You are on page 1of 2

HP425 Seminar 6: Censored/skewed costs

In this seminar, we will use a small simulated dataset by Chen et al (2015). Among
the patients in the data, 80 received the conventional therapy (group 0), and 80
received the new treatment (group 1). The first enrolled patient was followed for 69
months and the last for less than 1 month, with an average follow-up of 27 months.
Cost data were simulated with start and stop dates for each entry. Our objective in this
seminar is to estimate health care cost that are censored or/and skewed. You are
strongly advised to print out these instructions and bring them to the Seminar. You are
also strongly advised to bring printouts of the instructions for all previous Seminars,
as these will be helpful when you are asked to construct a Stata command on your
own (i.e. when the Stata command is not given in blue bold below).

1. Set up

• Double click on " HP425_Seminar6_dataset.dta" in Windows or else execute the


following command at the Stata command prompt:
use H:/Documents/HP425/Seminar6/HP425_Seminar6_dataset.dta", clear

• Create shortcut way of referring to file paths:

global home "H:/Documents/HP425/Seminar6"

• Open a log:
capture log close
log using "${home}/HP425_Seminar6_Log", replace

2. Descriptive analysis

• Browse the data: data description. Each patient has a unique id variable; a survival
time variable, surv; a death indicator variable, delta; a treatment variable, trt; and
one or more cost entries representing different types of costs or costs accumulated
at different time intervals defined by the start date variable, start, and the stop date
variable, stop.

sum
sum cost
sum cost if delta==1
hist cost

Question: Is it cost censored or skewed?

• Declare the data as survival data by stset, plot a survival curve and calculate the
mean survival time.

stset surv, failure(delta)


sts graph
stci,rmean
stci,emean

3. Non-parametric method to estimate the mean health care


costs

• Bang & Tsiatis method – hcost command (you might need to install it in Stata)

Question: Look at the hcost command and run a non-parametric model to compute
the cost mean for censored costs.

4. Parametric methods to estimate the mean health care costs


4.1. Carides – two-stage. Look at the slides from Lecture 8 to understand the
method and the regression needed.

4.2. Carides - Log transformation. Look at the slides from Lecture 8 to understand
the method and the regression needed

4.3. Box- Cox transformation. Look at the slides from Lecture 8 to understand the
method and the regression needed. The command is in Stata – boxcox. Use
several options (left-hand-side, right-hand-side, both-sides)

5. Close

capture log close

You might also like