Sunday, December 20, 2009

Don't let the median scare you to death

Further to Friday's post Advice to a Cancer Patient Facing News He Didn't Want:

One powerful factor in my case was the education I got about not putting too much meaning on the probabilities I read. They may be accurate but they're not useful to focus on, except as a motivator to get it in gear.

Here's more background on that, for people with varying appetites for math. Note: This isn't anti-science blather; it's scientifically valid. But often clinicians are taught not to give people "false hope" i.e. that they shouldn't get your hopes up "unrealistically." I understand that - it must be horrible to face a bereaved family, crying "But you told us there was hope." But frightening odds don't mean there's no hope; in fact it's unrealistic to think there's no hope.

The first thing to understand is "median survival time." The classic article about this is "The Median Isn't the Message," by famed Harvard author Stephen Jay Gould. The median is the statistic that's most often published about cancer survival, but it's not the whole story. Here's why.

The median is the middle number in a study. It tells you nothing about the other numbers. The others could be the same or vary vastly.

Example: let's say a study starts with 25 people who have a given condition. When the middle one (#13) dies, that's the median. Did everyone else die the same day? Can't tell. Did the others get completely better? Can't tell. You can't tell your personal odds by looking at the median.

Why do scientists publish the median? Because that's the first statistic they finish. They won't know the average survival until the last person dies, which may be never. And they'll never know yours.

As Gould says, "the median isn't the message."

And here's where the rubber meets the road: when you're in a crisis, the most important flaw of medians is that the median gives you no useful information about what to do.

So yes, the median can tell you how bad it might be, and that may motivate you to get your butt in gear to improve your odds. But it's not the whole story. Don't let it scare you to death.

p.s Another copy of that article was posted on the CancerGuide web site, which was handed coded by one of the most senior of all e-patients, Steve Dunn - a kidney cancer patient who founded the "KIDNEY-ONC" kidney cancer community that I used so heavily, on

In his "Cancer Guide" Steve assembled a phenomenal set of empowering teaching tools about statistics. They're listed in the side menu on that ugly old website :-). If you want, read it.


  1. Dave- this is a difficult (mathematical) topic. First of all, most medical statistics are interpreted incorrectly. I was just at a conference where they presented a distribution by the mean and standard deviation, however 100% of the data was less than one standard deviation above the mean. If data isn't normally distributed, standard deviation is meaningless. Let's take mean. If a disease kills 90% of those affected instantly but 10% live 50 years more, the mean survival time is 5 years. Does that tell you anything? Further, studies on survival include death by causes other than the disease. In Parkinson's disease, the median survival after diagnosis is 12 years, but 50% die from causes other than PD. In diseases that predominantly affect the elderly, it is important to consider the median age at diagnosis. One of the simplest ways to get a sense of a distribution is to see both the median and the mean. A difference between the two shows a skew in the distribution.

    My advice for anyone who is told a single statistic about a disease is to recognize that you've been told nothing. Any physician who doesn't know a disease well would be best off not trying to characterize it on the basis of a single metric. To understand survivability, you need to understand not just the survivorship curve of a disease but also the profile of typical patient on diagnosis (not just their age and general health but also the stage of their disease) and how the current patient deviates from that profile, and the therapies of the typical patient versus the therapies available to your current patient. If your doc doesn't tell you these things, ask; if she doesn't know, she ought to have been more careful about telling you the info in the first place.

  2. PNSchmidt,

    To amplify this - in my particular case, Stage IV kidney cancer, most people who are diagnosed at that stage are indeed in big trouble. I found a short questionnaire online that asked questions like "have you been losing your appetite," which graded by condition. From that answer (I think it was the Karnofsky Performance Scale) I looked up the median time.

    Only later, through my ACOR patient community, did I learn that those numbers were obsolete - the data was all collected in the early 1990s.

    That's just another example of why the "best available" information may not be USEFUL information. It still comes down to "Okay, what are my options?" Knowing the odds aren't good is just a motivator.


Your comments will be posted after they are reviewed by the moderator.