Decision tree
Figure 1 shows the decision tree. The decision node splits the tree into
two arms:
- Arm 1: the new intervention of providing intravenous
co-amoxiclav.
- Arm 2 : the current standard practice of offering no
antibiotic prophylaxis after operative vaginal delivery; this is
represented by the data for the placebo cohort of the ANODE trial.
The assumptions made in constructing the disease pathway reflected in
the tree were necessary to clean the data and are as
follows7:
- A patient with no perineal pain would not experience infection
- Endometritis is mutually exclusive to perineal wound infection
- All patients with perineal wound infections would progress to develop
perineal wound breakdown.
The unit cost of resources are outlined in Table
18-15. An explanation of how results from the trial
were used to estimate resource use and calculate final terminal node
costs can be found in Appendix S1.
The data provided in the ANODE trial posed two issues with regards to
modelling.
- The ANODE trial reported outcomes as independent
events1.
- Not all patients that reported primary outcomes had their secondary
outcomes measured. This was 3420 patients and 2593 patients
respectively1.
Solution to Problem 1: Constructing the decision tree with
the numbers reported in the
ANODE trial would imply that the outcomes were independent
events1. However, the
likelihood of a clinical outcome such as wound breakdown occurring is
affected by other
outcomes such as whether the patient experienced perineal pain. For this
reason, using medical literature and an understanding of the healthcare
delivery process associated with postpartum infection, conditional
probabilities were calculated.
Solution to Problem 2: The decision tree has mutually
exclusive nodes that are combinations of primary and secondary outcomes.
The nominal reported numbers of patients experiencing secondary outcomes
could not be used to calculate conditional probabilities as they would
be an underrepresentation. For this reason, the number of individuals
experiencing secondary outcomes was standardised by scaling the number
up to represent the number of expected outcomes for a cohort of 3420
participants.