Page 175 - 15Diarrhoeaandvomiting

Diarrhoea and vomiting caused by gastroenteritis in children under 5 years

Figure A.4 One-way sensitivity analysis varying the cost of ORT

probability distribution for some or all model parameters. A Monte Carlo simulation is then run,
which involves running the model many times over, where probabilistic parameter values are
sampled randomly from their probability distribution on each run.
For the PSA undertaken for this paper, we restricted the probabilistic parameters to those that
*
were derived from the Cochrane review. In the deterministic analysis, a point estimate was
taken from the Cochrane review. However, such point estimates are always subject to inherent
sampling errors. This is the basis of inferential statistics and is at the heart of the hypothesis
tests used to test for differences and the calculations of confidence intervals. The probability
distribution for the model parameters acknowledges this sampling uncertainty while using the
point estimate as the ‘best guess’ of the true value. A ‘beta distribution’ was chosen for each of the
probabilistic parameters. This is similar to the normal distribution but is constrained to an interval
between 0 and 1, a necessary requirement for probability parameters. For this PSA, 100 Monte
Carlo simulations were run and the results are shown in Figure A.5.

In this analysis, the probability of ORT being cheaper than IVT was 100%.

Discussion

†
The baseline result shown in Table A.12 suggests that, when ‘downstream’ costs are considered,
ORT is £630.48 cheaper than IVT. Table A.13 shows that, in the ‘worst case’ sensitivity analysis,
ORT is £351.84 cheaper than IVT.
The model that has been developed is essentially a cost-minimisation analysis. The model assumes
that all patients rehydrate even if at some stage they are classified as treatment ‘failures’. Using
rehydration as the measure of outcome means that the treatment alternatives do not vary in terms of
their effectiveness and therefore the cheapest option is also unambiguously the most cost-effective.
Of course, while it may be a reasonable approximation to assume equivalent effectiveness
(and hence a reasonable model assumption), in practice there are differences between the two
treatments. Firstly, the meta-analysis undertaken for the Cochrane review was not powered to
detect rare adverse events. It may be that there are rare but clinically important harms that do
differ systematically between the two treatment alternatives. Secondly, the Cochrane review did
show a higher rate of treatment ‘failure’ for ORT. It seems likely that such treatment failure

* Other model parameters were held constant as in the deterministic baseline analysis.
† Costs which are incurred as a result of the treatment but subsequent to it, e.g. costs arising from treatment complications.

150