Time-varying Covariate with Competing Risks

[Jacinto Aquino Pinto]

Hello,

I have to study the relationship between a drug exposure and outcomes, adjusting for other covariates.
For each subject the drug concentration was recorded once a month as a continuous value. This concentration may vary over time.
There are two possible competing Outcomes:

subject	Month#1	Month#2	Month#n	Outcome
1	160	240	240	0 (Censored)
2	360	340	360	1 (Event of Interest)
3	240	180	180	2 (Competing Event)

I tried the stcox solution (https://www.stata.com/support/faqs/s...ate-cox-model/) but Cox treat competing events as censored.
I would appreciate any sugestions.
Thanks in advance !

[Andrea Discacciati]

It's not clear from your post what quantities you want to estimate.

Using Cox (or other parametric regression models) to estimate cause-specific hazards/Hazard Ratios is perfectly ok. To do so, you treat the competing event times as censored. Time-varying covariates are not an issue in the interpretation of HRs (although landmarking is an alternative, see for example https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5711994/).

If you want to estimate Cumulative Incidence Functions, with time-fixed covariates you can "combine" results from cause-specific hazard regression models (see for example http://data.princeton.edu/pop509/justices.html) or you can model the sub-distribution hazard of the event of interest using Fine-Gray regression models. Note that with internal/endogenous time-varying covariates things become more complicated and you need to be extra careful if you want to predict CIFs/survival functions (eg, landmarking again) (see for example https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3702649/ https://academic.oup.com/biostatisti...9/4/765/259139 and references therein)

[Andrea Discacciati]

Relevant and just published: https://journals.sagepub.com/doi/10....62280218811837