Add arg alpha to gs_bound_summary() to report multiple efficacy bounds

[jdblischak]

Closes #522

xref: #468, #515

Still some open questions:

• Only supported for AHR design objects? (Generalize gs_bound_summary for supporting multiple alpha #522 (comment))
  • Confirmed. The initial implementation will only support AHR design objects. Support for WLR design objects will be added after gs_update_wlr() has been added (Add arg alpha to gs_bound_summary() to report multiple efficacy bounds #535 (comment))
• How to properly support design objects created by gs_power_ahr()? Currently I don't treat them any different that design objects created by gs_design_ahr() (Generalize gs_bound_summary for supporting multiple alpha #522 (comment))
  • Confirmed. It is ok to use x[["input"]][["alpha"]] for objects created by either function (Add arg alpha to gs_bound_summary() to report multiple efficacy bounds #535 (comment))

library("gsDesign2")

x <- gs_design_ahr(analysis_time = 1:3*12, alpha = 0.0125)
gs_bound_summary(x, alpha = c(0.025, 0.05))
##       Analysis               Value α=0.0125 α=0.025 α=0.05 Futility
## 1    IA 1: 31%                   Z   4.3506  3.8728 3.3437  -1.6192
## 2       N: 509         p (1-sided)   0.0000  0.0001 0.0004   0.9473
## 3  Events: 115        ~HR at bound   0.4449  0.4863 0.5366   1.3518
## 4    Month: 12    P(Cross) if HR=1   0.0000  0.0001 0.0004   0.0527
## 5              P(Cross) if AHR=0.8   0.0007  0.0032 0.0139   0.0032
## 6    IA 2: 74%                   Z   2.6778  2.3579 2.0022   1.1590
## 7       N: 611         p (1-sided)   0.0037  0.0092 0.0226   0.1232
## 8  Events: 278        ~HR at bound   0.7251  0.7535 0.7863   0.8701
## 9    Month: 24    P(Cross) if HR=1   0.0037  0.0092 0.0228   0.8768
## 10             P(Cross) if AHR=0.7   0.5336  0.6571 0.7770   0.0556
## 11       Final                   Z   2.2781  2.0096 1.7134   2.2775
## 12      N: 611         p (1-sided)   0.0114  0.0222 0.0433   0.0114
## 13 Events: 375        ~HR at bound   0.7903  0.8125 0.8378   0.7903
## 14   Month: 36    P(Cross) if HR=1   0.0122  0.0237 0.0445   0.9878
## 15             P(Cross) if AHR=0.7   0.9000  0.9252 0.9390   0.1001

[LittleBeannie]

Thank you, @jdblischak! Here, please find my answer to the questions above.

Only supported for AHR design objects?

Yes. If it is WLR design, some rownames are changed.

How to properly support design objects created by gs_power_ahr()? Currently I don't treat them any different that design objects created by gs_design_ahr()

The design objects created by gs_power_ahr and gs_design_ahr are gs_bound_summary in the same way. Their textual summary is different, but their table summary (as-gt table, or gs_bound_summary table) is the same.

[jdblischak]

Yes. If it is WLR design, some rownames are changed.

I was specifically asking about the new argument alpha described in #522. The rownames for WLR designs are already supported by my original PR #515.

Specifically, in #522, you describe how to use gs_update_ahr(). The function gs_update_wlr() does not exist, so I do not know how to implement the argument alpha for WLR designs.

[jdblischak]

The design objects created by gs_power_ahr and gs_design_ahr are gs_bound_summary in the same way. Their textual summary is different, but their table summary (as-gt table, or gs_bound_summary table) is the same.

Does that mean it is ok if gs_summary_bound() obtains the original alpha using x[["input"]][["alpha"]] for design objects created by either gs_design_ahr() or gs_power_ahr()?

[LittleBeannie]

Yes. If it is WLR design, some rownames are changed.

I was specifically asking about the new argument alpha described in #522. The rownames for WLR designs are already supported by my original PR #515.

Specifically, in #522, you describe how to use gs_update_ahr(). The function gs_update_wlr() does not exist, so I do not know how to implement the argument alpha for WLR designs.

Yes, you are correct that we need gs_update_wlr(). Its development is currently pending as I am working on the analytical derivation. For now, it is acceptable to have gs_bound_summary support multiple alpha levels for the ahr method only.

[LittleBeannie]

The design objects created by gs_power_ahr and gs_design_ahr are gs_bound_summary in the same way. Their textual summary is different, but their table summary (as-gt table, or gs_bound_summary table) is the same.

Does that mean it is ok if gs_summary_bound() obtains the original alpha using x[["input"]][["alpha"]] for design objects created by either gs_design_ahr() or gs_power_ahr()?

Yes

[jdblischak]

@LittleBeannie Thanks for answering my questions! I updated my original description above with these additional details. This PR is ready for review now

[LittleBeannie]

Thanks, @jdblischak! It looks good to me.