This is a brief posting about biostatistics.
In clinical trials, some subjects drop out.
The quality of a study is best if there are few drop-outs, and if data continues to be collected on those who have dropped out.
LOCF and MMRM are two different statistical approaches to dealing with study populations where some of the subjects have dropped out.
One technique or the other may generate different conclusions, different numbers to present.
The following article illustrates how these techniques can skew the presentation of data, and therefore change our conclusions about an issue, despite nothing "dishonest" taking place:
http://thelastpsychiatrist.com/2009/06/its_not_a_lie_if_its_true.html#more
While I agree with the general point of the above article, I find that the specific example it refers to is not necessarily more biased: as I research the subject myself, I find that LOCF is not necessarily superior to MMRM, although LOCF is the most commonly used method to deal statistically with drop-outs. The following references make a case that MMRM is less biased than LOCF most of the time (although it should be known that whenever there are any drop-outs which are lost to follow-up, the absence of data on these subjects weakens the study results--it is important to consider this issue closely when reading a paper):
http://www.stat.tamu.edu/~carroll/talks/locfmmrm_jsm_2004_rjc.pdf
http://www3.interscience.wiley.com/journal/114177424/abstract?CRETRY=1&SRETRY=0
In conclusion, I can only encourage readers of studies to be more informed about statistics. And, if you are looking at a study which could change your treatment of an illness, then it is important to read the whole study, in detail, if possible (not just the abstract).
No comments:
Post a Comment