Penalized estimation of complex, non-linear exposure-lag-response associations

Area of Science:

• Biostatistics
• Epidemiology
• Critical Care Medicine

Background:

• Time-to-event data analysis is crucial in medical research.
• Modeling complex exposure-response relationships, especially with time lags, presents challenges.
• Existing methods may not fully capture cumulative effects of past exposures on hazard rates.

Purpose of the Study:

• To propose a novel, flexible approach for modeling exposure-lag-response associations in time-to-event data.
• To estimate various effects, including smooth, time-varying, and cumulative effects with leads and lags.
• To apply the method to analyze artificial nutrition's impact on critically ill patients' survival.

Main Methods:

• Utilizes inference methods for generalized additive mixed models (GAMMs).
• Develops a flexible framework for complex exposure-lag-response associations.
• Applies the method to observational data from intensive care patients.

Main Results:

• The proposed method allows flexible estimation of diverse exposure-lag-response patterns.
• Demonstrates the association between artificial nutrition timing/amount and short-term survival.
• Simulation studies confirm the method's properties and provide comparisons to related approaches.

Conclusions:

• The novel approach provides a flexible tool for analyzing complex exposure-lag-response associations.
• Offers insights into the relationship between artificial nutrition and survival in intensive care.
• The method's performance is validated through simulations and comparisons.