Efficient sparse estimation on interval-censored data with approximated L0 norm: Application to child mortality.

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Variable selection in proportional hazards models with interval-censored data is challenging.
• Existing regularization methods often require extensive hyperparameter tuning.

Purpose of the Study:

• To develop a novel penalty for variable selection in interval-censored failure time data.
• To create an efficient method that avoids time-consuming hyperparameter tuning.

Main Methods:

• A new penalty is proposed, approximating information criteria like BIC or AIC by smoothing the ℓ0 norm.
• Sieve likelihood is employed for simultaneous estimation of coefficients and baseline cumulative hazards.
• The method ensures continuity, sparsity, and unbiasedness of penalties.

Main Results:

• The proposed sparse estimation method demonstrates high accuracy and efficiency.
• Numerical results validate the effectiveness of the novel approach.
• The method successfully identified key factors affecting child mortality in Nigerian children.

Conclusions:

• The novel penalty offers an accurate and efficient solution for variable selection in interval-censored survival data.
• This approach simplifies the modeling process by reducing the need for hyperparameter tuning.
• The method has practical applications in identifying significant risk factors in health studies.