I have a dataset where I'm comparing Survival (overall and cancer-specific survival) between two treatment groups (Surgery vs. Radiation) for prostate cancer. As suggested by Noordzij et al (PMID 23975843). I will be using this guide for choosing methodology:
NO Competing risks
- Prognostic research question (i.e.Calculation of Survival probability) a. Unadjusted: Kaplan Meir method b. Adjusted: Multivariate Cox regression
- Aetiological research question: Estimation of Treatment effect (Hazard Ratio ) a. Unadjusted: Univariate Cox regression b. Adjusted: Multivariate Cox regression
Competing risks Present
- Prognostic research question (Calculation of Survival probability) a. Unadjusted: Cumulative incidence competing risks method (CICR) b. Adjusted: Sub-distribution Hazards (Fine & Gray) method
- Aetiological research question: Estimation of Treatment effect (Hazard Ratio ) a. Unadjusted: Univariate cause-specific proportional hazards model b. Adjusted: Multivariate cause-specific proportional hazards model • * 4a and 4b done using standard cox-regression, but with censoring of the competing event
As the treatment groups are non-randomised, I will be using IPTW (inverse probability of treatment weighting) to derive propensity scores and create a weighted, balanced dataset.
MY QUERY: I will be using the IPTW-weighted dataset for 4a and 4b above ( i.e etiological question) but shouldn't I be using the weighted dataset for the Survival probability (prognostic question) too? In other words, I could do a crude dataset K-M curve analysis, but isn't it better if done on the weighted dataset? Similarly, should I be doing a crude dataset Competing risks analysis or one where the IPTW weights are applied (weighted Fine and Grey)?
THANK YOU SO MUCH, WONDERFUL FRIENDS
P.s: Medical Doctor here, no statistical experience doing this myself :(.
