MatchTime/TODOs.txt at main · RobinDenz1/MatchTime · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

match_time() methods
	- allow user-supplied models in methods "psm", "pgm", "dsm"
	- support things other than the cox model internally when using methods "psm", "pgm", "dsm"
	- maybe allow competing risks models in methods "pgm" and "dsm"
	- in method="pgm" / "dsm": allow fit of risk score model to be made on either all units or only on
	  untreated units as described in He (2020)
	- method="psm" / "pgm" / "dsm" fail when strata() is used in formula of cox model
	- method="brsm" man page: disentangle sequential stratification and actual brsm, add option "all" or similar to
	  ratio so that actual sequential stratification can be used

match_time() general
	- "d_longest" is calculated by ignoring inclusion criteria, an option to change this might be useful
	- i am not sure if the "hazard" calculation at t is correct for methods "psm", "pgm" and "dsm" (probably not)

merge_start_stop()
	- allow multiple event variables in event_times (probably best to allow a list of data.tables)
	- do competing risks work with "event_times" argument?
	- allow previous event counts to be generated as well
	- in the input checks, the start_types thing might be wrong (checking only the first?)

long2start_stop()
	- allow event variables to be specified

add_outcome()
	- maybe allow different types of outcomes (recurrent events, event counts, binary etc.)

New features:
	- add_covariate(): similar to add_outcome() but for covariates of different types
	- estimate_ipcw(): function to estimate inverse probability of censoring weights
	- implement clone-censor-re-weighting approach
	- implement various analysis functions
	- allow categorical treatment variables (big reworks + theoretical work needed)

General TODOs:
	- add vignette showcasing time-dependent matching with match_time()
	- in this vignette (or another vignette) give a table with supported funtionality
		- what treatments are allowed (timings, binary vs. categorical, ...)?
		- what outcomes are allowed per method (recurrent, competing events, ...)?

Theoretical:
	- categorical treatments?
	- what about treatments that do not stay at "treated" value once there?

Simulation:
	- will currently fail because a lot has changed in match_time() and related functions