Thread by @RikEekelen

The main research project from my postdoc just got published in Human Reproduction: cost-effectiveness of medically assisted reproduction for unexplained subfertility. tinyurl.com/y36hkrbs (open access)
Time for a tweetorial on cost-effectiveness!
#epitwitter #statstwitter

Thinking of our health care in terms of finances can come across as cold and calculating. Unfortunately we do not have infinite amounts of medical staff and hospital beds. The harsh truth is that we DO have to make choices, for instance more effective or less expensive treatments

In practice, we often encounter the situation that a (new) treatment seems more effective but also more expensive than usual care or control. This constitutes a decision problem: is it ‘worth’ to set the more expensive treatment as the new standard of care?

To tackle this problem, we can conduct a cost-effectiveness analysis. Broadly speaking, there are two types of study designs: one in which individual patient data is available (often a RCT) and one in which data from other sources is combined in a new model.

In the first, the RCT provides data on the Tx effect and costs made by individuals during the trial. The differences between groups in average cost divided by Tx effect yields the incremental cost-effectiveness ratio (ICER): the amount we'd spend to gain one additional 'effect'.

The uncertainty around ICERs can be evaluated by bootstrap due to skewed costs distributions. Even after bootstrapping, the uncertainty is tricky to express. For instance, a simple 95%CI around the ICER can be misleading as a positive or negative ICER can mean multiple things.

The second design, combining different sources, was our approach. The question: for couples who did not conceive after 12 months of trying with idiopathic disease, do we treat them with insemination or IVF? Or can we treat later to save costs, as they might conceive naturally?

We assumed the (first) treatment decision would have to be made within 3 years after diagnosis with every period constituting 1 year. The ‘policies’ we considered were different orderings of the three treatment options: keep trying to conceive naturally, insemination and IVF.

For our model, we combined our dynamic prediction model for natural conception (i.e. keep trying) with relative effects of insemination/IVF from RCTs from a network meta-analysis. Costs were from hospital cost sheet data, the number of cycles were from literature.

One ‘run’ of the economic model does not represent the truth. All parameters such as probabilities, relative effects and costs had their own uncertainty. Overall uncertainty was expressed by drawing parameters from distributions and repeating the process 20,000 times.

As we considered 5 different policies, the decision problem was more complicated than my earlier example. If policy B is cost-effective compared to A, does the same hold comparing B to C? Many comparisons are possible. The solution is to describe results in terms of net benefit.

We express the health outcome (here live birth) as costs. If we deduct the costs of treatment from that, we are left with a net benefit. For example, let’s say a live birth is worth €1000. A policy for which 60% had a live birth after 3 years then ‘yields’ 1000*0.60 = €600.

Let's say the costs were €500, so net benefit is €600-€500=€100. So, which policy gains us the most? We calculate the proportion of each policy yielding the highest net benefit in all 20,000 replications for different monetary values of live birth and express this in a curve.

This curve is very informative: all policies are compared simultaneously without the need to choose references. The proportion is a quantification of the overall uncertainty and the range in monetary values represent different perspectives of ‘importance of the health outcome’.

That was my brief summary! This was fun, I hope I was clear. This type of modelling is not an exact science by any means but we try to quantify uncertainty as well as we can.
Many thanks to Madelon and @ReneEijkemans
PS the paper can be found at tinyurl.com/y36hkrbs