Source: Van de Weijer PHM, Barentsen R. Isoflavones from red clover (Promensil®) significantly reduce menopausal hot flush symptoms compared with placebo. Maturitas 2002;42:187-193.
Design: Randomized placebo controlled trial.
Subjects: Thirty menopausal women with more than 12 months of amenorrhea who were experiencing more than five hot flushes per day.
Treatment: Four-week placebo run-in followed by two tablets each morning of placebo or 40 mg Promensil for 12 weeks. Promensil is advertised as being manufactured from three varieties of red clover, using a standardized extraction and blending procedure to yield a proprietary ratio of daidzein, genistein, biochanin, and formononetin.
Outcome Measures: Hot flash counts and urinary isoflavones, not otherwise specified. "Overall menopausal symptoms," were measured by the Green score, a menopause symptom score.
Results: During the four-week placebo run-in, median hot flash number decreased from six per day to five per day. The median number of hot flashes continued to decrease among women randomized to Promensil by as much as 56%, but not among women randomized to remain on placebo. The proportion of women with hot flashes less than the median for the entire group was significantly lower in the Promensil group than in the placebo group. There was no difference between groups in the Greene score.
Funding: Novogen Ltd, Australia (the manufacturer of Promensil).
Comments by Anthony R. Scialli, MD: There is an old saying in the clinical trial biz: Statistics are like political prisoners; if you torture them long enough, they will tell you anything you want to hear. Nowhere is this adage more demonstrated than this effort on the part of these authors to show that Promensil works for hot flashes.
This randomized placebo-controlled double-blind study used an excellent design, including a four-week placebo run-in (important because menopausal hot flushes are believed to be placebo-responsive). This trial, then, had the potential to show whether Promensil is any good for hot flashes.
While the authors get an A for study design, they get an F for analysis. Make that an F-. The results of this study are completely obscured by statistical bamboozling and a refusal to give real numbers so the reader can tell what actually happened. The use of medians for number of hot flashes is appropriate because the number of hot flashes among women in both groups was not distributed normally. The median, however, is only an indication of central tendency of a distribution, and gives no information about the range of values that was obtained. When the authors report that the median number of hot flashes decreased in the Promensil group by 56% at week 10, they are carefully avoiding telling us how many women got better and how many women got worse. In fact, we cannot even tell what they mean by a 56% decrease: The median hot flash number at week 0 was said to be 6. A 56% in a median of 6 would be a median of 2.64. The median is the middle value in a distribution. As there are 15 subjects in this study, the median is the eighth number when you put all the numbers in ascending or descending order. In other words, the median in this sample must be a whole number and can’t be 2.64!
More subterfuge: The evaluation of improvement was based on the proportion of each group above and below the median for the total group. This method of analysis is tortured indeed. Why not simply use a nonparametric test on the number of hot flashes in each group? There are only two reasons why they did not use a straightforward statistical test: They know less about statistics than my dogs, or when the appropriate analysis was performed, it showed no statistical difference.
Although they withheld data that might have permitted us to understand what actually happened, one of the tables in the paper hints at the problems they must have had creating the appearance of effectiveness: The placebo group at the end of the study had a mean daily hot flush count of 6.04 ± 5.5. We understand that the mean is not very ¼ er, meaningful with a skewed distribution, but we can guess from this standard deviation that there must have been women on placebo without any hot flushes at week 12, as well as women with quite a few hot flushes. These pesky placebo-responders probably made it harder to show that Promensil is worthwhile, thus making trickery the only viable statistical technique.