Yesterday Wavylines produced a popular debate on a paper recently published in the journal Thyroid. This paper suggested a link between suppressed TSH in T4 treated thyroid cancer patients and an increased likelihood of later osteoporosis (but not AF). This is the comment of Prof Rudolf Hoermann, who is expert in this field and I quote:
But this paper really is unacceptable by general standards even if you would not subscribe to our findings on TSH.
1) On clinical grounds, it fails to categorise patients in euthyroid and hyperthyroid. TSH doesn’t do that according to any guideline.
2) On statistical grounds, it fails to justify the use of the Cox Proportional Hazard model.
If the risk is proportional, which needs to be tested and reported, the multivariable Cox model should use TSH as a continuous parameter rather than a category.
Either way they should first have tested for and established the nature of the relationship and then chosen the statistical model that applies.
Some of the argument is difficult and critical of the way the statistics are used, but the general tenor is: it is poor work that does not prove the point it is trying to show.