Propensity Score Matching with Survival

Propensity score matching with survival analysis pairs treated and untreated subjects with similar estimated probabilities of treatment, then compares time-to-event outcomes between the matched groups using Kaplan-Meier curves and Cox models.

생존분석을 결합한 성향점수매칭은 처치를 받을 추정확률이 비슷한 처치군과 비처치군을 짝지은 뒤, 매칭된 집단 간 사건 발생 시간을 카플란-마이어 곡선과 콕스 모형으로 비교하는 방법입니다.

How do I estimate a treatment's effect on survival from observational data?

관찰자료에서 처치가 생존에 미치는 효과는 어떻게 추정하나요?

Propensity score matching with survival analysis reduces confounding by measured covariates by matching treated and untreated subjects on their estimated probability of treatment before comparing time to event. It fits observational cohort data with a binary treatment, censored outcomes, and enough overlap in propensity scores, after which balance is checked with standardized mean differences.

When to use Propensity Score Matching with Survival

• Observational cohort comparing a treated and an untreated group on time to an event
• Many baseline covariates that differ between treatment groups
• Censored follow-up such as mortality, relapse, or readmission

When to use something else

• Matching would discard many subjects: use inverse probability weighting (IPW)
• Treatment changes over follow-up: use marginal structural models

Assumptions and requirements

• No unmeasured confounding
• Overlap (positivity) in propensity scores between groups
• Covariate balance after matching, commonly standardized mean differences below 0.1
• Non-informative censoring and proportional hazards for the Cox model

What the analysis reports

• Propensity score distributions before and after matching
• Covariate balance table with standardized mean differences
• Number of matched pairs
• Kaplan-Meier curves and log-rank test in the matched sample
• Hazard ratio with robust confidence interval

How to report it

After 1:1 nearest-neighbor matching ([n] pairs; all standardized mean differences < 0.1), treatment was associated with a lower hazard of [event], HR = [value], 95% CI [[value], [value]], p = [value].

Related methods

• Inverse Probability Weighting (IPW)
• Cox Proportional Hazards Regression
• Propensity Score Matching

Run Propensity Score Matching with Survival on Quorant