Critical health literacy, statistics, and treatment decisions
I was asked to choose between ACT chemotherapy and TC chemotherapy. Both were shown to be as effective, so the decision was mine. In making the decision, I looked at the potential side effects of each, as well as what the standard of care would be in Canada. Further, I looked at the literature and saw that there was more data on the ACT regime. I used that information together with knowledge of how my body reacts to things and my gut-feelings to decide on the ACT chemotherapy. The medical oncologist confirmed my decision by saying that 80% of women who need chemotherapy for hormone positive breast cancer get ACT chemo.
Now that I am much further out from the decision, I can see the problem with this statistic and some of the other data I used to make the decision. At the time, I felt comforted by knowing that this was a chemotherapy regime that was used in other places, and that it was common. If 80% of others were getting it, then it had to be the best choice right?
The problem is that the statistic is skewed by time. The longer a treatment is available, the higher the statistic will be. In addition, the longer a treatment option is available the more 5 year and long-term survival data there is (academic measures of treatment used in evidence-based medicine). Once a treatment becomes standard of care internationally, the number of people that get that treatment becomes much higher than other options. It means that newer treatments will always have lower numbers – at least until they are proven to be more effective and take over as standard of care. So, the percentage of other people who have had that treatment isn’t necessarily a useful number when it is used to make the treatment decision. Further, the older a treatment is, the more likely there is long-term survival data on that treatment. It doesn’t mean the treatment is better, it just means there is more data about it!
I also did not appreciate that looking to the standard of care decision was also not the best information for making my decision. It was comforting to know, but I didn’t appreciate that newer treatments take time to become the standard of care. So the newer treatment might actually be a better choice, but the statistics and standard of care data are not data that would support choosing the newer treatment options.
This is one place where the practice of “evidence-based medicine” can fall apart when it comes to decisions in care. The evidence will almost always suggest the older treatment options. There is a bias towards the older and better understood option. That is a nuance that was not appreciated by me, an academic, when I was in the position to have to make the treatment decision. It was not something that anyone pointed out to me.
I think this might also be an important factor when promoting clinical trials. In order to advance care, clinical trials are important, and often provide better treatment options – or at least that is their goal. However, patients can shy away from them in part because of a misunderstanding of the data, and not appreciating the biases associated with the current standard of care.
Another non-medical place where I see this bias is on YouTube – I have been using YouTube for years to provide how-to tutorials for technology. My most viewed video is an old one. But because it has been there for years, it has a much higher view count than newer and better versions of the same tutorial. The higher hit count of the older one causes more people to watch the older one, which in turn increases the hit count more. The newer one cannot catch up. There is a bias towards the older one because it has a history. It has been available longer and therefore has had more time to get hits. I see this bias all over the internet. Whenever an already known entity needs to compete with something new – the older known entity goes into the competition with a bias because of its history.
I think this is another way in which we can look at health literacy from a critical perspective – what are the biases in the data itself?