Example: Replicating st0399 results

Lars Børty Nielsen, Martin Bøgsted, and Rasmus Brøndum

library(ccostr)

The ccostr package includes a dataset called “hcost”. The dataset comes from the Stata package of the same name (Chen, Rolfes, and Zhao 2015), and we can therefore use it both as an example and to verify that the two packages reach similar estimates. First we look at the data:

head(hcost)
#> # A tibble: 6 x 7
#>      id start  stop  cost   trt delta  surv
#>   <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1     1     1     1  3694     0     0   575
#> 2     1     1     9     1     0     0   575
#> 3     1     1     9    12     0     0   575
#> 4     1     1    34   106     0     0   575
#> 5     1     1   237    68     0     0   575
#> 6     1     1   237    86     0     0   575

Now we use the ccmean function with a limit of 4 years (1461 days), we also add the interpolation term which affects the interpolation of cost between two observed time points. By default this parameter is set to zero, but to replicate the results of the Stata implementation we set it to 1.

est <- ccmean(hcost, L = 1461, addInterPol = 1)
est
#> ccostr - Estimates of mean cost with censored data
#> 
#>   Observations Individuals FullyObserved Limits TotalTime MaxSurvival
#> N         9704         160            61   1461    122401        2082
#> 
#>    Estimate Variance       SE  0.95LCL   0.95UCL
#> AS 63725.42 19193502 4381.039 55138.59  72312.26
#> CC 74779.13 37572385 6129.632 62765.05  86793.21
#> BT 86175.16 51593885 7182.888 72096.70 100253.62
#> ZT 80134.84 23726332 4870.968 70587.74  89681.94
#> 
#> Mean survival time: 1165.04 With SE: 41.94

These estimates are similar to what is found with the hcost Stata package. We can also plot these with the included generic plot function:

plot(est)