Identifying risk factors for Alzheimer's disease from multivariate longitudinal clinical data using temporal pattern...

Area of Science:

• Computational biology
• Data science
• Medical informatics

Background:

• Clinical data presents challenges for modeling due to heterogeneity and varied time series.
• Existing temporal pattern mining algorithms are not optimized for complex clinical datasets.
• Understanding disease onset and progression requires effective modeling of longitudinal patient information.

Purpose of the Study:

• To introduce an efficient and scalable framework, Clinical Temporal Pattern Mining (C-TPM), for temporal pattern mining of real-world clinical data.
• To identify clinically relevant and high-risk temporal patterns from multivariate, longitudinal patient data.
• To improve the computational efficiency and interpretability of clinical data analysis.

Main Methods:

• Developed C-TPM, integrating temporal abstraction, an extended TPMiner algorithm, relative risk, odds ratio, and multiprocessing.
• Applied a complete set of cut-off values for data discretization and interpretation.
• Utilized a visualization module for clear pattern representation.

Main Results:

• Applied C-TPM to Alzheimer's disease (AD) datasets, discovering patterns predictive of AD.
• Achieved a high Concordance index (up to 0.87) in survival analysis models using discovered patterns.
• Identified clinically relevant variables within the discovered temporal patterns.

Conclusions:

• C-TPM offers an effective and scalable method for modeling complex longitudinal clinical data.
• The framework can identify patterns in rare or slowly progressing diseases.
• C-TPM is generalizable to various medical and non-clinical data domains.