Thursday, January 22, 2015

Solving the Mystery of Swedish Mammography Trials

Kenneth Branagh as Kurt Wallander
I felt like Kurt Wallander on the trail of a suspect in a small Swedish town with the murder rate of Cabot Cove.  The quarry was causality in the Swedish Two County randomized controlled trial.  

The trial, conducted some thirty years ago, randomized women in two Swedish counties (thus the name) to receive either an invitation to receive a mammogram or no invitation.  These invitations were given over the next several years and the trial participants were followed for the next twenty plus years.

The results were that 339 women of 77,080 passed away from breast cancer in the invitation group and 339 women of 55,985 passed away in the control group.  The claim is that this result proves that general mammography screening reduces death from breast cancer.  The issue is that this doesn't tell us that mammography reduces death from breast cancer.  

This is an intent-to-treat analysis.  

We know that these women were randomized into two groups and we know the average death rate from breast cancer in the two groups, but we don't know how many people actually got a mammogram in each of the two groups.  It was this number that I was searching for.  How many people actually got mammograms?

After trudging through the many papers written on the study I found one of the two numbers I needed.  Approximately, 85% of women in the invitation group accepted the invitation and received a mammography.  As far as I can tell the researchers never determined the take up rate of mammography in the group that didn't receive mammography.

So what can we say from this trial?

As long as the mammography rate in the invitation group was higher than in the control group then the analysis implies that mammography causes a reduction in death rates from breast cancer.  If we are willing to make a behavioral assumption - that Swedish women are motivated to get mammography by receiving invitations, then this large randomized controlled trial can be used to infer causality.  Our prime suspect must have had an accomplice.

We can't really say much more about the value of mammography without knowing how many women in the control group received a mammogram. 

No comments:

Post a Comment