Time-to-event estimation by re-defining time.

Area of Science:

• Biostatistics
• Machine Learning
• Bioinformatics

Background:

• Time-to-event estimation models are crucial for predicting event occurrences.
• Existing models primarily focus on utilizing censored data.
• A fundamental constraint is that time is treated as a univariate number.

Purpose of the Study:

• To propose a novel time-to-event estimation model.
• To represent event occurrence times as vector concepts.
• To enhance model accuracy and interpretability by capturing feature-time vector relationships and relationships among time vectors.

Main Methods:

• Relaxing the univariate constraint of time, representing it as a vector.
• Developing a scalable framework for simultaneous learning of model parameters and time concept vectors.
• Applying the model to medical event prediction using gene expression data.

Main Results:

• The proposed model demonstrates efficiency and effectiveness in medical event prediction.
• Experiments on seven gene expression datasets validate the model's performance.
• Similarity analysis of time concept vectors identified potential cancer-related time regimes.

Conclusions:

• The vector representation of time concepts offers a new perspective for time-to-event analysis.
• The model enhances prediction accuracy and provides interpretable insights.
• This approach facilitates knowledge discovery in medical research, particularly in cancer studies.