# Confidence Interval for CER, CI for NMB and acceptability curves

This page contains 3 sets of programs for estimating confidence intervals for cost-effectiveness ratios, confidence intervals for NMB, and acceptability curves.

One set of programs (iprogs.do) represent "immediate form" programs, in which you have already calculated measures of incremental cost and QALYs, standard errors of cost and QALYs, and the correlation between cost and QALYs. The programs use these data to calculate CI for CER, CI for NMB, and acceptability curves.

A second set of programs (uprogs.do) calculate CI for CER, CI for NMB, and acceptabilty curves based on univariate t-tests of cost and QALYs.

The third set of programs (bsceaprogs.do) use an already constructed sample of incremental cost and QALYs (e.g., 1000 replicates of the difference in mean cost and QALYs) to calculate CI for CER, CI for NMB, and acceptabilty curves.

We provide data from the 3 hypothetical experiments that we presented in chapters 8 and 9 as examples for these programs. For each experiment we provide a patient level dataset as well as a bootstrapped dataset.

Finally, we provide annotated log files for use with each of the three sets of programs.

IPROGS.DO: "Immediate form" programs to estimate Fieller's method confidence intervals for cost-effectiveness ratios (fielleri), confidence intervals for NMB (nmbi), and the acceptability curves (accepti). Also included is a program that defines the CI below which the lower limit is less than the upper limit; the CI between which the lower limit is greater than the upper limit; and the largest defined CI (ciboundi). Third, there is a program that uses univariate evaluation of means to calculate the inputs for these four immediate form programs (ipinputs). Finally, the program called iprogsdoc provides brief documentation for the programs. All programs are included in iprog.do. Three sample datasets, c8exper1.dta, c8exper2.dta, and c8exper3.dta are provided. As in chapters 8 and 9, these experiments provide examples of pattern 1, 2, and 3 results, respectively. Each dataset has 4 variables: id represents the participant ID number; the 0/1 treat represents treatment group; cost equals the patients' cost and qaly equals the patients' QALYs. Finally, a log file (iprogslog.pdf) is provided which demonstrates use of the programs. Programs particularly useful if one has estimated costs and outcomes by use of tests other than the t-tests.

**Data**

UPROGS.DO: Programs that use t-tests to estimate Fieller's method confidence intervals for cost-effectiveness ratios (fielleru), confidence intervals for NMB (nmbu), and the acceptability curves (acceptu). Also included is a program that defines the CI below which the lower limit is less than the upper limit; the CI between which the lower limit is greater than the upper limit; and the largest defined (ciboundu). Finally, uprogsdoc provides brief documentation of the programs. All programs are included in uprog.do. The 3 sample datasets above can be used with these programs as well. Finally, a log file (uprogslog.pdf) is provided which demonstrates use of the programs. These programs are useful if one wants to estimate costs and outcomes by use of t-tests.

BSCEAPROGS.DO: Programs to estimate confidence intervals for cost-effectiveness ratios (bscicer), confidence intervals for NMB (bsnmb), and the acceptability curve (bsaccept) from bootstrapped estimates of cost and outcome. bsceaprogsdoc provides brief documentation of the programs. All 4 programs are included in bsceaprogs.do. They presume you have already constructed the bootstrap sample. Three sample datasets, c8bsexp1.dta, c8bsexp.dta, and c8bsex3.dta, are provided. These datasets represent bootstrapped replicates of c8eper1 - c8cexper3, above. Readers should understand that unless one bootstraps an estimate in the 1000s of times, it is unlikely that the bootstrapped point estimates and standard errors will be precisely those we've described. They should also understand that we've manipulated our bootstrapped examples so that with 1000 replicates, they have estimates that are nearly identical to those we've reported in the chapter. Finally , a .log file (bsceaprogslog.pdf:) is provided which demonstrates use of bscicer. bsnmb and bsaccept.

**Data**

Last Update 12/02/2007