject at time point \( t_{ij} \).
- **Model form**:
\[
y_{ij}=\beta_{1i}+\beta_{2i}(t_{ij}-\omega_i)I(t_{ij}\leq\omega_i)+(\beta_{2i}+\beta_{3i})(t_{ij}-\omega_i)I(t_{ij}>\omega_i)+\epsilon_{ij}
\]
where:
- \( I \) is an indicator function.
- \( \omega_i \) is an individual - specific change point.
- \( \beta_{1i} \) is an individual - specific intercept, representing the expected value of the cognitive test score at the change point.
- \( \beta_{2i} \) is the individual - specific linear slope before the change point (pre - change slope).
- \( \beta_{3i} \) is the individual - specific value of the slope difference before and after the change.
- \( \epsilon_{ij} \) is the individual measurement noise.

- **Random effects**:
\[
\begin{bmatrix}
\beta_{1i}\\
\beta_{2i}\\
\beta_{3i}\\
\omega_i
\end{bmatrix}=
\begin{bmatrix}
\beta_{10}\\
\beta_{20}\\
\beta_{30}\\
\omega_0
\end{bmatrix}+
\begin{bmatrix}
u_{1i}\\
u_{2i}\\
u_{3i}\\
u_{4i}
\end{bmatrix}
\]
where \( \beta_{10}, \beta_{20}, \beta_{30} \) and \( \omega_0 \) represent the fixed (group - average) intercept, pre - change slope, slope difference and change - point age respectively, and \( u_{1i} \)