Using trajectories from a bivariate growth curve as predictors in a Cox regression model

Area of Science:

• Psychiatric clinical trials
• Longitudinal data analysis
• Biostatistics

Background:

• Identifying predictors of illness recurrence is a key objective in psychiatric maintenance treatment trials.
• Longitudinal response measures during acute treatment are believed to predict future recurrences.
• Existing methods may not fully account for estimation errors in predictive modeling.

Purpose of the Study:

• To develop a novel statistical method for predicting illness recurrence in psychiatric maintenance treatment.
• To utilize bivariate longitudinal response trajectories from the acute treatment phase for prediction.
• To adjust for parameter estimation errors in Cox regression models for improved accuracy.

Main Methods:

• Employed a bivariate growth curve model to estimate individual subject trajectories from longitudinal data.
• Integrated these estimated trajectories into a Cox regression model to predict recurrence.
• Applied a full likelihood approach, using conditional expectations, to correct for parameter estimation errors.

Main Results:

• Simulation studies demonstrated that the proposed estimation error-corrected Cox model estimators are less biased than naive estimators.
• The method effectively estimates trajectories from bivariate, unequally spaced longitudinal response measures.
• The approach was illustrated using data from a major depression maintenance treatment trial in the elderly.

Conclusions:

• The developed method provides a statistically robust approach to predicting psychiatric illness recurrence using longitudinal data.
• Accounting for estimation errors in predictive modeling leads to more accurate parameter estimates.
• This technique offers a unique way to leverage complex longitudinal data for improved clinical trial outcomes.