(no subject)
I hate it when the news media report only the relative change in risk, rather than the absolute change. For example:
The Vancouver Province places this text on its cover: “Cancer Risk: Children who use cellphones are five times more likely to develop malignant brain tumours, say researchers”.
First, no one should ever say “five times more likely” because it is ambiguous. They mean “five times as likely”. Literally, “five times more likely” means “six times as likely”.
But that’s incidental. My main objection is that it’s useless to convey the amount of the increased risk as a multiple of the original risk when they don’t tell you the original risk.
For example, suppose the research had found that among children who don’t use cell phones, the risk of developing a malignant brain tumour was 1 in 200, but among children who use cell phones the risk was 1 in 40. If that’s what the research had found, it would be extremely worrying and a very persuasive case to cease using cell phones, at least for children.
But on the other hand, suppose the research had found that among children who don't use cell phones, the risk of developing a malignant brain tumour was 1 in 2,000,000, and among children who do, the risk was 1 in 400,000. That result would not be a strong argument at all against cell-phone use: it would represent a very small increase in the risk of a malignant brain tumour (comparable to the risk of being injured in a short car trip, for example).
In both these examples, it would be accurate to say that the risk among children who use cell phones was five times as high as among children who don’t. But that’s not enough information to make any kind of rational decision. You need to know the absolute difference in the risk, not the relative difference. In these examples, the absolute increase in risk would be 0.02 in the first example and 0.000002 in the second example.
There was a great article on this topic in Scientific American: Knowing Your Chances: What Health Stats Really Mean. As the authors of that article put it: Absolute risks are more informative because they take into account information about background rates. Given the absolute risks, a person can derive the relative risks-but not vice versa. After all, a relative risk reduction of 50 percent could describe either a substantial mortality reduction from 200 to 100 in 10,000 patients or a much smaller one from two to one in 10,000 patients. Randomized trials provide some of the best information in medicine, but unless the results are reported adequately, people will not be able to assess them.
Moreover, when reporting on studies like this, without a randomized control group, the news media ought to caution against inferring a causative link.
Thus it appears that the studies linking cell phone use to brain tumours were conducted by interviewing people already diagnosed with brain tumours, asking them whether or not they had used cell phones as a child, and then interviewing a control group of people without a brain tumour, and asking the same question. Regardless of the result of this type of study, you could not conclude that cell phone use had caused an increase in the tumour risk. To prove causation, you would have to start out with a large number of children and then randomly choose some to use cell phones while the rest did not. Only then could you be reasonably sure that the cell phone use caused the increased tumour risk (rather than, for example, some other factor being responsible for both the cell phone use and the increased tumour risk).
Neither the Canwest News Service article nor an article from CTV BC mention the absolute risk. These articles refer to a paper in the journal Pathophysiology. The abstract of the paper is available here but I didn’t feel like paying $31.50 for the full text.
The Vancouver Province places this text on its cover: “Cancer Risk: Children who use cellphones are five times more likely to develop malignant brain tumours, say researchers”.
First, no one should ever say “five times more likely” because it is ambiguous. They mean “five times as likely”. Literally, “five times more likely” means “six times as likely”.
But that’s incidental. My main objection is that it’s useless to convey the amount of the increased risk as a multiple of the original risk when they don’t tell you the original risk.
For example, suppose the research had found that among children who don’t use cell phones, the risk of developing a malignant brain tumour was 1 in 200, but among children who use cell phones the risk was 1 in 40. If that’s what the research had found, it would be extremely worrying and a very persuasive case to cease using cell phones, at least for children.
But on the other hand, suppose the research had found that among children who don't use cell phones, the risk of developing a malignant brain tumour was 1 in 2,000,000, and among children who do, the risk was 1 in 400,000. That result would not be a strong argument at all against cell-phone use: it would represent a very small increase in the risk of a malignant brain tumour (comparable to the risk of being injured in a short car trip, for example).
In both these examples, it would be accurate to say that the risk among children who use cell phones was five times as high as among children who don’t. But that’s not enough information to make any kind of rational decision. You need to know the absolute difference in the risk, not the relative difference. In these examples, the absolute increase in risk would be 0.02 in the first example and 0.000002 in the second example.
There was a great article on this topic in Scientific American: Knowing Your Chances: What Health Stats Really Mean. As the authors of that article put it: Absolute risks are more informative because they take into account information about background rates. Given the absolute risks, a person can derive the relative risks-but not vice versa. After all, a relative risk reduction of 50 percent could describe either a substantial mortality reduction from 200 to 100 in 10,000 patients or a much smaller one from two to one in 10,000 patients. Randomized trials provide some of the best information in medicine, but unless the results are reported adequately, people will not be able to assess them.
Moreover, when reporting on studies like this, without a randomized control group, the news media ought to caution against inferring a causative link.
Thus it appears that the studies linking cell phone use to brain tumours were conducted by interviewing people already diagnosed with brain tumours, asking them whether or not they had used cell phones as a child, and then interviewing a control group of people without a brain tumour, and asking the same question. Regardless of the result of this type of study, you could not conclude that cell phone use had caused an increase in the tumour risk. To prove causation, you would have to start out with a large number of children and then randomly choose some to use cell phones while the rest did not. Only then could you be reasonably sure that the cell phone use caused the increased tumour risk (rather than, for example, some other factor being responsible for both the cell phone use and the increased tumour risk).
Neither the Canwest News Service article nor an article from CTV BC mention the absolute risk. These articles refer to a paper in the journal Pathophysiology. The abstract of the paper is available here but I didn’t feel like paying $31.50 for the full text.