Lasso Cox Regression

Lasso Cox regression is a penalized proportional hazards model that adds an L1 penalty to the Cox partial likelihood, shrinking some coefficients to exactly zero to select predictors of time-to-event outcomes.

Lasso Cox 회귀는 Cox 부분우도에 L1 벌점을 더해 일부 계수를 정확히 0으로 축소함으로써 생존시간 결과의 예측변수를 선택하는 벌점화 비례위험모형이다.

How do I select predictors of survival when there are many candidate variables?

후보 변수가 많을 때 생존에 영향을 주는 예측변수를 어떻게 선택하나요?

Lasso Cox regression fits a Cox proportional hazards model with an L1 penalty, choosing the penalty strength by cross-validation so that only informative predictors keep non-zero coefficients. It fits high-dimensional survival data, such as genomic or radiomic features, or settings with many predictors relative to events.

When to use Lasso Cox Regression

• Many candidate predictors relative to the number of events
• Building a prognostic signature from genomic, imaging or clinical features
• Time-to-event outcomes with censoring

When to use something else

• Few predictors chosen in advance and interest in inference: use standard Cox proportional hazards regression
• Predictors are highly correlated and should be retained as groups: use elastic net Cox regression

Assumptions and requirements

• Proportional hazards
• Non-informative censoring
• Standardized predictors before penalization
• Penalty parameter chosen by cross-validation

What the analysis reports

• Selected predictors and shrunken coefficients
• Hazard ratios for selected predictors
• Cross-validated partial likelihood deviance versus lambda
• Coefficient path plot
• Concordance index (C-index)

How to report it

Lasso Cox regression with [k]-fold cross-validation selected [n] predictors at λ = [value], and the resulting risk score achieved a C-index of [value].

Related methods

• Cox Proportional Hazards Regression
• Kaplan-Meier Survival Analysis
• Random Survival Forest

Run Lasso Cox Regression on Quorant