A Bayesian joint longitudinal-survival model with a latent stochastic process for intensive longitudinal data

Madeline R. Abbott,Walter H. Dempsey,Inbal Nahum-Shani,Lindsey N. Potter,David W. Wetter,Cho Y. Lam,Jeremy M.G. Taylor

DOI: https://doi.org/10.48550/arXiv.2405.00179

2024-05-01

Abstract:The availability of mobile health (mHealth) technology has enabled increased collection of intensive longitudinal data (ILD). ILD have potential to capture rapid fluctuations in outcomes that may be associated with changes in the risk of an event. However, existing methods for jointly modeling longitudinal and event-time outcomes are not well-equipped to handle ILD due to the high computational cost. We propose a joint longitudinal and time-to-event model suitable for analyzing ILD. In this model, we summarize a multivariate longitudinal outcome as a smaller number of time-varying latent factors. These latent factors, which are modeled using an Ornstein-Uhlenbeck stochastic process, capture the risk of a time-to-event outcome in a parametric hazard model. We take a Bayesian approach to fit our joint model and conduct simulations to assess its performance. We use it to analyze data from an mHealth study of smoking cessation. We summarize the longitudinal self-reported intensity of nine emotions as the psychological states of positive and negative affect. These time-varying latent states capture the risk of the first smoking lapse after attempted quit. Understanding factors associated with smoking lapse is of keen interest to smoking cessation researchers.

Methodology

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively process and analyze Intensive Longitudinal Data (ILD) in the context of mobile health (mHealth) technology. Specifically, the existing joint longitudinal - survival models have the problem of high computational cost when dealing with multivariate longitudinal data and event - time data, especially in the case of large amounts of data and frequent measurements. To solve this problem, the author proposes a new joint model, which is suitable for analyzing intensive longitudinal data and can capture the relationship between rapidly changing factors in longitudinal data and the risk of event occurrence.

### Main contributions of the paper

1. **Dynamic factor model**: Use the dynamic factor model to reduce the dimension of multiple longitudinal outcomes and extract a small number of time - varying latent factors. These latent factors can better reflect the changes in individual states.
2. **Continuous - time multivariate stochastic process**: Use the Ornstein - Uhlenbeck (OU) stochastic process to model the dynamic changes of latent factors. This model can flexibly capture the sudden changes of latent factors and avoid the complex specification of the number and position of knots in the spline - based method.
3. **Hazard regression model**: Incorporate latent factors as time - varying covariates into the parameterized hazard regression model to evaluate the impact of latent factors on the risk of event occurrence.

### Specific components of the model

1. **Measurement sub - model**: Use the dynamic factor model to describe the observed multivariate longitudinal outcomes. Assume that each longitudinal outcome is a measurement of latent factors, plus measurement error and individual - specific random intercept.
2. **Structural sub - model**: Assume that the dynamic changes of latent factors follow a multivariate OU stochastic process. This process can capture the autocorrelation and mean - reversion characteristics of latent factors.
3. **Survival sub - model**: Use a parameterized method to model the risk of event occurrence and incorporate the time - varying values of latent factors as covariates into the model.

### Application background

The paper takes a mobile health study on smoking cessation as an example to show the application of this model. In this study, the real - time emotional states and smoking behavior data of participants were collected through Ecological Momentary Assessments (EMAs). The researchers hope to use this model to explore the relationship between emotional states (including positive and negative emotions) and the time to the first smoking relapse.

### Method validation

To verify the validity of the model, the author conducted a simulation study to evaluate the bias of model parameter estimates and the coverage probability of confidence intervals. The simulation results show that the model performs well under different data - generation settings and can accurately estimate parameters and provide reliable inferences.

### Conclusion

This paper proposes a new joint model that can effectively handle intensive longitudinal data and achieves a good balance between computational efficiency and model flexibility. This method has important application value for understanding the relationship between emotional states and the risk of event occurrence, especially in mobile health research.