Merge pull request #53 from Merck/cranprep

Edits to enable win-builder pass

NEWS.md

@@ -1,17 +1,16 @@
-# psm3mkv 0.3.2
-
-# psm3mkv 0.3.2
+# psm3mkv 0.3.2 (7 June 2024)
 
 - Revised calculations of constrained restricted mean durations in internal function `calc_drmd()` and the accompanying `vignette("background-mortality")`.
+- Add new data option: `create_dummydata("pharmaonc")` provides a dataset based on `pharmaverseadam::adsl()` and `pharmaverseadam::adrs_onco()`. This requires a dependency to `admiral` and `pharmaverseadam` packages.
 - Provide a convenience function, `compare_psm_likes()`, to compare the total log-likelihood values for the patient-level dataset after fitting PSM-simple and PSM-complex models to each combination of endpoint distributions.
 - Add citation to publication in *Applied Health Economics and Health Policy*, [DOI: 10.1007/s40258-024-00884-2 ](https://doi.org/10.1007/s40258-024-00884-2).
 - Updated license statements to 2024
 
-# psm3mkv 0.3.1
+# psm3mkv 0.3.1 (7 May 2024)
 
 - Submission to CRAN, including changes requested by CRAN
 
-# psm3mkv 0.3.0
+# psm3mkv 0.3.0 (5 May 2024)
 
 - First submission to CRAN, not accepted
 

R/likepsm.R

@@ -30,16 +30,16 @@
 #' @param fitslist List of distribution fits to relevant endpoints, after calling `fit_ends_mods_par()` or `fit_ends_mods_spl()`
 #' @importFrom rlang .data
 #' @return List containing
-#' - `res`: Dataset of calculation results for each model
-#' - `best`: Tibble indicating which is the best fitting model individually or jointly, to each endpoint, according to AIC or BIC
+#' - `results`: Dataset of calculation results for each model
+#' - `bests`: Tibble indicating which is the best fitting model individually or jointly, to each endpoint, according to AIC or BIC
 #' @export
 #' @examples
 #' # Fit parametric distributions to a dataset
 #' bosonc <- create_dummydata("flexbosms")
 #' parfits <- fit_ends_mods_par(bosonc)
+#' \donttest{
 #' splfits <- fit_ends_mods_spl(bosonc)
 #' # Present comparison of likelihood calculations
-#' \donttest{
 #' compare_psm_likes(bosonc, parfits)
 #' compare_psm_likes(bosonc, splfits)
 #' }

vignettes/example.Rmd

@@ -53,7 +53,7 @@ Once we have a suitable dataset, we will fit statistical models to these six end
 
 ### Parametric distributions
 
-Let us start by considering parametric distributions. This uses the function `fit_ends_mods_par()`, so called because it cycles through fitting endpoints and models. The original dataset contained only three of these endpoints, the other three endpoints are calculated within the function.
+Let us start by considering parametric distributions. This uses the function `fit_ends_mods()`, so called because it cycles through fitting endpoints and models. The original dataset contained only three of these endpoints, the other three endpoints are calculated within the function.
 
 ```{r fit_ends_mods_par}
 # Create a vector of distributions of interest (flexsurv notation)

vignettes/mortality-adjustments.Rmd

@@ -260,9 +260,9 @@ res3 |> mutate(
 proppf3 <- res3$pf / res3$os
 ```
 
-There is now a large difference between the STM-CR compared to the STM-CF and PSM structures. All structures estimate fairly similar RMD estimates of being alive (range: `r round(min(res3$os),1)` to `r round(max(res3$os),1)` weeks). The estimate of mean time alive in the STM-CR structure has reduced by `r round(res1$os[3]-res3$os[3],1)` weeks from `r round(res1$os[3],1)` to `r round(res3$os[3],1)` weeks.
+There is now some difference between the STM-CR compared to the STM-CF and PSM structures. All structures estimate fairly similar RMD estimates of being alive (range: `r round(min(res3$os),1)` to `r round(max(res3$os),1)` weeks). The estimate of mean time alive in the STM-CR structure has reduced by `r round(res1$os[3]-res3$os[3],1)` weeks from `r round(res1$os[3],1)` to `r round(res3$os[3],1)` weeks.
 
-The proportion of that time in the PF rather than PD state varies considerably (PSM: `r round(100*proppf3[1], 1)`%, STM-CF: `r round(100*proppf3[2], 1)`%, STM-CR: `r round(100*proppf3[3], 1)`%). Given that the hazard of mortality would typically be lower before rather than after progression, the pre-progression mortality is likely to be more affected by any mortality constraint that post-progression mortality. The STM-CR model of post-progression survival depends on time from progression and is therefore rather different from the STM-CF and PSM structures, which model mortality based on time from baseline.
+The proportion of that time in the PF rather than PD state varies somewhat (PSM: `r round(100*proppf3[1], 1)`%, STM-CF: `r round(100*proppf3[2], 1)`%, STM-CR: `r round(100*proppf3[3], 1)`%). Given that the hazard of mortality would typically be lower before rather than after progression, the pre-progression mortality is likely to be more affected by any mortality constraint that post-progression mortality. The STM-CR model of post-progression survival depends on time from progression and is therefore rather different from the STM-CF and PSM structures, which model mortality based on time from baseline.
 
 We can check whether using a 'complex' type of PSM structure rather than the default 'simple' affects RMD estimates from the PSM structure.
 
@@ -281,7 +281,7 @@ res4 |> mutate(
 proppf4 <- res4$pf / res4$os
 ```
 
-All structures estimate fairly similar RMD estimates of being alive (range: `r round(min(res4$os),1)` to `r round(max(res4$os),1)` weeks). The proportion of that time in the PF rather than PD state varied again considerably by model structure (PSM: `r round(100*proppf4[1], 1)`%, STM-CF: `r round(100*proppf4[2], 1)`%, STM-CR: `r round(100*proppf4[3], 1)`%).
+All structures estimate fairly similar RMD estimates of being alive (range: `r round(min(res4$os),1)` to `r round(max(res4$os),1)` weeks). The proportion of that time in the PF rather than PD state varied a little by model structure (PSM: `r round(100*proppf4[1], 1)`%, STM-CF: `r round(100*proppf4[2], 1)`%, STM-CR: `r round(100*proppf4[3], 1)`%).
 
 ### Comparing the results
 
@@ -316,9 +316,8 @@ A summary of the PSM estimates of mean time in PF state and mean time alive is g
 
 The change in the RMDs for the 'complex' PSM structure compared to the 'simple' PSM structure were `r round(res4$pf[1]-res3$pf[1], 1)`, `r round(res4$pd[1]-res3$pd[1], 1)` and `r round(res4$os[1]-res3$os[1], 1)` weeks respectively for time in PF state, PD state and time alive.
 
-### For further investigation
+### Possible extensions
 
-- The effect of the lifetable constraint on the STMs was much greater - especially to time in the PF state - in the STMs rather than the PSM.
-- The package also allows application of discounting through the `discrate` optional call to `calc_allrnds()`.
+Not applied here, the package also allows application of discounting through the `discrate` optional call to `calc_allrnds()`. This work could also be easily extended through calculating RMDs or Life Years that are Quality Adjusted (QALYs).
 
 ## References