生存分析¶

方法简介¶

Kaplan-Meier 生存曲线和 Cox 比例风险模型是时间-事件数据（OS、PFS、TTR 等）的标准分析方法。

代码实现¶

RPythonSAS

library(survival)
library(survminer)

# KM 估计
km_fit <- survfit(Surv(avalu, event) ~ treatment, data = adtte)

# Log-rank 检验
survdiff(Surv(avalu, event) ~ treatment, data = adtte)

# Cox 回归
cox_fit <- coxph(Surv(avalu, event) ~ treatment + strata, data = adtte)
summary(cox_fit)

from lifelines import KaplanMeierFitter, CoxPHFitter

# KM
kmf = KaplanMeierFitter()
kmf.fit(durations=adtte['avalu'], event_observed=adtte['event'])

# Cox
cph = CoxPHFitter()
cph.fit(adtte[['avalu', 'event', 'treatment', 'strata']], 
        duration_col='avalu', event_col='event')
cph.print_summary()

proc lifetest data=adtte plots=survival;
    time avalu * event(0);
    strata treatment;
run;

proc phreg data=adtte;
    class treatment strata;
    model avalu * event(0) = treatment strata / risklimits;
    hazardratio treatment;
run;

相关可视化¶

注意事项¶

比例风险假设检验 (Schoenfeld residuals)
非比例风险时的替代方法 (RMST, 加权 log-rank)