Clustering with missing and left-censored data: A simulation study comparing multiple-imputation-based procedures.

Area of Science:

• Biostatistics
• Bioinformatics
• Computational Biology

Background:

• Cluster analysis is vital for exploring large biomedical datasets.
• Missing or left-censored data present significant challenges in clustering.
• Existing methods like complete-case analysis or simple imputation are often inadequate.

Purpose of the Study:

• To develop a novel consensus-based clustering algorithm that addresses left-censored data.
• To incorporate a modified multiple imputation method for handling censored data.
• To allow the number of clusters to vary across imputed datasets for improved accuracy.

Main Methods:

• Developed a consensus clustering algorithm incorporating a modified multiple imputation technique.
• Accounted for left-censored data by adjusting the imputation process.
• Estimated the number of clusters for each imputed dataset independently.

Main Results:

• Simulation studies demonstrated superior performance compared to alternative methods.
• The algorithm accurately estimated the number of clusters and reduced unclassified observations.
• Achieved a high adjusted Rand index, indicating robust clustering performance.

Conclusions:

• The proposed consensus clustering method effectively handles left-censored data in biomedical analyses.
• This approach reveals novel patient clusters, as evidenced by a real-world breast cancer study.
• Offers a more accurate and robust clustering solution for challenging biomedical datasets.