Subscribe to Blog via Email
February 2018 S M T W T F S « Oct 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
At the annual meetings of the American Thoracic Society a paper was presented that claimed a link between Sleep Apnea and the risk of getting Cancer. It gained national headlines with a claimed five fold risk dying from cancer for those suffering from the most sever forms of sleep apnea. Those with a moderate form of the sleep apnea had doubled the risk of dying from cancer. View the NY times blog on the paper here and the Fox News report here.
I contacted one of the authors for a copy of the paper that had been presented and was given a link where I can for a fee obtain a copy of the paper as is soon to be published in the American Journal of Respiratory and Critical Care Medicine. That is unacceptable. I am aware that journal need to be funded in some manner. But my position is that once the authors release the results of their findings to the public through the media full disclosure of the paper is required so that the users of the information, who are no longer just the doctors treating their patients, can properly consider the claims being made.
All is not lost because I was not provided with a copy of the paper. The analysis was based on data from the Wisconsin Sleep Study. The study design is well documented. In addition some of the same authors to the current paper published an earlier paper in 2008 using some of the same data. There they only linked sleep apnea to mortality. The Wisconsin Sleep Study has been following patients since 1988. Both the earlier paper and the current paper use much of the same data. The earlier paper involved followup for 18 years, while the current paper involved followup for 22 years.
The first red flag came when I looked that the abstract for the paper and found that the five fold increased risk of dying from cancer had a confidence interval from 1.7 to 13.2. This is a very large confidence interval around the actual estimated relative risk of 4.8. Such large confidence intervals are symptomatic of the use of very little data. Looking further, the 2008 paper showed that of the 145 people in the original study with moderate to severe sleep apnea nine had died from cancer by that time. The number likely has increased by a few cases in the succeeding four years since the first paper. My first take away it that this is too little data to draw any conclusions. The authors included things like age, sex, obesity, smoking, and a few other variables in their models. But I saw no mention that other obvious and important factors like family history for cancer and heart disease were considered.
The next comment I have is on the nature of the statical testing that was done and just how the variances were calculated. The approached used in the paper uses what is usually called classical statistical testing. This scenario, and the associated variance calculations, assumes that the researcher decided ahead of time what data they were going to collect. That also means they decided how many years into the study they would do the statistical testing for the association between sleep apnea and cancer. That was apparently not done in this case. Rather the 2008 paper was written when the association between mortality and sleep apnea first become statistically significant. I fully expect a third paper to be written in a few years when the authors can claim statistical significance between heart disease and sleep apnea. This type of analysis goes under the name sequential decision theory and requires very different calculation of the associated variances than was done for this paper.
Beyond this the major weakness in the paper is the data availability. As the study was constructed it will only be able to measure mortality risk as it relates to cancer and heart disease. If, as I expect, they produce a third paper linking heart disease to sleep apnea then the results will only say that those with sleep apnea have a higher mortality risk. To demonstrate that there is a specific risk for cancer they need to go beyond the available data in the Wisconsin Sleep Study.