So, apologies again for the radio silence. Good news though: the PhD has finally been submitted! That’s not quite the end of that chapter in my life though, as I still have a viva to complete and six more publications to prepare to add to the two that have been published in the last 18 months or so. Hopefully I’ll be able to share more of my PhD outputs from the start of 2016 onwards, depending (of course) on the vagaries of the peer-review process.
Anyway, I now have time to read and then write about all of the publications that I’ve been putting to one side over the last few months. I’m going to start with a paper by Carlos Gallego et al. which was published in JCO in May, and which considered the cost-effectiveness of next generation sequencing (NGS) panels for the diagnosis of colorectal cancer and polyposis (CRCP) syndromes.
Colorectal cancer is interesting from the perspective of health economists working in genomics because, along with breast cancer, it feels like this is where the evidence that we produce is making the most difference at the moment. This paper used a fairly standard decision modelling approach to conduct a CEA and CUA which compared the costs and outcomes (life-years and QALYs) of different NGS panels with standard care in patients referred to US cancer genetics clinics. The main result highlighted by the authors was that evaluation with a NGS panel that included Lynch syndrome genes and other genes associated with highly penetrant CRCP syndromes was the most cost-effective strategy, with an ICER of $36,500 per QALY compared with standard care and a 99% probability that this panel was cost effective at a threshold of $100,000 per QALY.
This is quite a nice result, especially when you realise that standard care can involve a sequence of 3-4 genetic tests, hence the targeted NGS approach is also providing patients with a definitive result more quickly. However, it wasn’t the most interesting result in the paper, which was almost ignored by the authors. They also conducted a scenario analysis in which the NGS panel was used for universal testing on all patients diagnosed with colorectal cancer and found that this strategy also had an ICER below $100,000 ($70,600 to be precise). I was surprised that the authors didn’t make a bigger deal of this finding as it has implications for how we might wish to organise genetic and genomic testing in a variety of other disease areas where we have a reasonably good idea of which genes – in combination – have clinical utility (breast cancer being the example that immediately jumps out).
Putting that to one side, overall the paper is nicely written for the non-specialist and is perhaps a good entry point for health economists coming to genomics for the first time. As always, it’s good to reflect on the things that we could do better in health economics and genomics, and there were a couple of things that I wanted to highlight in this paper (hopefully in a constructive way). First, it’s often the case in genomics that we’re forced to select a small number of strategies to evaluate for our models from a wide array of possible comparators. This isn’t ideal, but we have to be pragmatic about providing an appropriate amount of information to facilitate decision-making. In this instance, I felt that more time could have been spent outlining the clinical utility of the different NGS panels (and, possibly, the lack of clinical utility of panels not considered) as this crucially drives the CEA and CUA results.
Second, the cost of the NGS panel was reported to be $2,700. I am struggling a bit with this cost. I know it’s the USA, and I guess it’s a marketed panel (going by the name) but in reality this is not a cost, it’s a price. I know targeted NGS doesn’t cost this much. It’s closer to 20% of that figure. What are the implications of this? Well, I’d be hesitant to assume that these results are transferable to another setting for starters, and, as variations in the cost of this panel do have an effect on the overall results, I’d be reporting these results with a little bit more emphasis on their uncertainty. Also, if the test really costs this much then if we’ve learnt anything from Oncotype DX it’s that there will be 10 replica tests on the market charging lower prices asap. This means that these results might not be applicable in 6/9/12 months’ time, but this isn’t really acknowledged anywhere.
My remaining quibbles are more minor. I’m not sure why Figure 2 reports the sensitivity analysis results for panel 4 and not panel 3, and I’d prefer to see CEACs for all the comparators in Figure 3, not just panel 3. Finally, the authors highlight limitations around the evidence informing a few key parameters. Why not just do EVPI/EVPPI and quantify this? But these are all relatively minor queries. The paper is worth a read, and if you only take one thing from the paper, note the cost-effectiveness of population NGS screening, because I suspect this will hit clinical practice sooner rather than later.