Methods for predictor analysis of repeated measurements: application to psychiatric data

Area of Science:

• Psychiatry
• Biostatistics
• Computational Neuroscience

Background:

• Schizophrenia research often overlooks methods for analyzing longitudinal data.
• Early intervention strategies for schizophrenia are crucial for patient outcomes.
• Prodromal symptoms require robust analytical approaches for effective prediction.

Purpose of the Study:

• To evaluate Generalized Estimating Equations (GEE) and Artificial Neural Network (ANN) methods for analyzing correlated response data in schizophrenia research.
• To investigate the predictive validity of prodromes for relapse in first-episode schizophrenia patients.
• To guide the development of prodrome-based early intervention strategies.

Main Methods:

• Application of Generalized Estimating Equations (GEE) for correlated response data analysis.
• Utilization of Artificial Neural Network (ANN) approach for data analysis.
• Analysis of data from the A.N.I. study, a large German multicenter long-term schizophrenia treatment study (1983-1989).

Main Results:

• Both GEE and ANN methods are demonstrated to be applicable to realistic, complex datasets.
• Statistical model selection was performed prior to GEE analysis.
• The prodrome 'trouble sleeping' emerged as the most informative predictor of relapse in the A.N.I. data.

Conclusions:

• Both GEE and ANN methods are suitable for predictor analysis in schizophrenia research, incorporating all variable time points to minimize bias.
• GEE allows for individual predictor association testing, while ANNs provide generalizable propositions for prodromes.
• 'Trouble sleeping' is identified as a significant predictor, highlighting its importance in early intervention strategies.