Sensitivity analysis
This section explores three sensitivity analyses. Two one-way
sensitivity analyses were conducted in order to test the uncertainty of
the modelling given our assumptions. A further probabilistic sensitivity
analysis was performed to test the model given uncertainty in the true
population parameters in the available sample.
Preliminary analysis demonstrated that the cost of perineal wound
breakdown treatment impacted the ICER most greatly. In the given model,
the estimate of this cost incorporated maternal readmission to hospital.
The proportion of cases that were referred to hospital was based on the
ANODE trial’s maternal readmission rate. However, wound breakdown can
also be treated in the community1. Therefore,
sensitivity analysis was conducted based on community treatment costs.
Assuming QOL remains consistent over a period of a year, this analysis
yielded a new ICER of -£3635.35/QALY, indicating cost savings of
£3635.35 for an extra QALY gained.
A second sensitivity analysis was conducted on the utility parameter for
patients that had perineal pain but no infection. In the original model,
the utility score assigned to this patient
group was 1. However due to the short time-period of 6-weeks, it is
possible that this group of patients also experienced a lower QoL.
Therefore, sensitivity analysis was conducted by attributing this
patient group the mean utility values provided in the trial of 0.935 for
the co-amoxiclav cohort and 0.927 for the placebo cohort. This yielded
an ICER of -£7203.33/QALY, corresponding to cost savings of £7203.33 per
extra QALY gained.
Under both sensitivity analyses, co-amoxiclav continued to be a dominant
activity (Figure 2) and all ICERs reported were negative. Overall, the
analysis showed that the intervention became less cost effective when
the cost of perineal wound breakdown varied but more cost effective when
the utility of patients with pain but no infection was reduced.
A third sensitivity analysis was conducted testing uncertainty of the
true population parameters of the model, given the sample provided by
the ANODE trial. Figure 3 demonstrates the results of a Monte Carlo
simulation, run 10,000 times, using the sample probabilities presented
in the decision-tree in Figure 1. The red-ellipse captures the 95%
confidence interval of the simulation results, where the 95% confidence
interval for the intervention ICER has lower-bound of -£6701.33/QALY and
an upper-bound ICER of £1874.36/QALY. Therefore, the intervention
remains cost-saving and QALY improving with a high-degree of
significance when presented with parameter uncertainty, and the
intervention is far below the NICE cost per QALY thresholds.