-
Notifications
You must be signed in to change notification settings - Fork 1
/
TODO
111 lines (70 loc) · 4.13 KB
/
TODO
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
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
----------------------------------------------------------------
ROLLING CRM:
- Not critical but worth to check - could be known limitation for PEMs:
Using an informative Gamma prior for lambda did not produce stable results.
How could this be done? would need proper documentation in help files/vignette
- Still todo before submission: C_par needs better documentation (name is OK)
- For `trueTmax`: add argument documentation here
- `DADesign` simulations: there could be errors if DLT occurs between the 2-3 patients,
when the 3rd patient has't been dosed yet.
- `simulations.Rmd`: Fix the problem with `factSurv` being negative and
then reenable simulation runs.
- Later: write a vignette (in Rmd format?) for rolling CRM, to make it easier
to start with this kind of design
- Later: Convert TITE CRM demo file to vignette in Rmd format (same as for rolling CRM
demo). Reason is that vignettes are routinely checked and easier to find
than demos.
- make nextOpen a part of the DAdesign object
- update method for `DataDA`: need to do like in `DataParts`, i.e. use underlying Data update
method first and then more changes according to `DataDA`.
Also we would expect arguments `x`, `y`, `u`, and `ID`.
- plot method for `DataDA`: insert example code in example file (still empty)
- update `SafetyWindowConst`with different slots
- `NextBestNCRM`: need an example here for `DataParts`.
- Document `safetyWindow` in `DADesign`.
- `prob`: simplify this to always vectorize internally over points so that we
always take/return matrix.
- Finish documenting `plotDualResponses`.
- `fitPEM`: Document `hazard` completely.
- `PseudoSimulations`: Document `stopReasons`.
- `DALogisticLogNormal`: Better would be to sample here also the interval probabilities
for having a DLT. This can also be nicely summarized later. We can then
put the additional parameter name in `sample`. Currently they are
calculated in the plot step.
Also explain why `l` is defined as is. Starting point:
Using the same assumption in Liu, Yin, Yuan 2013, it is assumed that a priori toxicity occurs
uniformly throughout the assessment period (0, Tmax), which represents a neutral prior opinion
between early-onset and late-onset toxicity. Under this assumption, the hazard at the middle of
the kth partition is caluclated as the `lambda_prior` function.
- `DADesign`: The `examine` method does not work. See the example script
`design-method-examine-DADesign.R`.
- Add demos, see design folder for one example.
----------------------------------------------------------------
BUGFIXES:
no current known bugs
----------------------------------------------------------------
IMPROVEMENTS:
- add warning if the MinimalInformative prior or similar return a prior variance
that is very low (to prevent undeliberately strong priors!)
- add a customized simulation summary function for the dual endpoint models,
which also reports the frequency of simulated trials selecting the correct OBD
- also optimize refDose for quantile->logistic functions
- make sure that it is clear in the documentation that the log transformation
only applies to the slope (beta) and not to the intercept (alpha)
- the minimalInformative function should already produce a graph comparing the
required and the resulting quantiles, instead of the user having to do it.
- record number of times the model is overruled in the simulations (Uli Beyer
request)
- mixture prior: add documentation in vignette
- allow different x values (doses) to be specified in simulate() to be passed to fit()
----------------------------------------------------------------
NEW FEATURES:
- add slide production with ReporteRs package examples (to demo folder?)
- be able to run simulations on a cluster: use the BatchJobs package that is installed
on the new Roche HPC - potentially as different fork of the package??
- production of Word output tables from simulation summaries
- minimal informative prior construction for other model(s), especially the probit model
to allow construction for the dual endpoint models.
- other prior in the Kadane model -> how?
- historical data prior,
pseudodata prior