02.06.2013 Views

Aanesthetic Agents for Day Surgery - NIHR Health Technology ...

Aanesthetic Agents for Day Surgery - NIHR Health Technology ...

Aanesthetic Agents for Day Surgery - NIHR Health Technology ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Costs<br />

The total cost was calculated <strong>for</strong> each of the<br />

patients enrolled in the trial and allocated to one<br />

of the comparators. The total cost was the sum of<br />

all costs incurred on behalf of the patient, from<br />

the perspective of the NHS. The variable costs<br />

and costs associated with postdischarge resource<br />

use were reported separately. The mean cost per<br />

patient <strong>for</strong> each comparator was estimated as the<br />

sum of the total costs <strong>for</strong> all patients randomised<br />

to that intervention, divided by the number of<br />

patients randomised. The costs incurred by the<br />

patient, from the perspective of the patient or<br />

parent/guardian, were reported separately.<br />

The analysis excluded cost estimates <strong>for</strong> missing<br />

patients (patients who did not complete<br />

follow-up).<br />

Incremental cost-effectiveness ratios<br />

ICERs were calculated <strong>for</strong> the trial-based analyses<br />

and sensitivity analyses. The effectiveness measure<br />

<strong>for</strong> the calculation of the ICER was PONV (cost<br />

per case of PONV avoided). Variable costs were<br />

used in this analysis because the fixed costs<br />

component did not differ between randomisation<br />

arms. The interventions were ranked from highest<br />

to lowest cost. Interventions with high rankings on<br />

cost, which also have lower outcomes than the next<br />

most costly comparator, were treated as inefficient<br />

and excluded from further analysis. If the lowest<br />

cost intervention was also associated with better<br />

outcomes than more costly comparators, this was<br />

treated as efficient. Incremental ratios would not<br />

be calculated <strong>for</strong> this intervention, since its use<br />

would lead to both net savings and greater<br />

benefits than any other comparator.<br />

ICERs were calculated <strong>for</strong> the remaining<br />

interventions. Each intervention was compared to<br />

the comparator ranked immediately below it in<br />

terms of cost. The incremental ratios were<br />

calculated as:<br />

ICER = (Cost A – Cost B)/(Outcome A –<br />

Outcome B)<br />

Statistical analyses<br />

The SAS ® and SPSS ® statistical software packages<br />

were used. The objective of the statistical analysis<br />

was to test whether there were statistically significant<br />

differences between groups in the primary<br />

outcome (PONV), willingness-to-pay values,<br />

resource use and costs. Nominal data (e.g.<br />

incidence of PONV) were mainly analysed<br />

© Queen’s Printer and Controller of HMSO 2002. All rights reserved.<br />

<strong>Health</strong> <strong>Technology</strong> Assessment 2002; Vol. 6: No. 30<br />

using the χ 2 test. So-called ‘exact’ tests were<br />

used when appropriate. Regression analyses<br />

were employed to confirm the findings from<br />

tabular analyses and to explore fine detail<br />

and interrelations not easily studied<br />

through tabulation.<br />

Differences in mean values <strong>for</strong> continuous<br />

economic data were analysed using parametric<br />

tests of differences in mean values (length of stay,<br />

cost, CV and net benefit). Typically, these variables<br />

have positively skewed distributions. 253 The main<br />

options <strong>for</strong> statistical analysis were: standard<br />

non-parametric methods, data trans<strong>for</strong>mation,<br />

standard parametric methods, and nonparametric<br />

bootstrapping.<br />

Arithmetic means provide a measure of central<br />

tendency that incorporates the full distribution of<br />

observations. The arithmetic mean is considered to<br />

be the most relevant measure <strong>for</strong> healthcare policy<br />

decisions, which should be based on in<strong>for</strong>mation<br />

about the distribution of the costs of treating a<br />

patient group, as well as the average cost. The<br />

choice of statistical approach was based on the<br />

need to calculate and test <strong>for</strong> significant differences<br />

in the arithmetic mean in potentially<br />

skewed data.<br />

Non-parametric statistical tests were considered<br />

inappropriate because they do not test differences<br />

in arithmetic means. 253 Similarly, data trans<strong>for</strong>mation<br />

to achieve approximate normality does not<br />

result in a comparison of arithmetic means<br />

(e.g. geometric means are derived during<br />

log trans<strong>for</strong>mation).<br />

Bootstrapping compares arithmetic means,<br />

while avoiding distributional assumptions. This<br />

technique is most useful where the sample size<br />

is small to medium. Work carried out to compare<br />

the per<strong>for</strong>mance of bootstrapping with parametric<br />

t-tests has shown the t-test to be “remarkably robust<br />

to non-normality”. 253 This robustness requires the<br />

sample size to be large enough <strong>for</strong> the central<br />

limit theorem to act sufficiently, or <strong>for</strong> the sample<br />

size and skewness to be similar in the groups<br />

under comparison. 254<br />

The t-test has been shown to be robust and give<br />

similar results to non-parametric bootstrapping<br />

with sample sizes of 148, where there was “severe<br />

non-normality”. 253 The sample sizes in this study<br />

were much larger. It is suggested that in trials<br />

like this study, which are large enough to<br />

influence healthcare policy, standard t-test based<br />

approaches will be robust and give results very<br />

41

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!