This blog post is a follow on from this post I wrote about cancer survival, and is about some really interesting quirks we encounter when we try to study it. These problems arise when we increase our ability to detect the cancer at earlier stages.
Lead time bias
The first of these is known as lead time bias. Because of better technology, we can now diagnose cancer earlier. So imagine a case where a new screening technique lets us catch a cancer 1 year earlier. The catch is that no matter how early we catch it, in this case the disease is going to progress and ultimately kill the patient.
So before the introduction of this new technology the patient might only survive 1 year after diagnosis, because we are catching it late. After the introduction, we can now catch the disease early, and the patient will survives 2 years after diagnosis.
If we were just looking at the numbers, we might think that we are making significant progress with this disease, because patients are now surviving for 2 years rather than just 1. However, we have done nothing to increase the patient’s lifespan. If we hadn’t caught it at all, the patient would have died on the same day.
This is known as lead time bias. Because we are catching a disease earlier, it can look like patients are surviving longer, when in fact they are not. It is an easy mistake to make, but a very important bias to consider when we are talking about cancer survival.
Length time bias
The second problem appears when you realize that if we detect a cancer early, it is possible that we are detecting a cancer that might never have progressed at all. It isn’t well-known, but there are cancers that never progress to a dangerous level. If we look at autopsies, nearly half of all men have prostate cancer when they die. However, only a small proportion of them actually die from prostate cancer. The rest have the cancer, but it will never progress. So the men are dying with prostate cancer, but not from prostate cancer. Most of these men will never have symptoms, so will never be diagnosed.
If a patient has symptoms, then it is quite likely that the cancer will progress if we don’t treat it. However, if we use a screening technology and catch the cancer before symptoms appear, then it is possible (likely) that some of those cancers were never going to progress to a dangerous level.
For example, before a screening technology is developed, survival from a particular cancer might be quite low, because we don’t detect the cancer until symptoms appear. Then we develop the new screening technology, and suddenly we are detecting all the cancers, regardless of whether they have symptoms or not.
If we were just looking at the numbers it would look like the incidence of that cancer is increasing (we now detect extra cancers that we wouldn’t have before), but it would also look like we are successfully treating these additional patients. Even though those cancers would never progress, patients would still get (un-needed) chemotherapy, and it would look like the treatment was successful.
The result of this is that we would think that the survival is increasing, but in reality, we are just identifying cancers who we wouldn’t have previously.
This is just two examples, but understanding this kind of bias helps us realise just how easily we can be fooled into thinking the wrong thing. This lets you look more critically at studies, and hopefully means we are less prone to bias when carrying out these kind of studies.