Mr. Nelson set the tone of his comments in the first paragraph of his remarkable letter when he accuses me of ``apparent bias in singling out finasteride studies from other medical treatment for BPH''. The fact is that I did not even know the name ``terazosin'' at the time I did my finasteride study! He comes back to what the newspapers call the ``Proscar-Hytrin battle'' again in his paragraph 5, again claiming that I ``singled out finasteride data to analyze''. The accusation is is as reprehensible as it is wrong.
Mr. Nelson should be aware of the double blind study of finasteride and terazosin, reported by H. Lepor, which showed no difference betweenn finasteride and placebo, (1) substantiating and even enforcing my conclusion. A huge amount of time and money went into that study as compared with my study of the literature. The reader might appreciate Voltaire's remark on the expedition of Maupenius to Peru to confirm the bulging of the earth at the equator: ``You have confirmed in these tedious places what Newton found out without leaving his room''.
I now turn to Nelson's criticism of my using results published in the New England Journal of Medicine by R. S. Rittmaster (2). It is standard practice for researchers to use results published in peer-reviewed journals. He is addressing his complaint to the wrong journal.
I never claimed, from the earlier results, that the finasteride score improvement was not statistically slightly better than the placebo. I took all the statistical values at face value. I only pointed out that the improvement in score had to be the difference between the drug and placebo scores. It was 2.4 units for severe cases when the score itself varied between about 24 and 44. That is clearly a negligible difference. I then went further and examined the same effect in the study of E. Stoner et al. (3) which used a different score basis but showed the same small effects.
As to Mr. Nelson's reference to the three year study, I would like to emphasize that he does not mention that there was no placebo control after the first year. Thus one could not properly evaluate the changes in the following two years. (4) Table III in the Stoner study shows that 35% of the participants in the North American study and 57% in the International study dropped out at 12 months. What systematic errors are produced by this change in participants? One crucial number worth publishing would have been the relative variable results of the dropouts and nondropouts at month 12. Note that in most of the plots in Figs 1,3 and 4, a statistically significant drop occurs right after the 12 month points and then levels out for the rest of the study. The correlations between dropouts and non-dropouts are worth investigating. So, I did not misread the reported symptom scores as Nelson states. I chose to play safe and use the controlled study data. Nelson picks the result most favorable to finasteride, the uncontrolled study.
And then there is the problem of logical inconsistencies: Here is an example that I regret I did not stress and which I thank Mr. Nelson for giving me the opportunity to elucidate. It is argued by advocates of finasteride that the drug reduces prostate size and, as a consequence, there is to be improvement of flow and symptom score. The published data which I reproduced in my Fig. 4 (5) shows conclusively that there is no statistical difference between reduction of prostate size for the 1 mg and 5 mg doses at the end of the 12 month period. (At 3 and 6 months the 1 mg dose is even one standard deviation smaller!) But Figs. 2 and 3 show an improvement in both the symptom score and urinary flow rate that is statistically significant (2 standard deviations) for the two dosages, 5mg being more effective than 1mg. How can the flow and scores improve with dose when there is no difference in the prostate size reduction with dose? Simply, why should anyone care that finasteride shrinks the prostate if that is uncorrelated with flow improvement or symptom improvement? (I confess that Mr. Nelson is correct on one point. I did not make a study of the Federal VA pricing schedule. I simply multiplied the cost of a one month supply of PROSCAR from my pharmacist by 12, not a terribly bad approximation. The important question, unconsidered by Nelson, is the the incidence of side effects from the five times larger dose.) And is it not important to know, if there is a chance to become impotent, whether one will simultaneously grow breasts? The researcher might not care, but the patient and physician making treatment judgements do.
There were two main points in my paper. The first was the marginal effectiveness of the drug but, more important, was the unbelievable lack of attention in the medical literature to what is standard methodology in every science that handles data - the importance of publishing correlations among variables as well as their means.
Although I chose not to mention this in my paper, I was told by a key researcher at Merck that ``there were no observed correlations'' among the variables. Yet Merck claims that the drug shrinks the prostate which simultaneously increases the urinary flow and improves the symptoms. If that were all true it would be easy to determine whether they were indeed correlated. Since each patient was described by a triplet of points describing the improvement before and after the test on all three variables, the data existed for the determination of any of the standard correlation moments. When I offered to do the analyses I was told by Dr. Stoner of Merck (6) that the data was ``raw'' data and could not be released. I found this remark quite strange since it was the same data that was used in the author's studies. The data was also called ``proprietory''. Apparently Merck's knowledge that there were no correlations also could not be released.
Mr. Nelson claims that my plea for correlation studies ``simply is not well accepted in any of the BPH studies''. He is correct. A commentator on my paper once wrote ``By the current (italics mine) standards of reporting therapeutic studies, failure to report such evidence does not represent a deficiency of reported studies''. The commentator also added: `` Rather, this represents an unexplored aspect of the data whose utility and informativeness must be demonstrated.''
Because of these strange attitudes towards data analysis, not shared in other disciplines, I feel confortable about my attempt to bring medical studies (and BPH is only one) into line with established scientific appreciation for the importance of correlations.
In his last paragraph Mr. Nelson recommends taking drugs like terazosin and finasteride simultaneously even though they work on entirely different parts of the urinary system. This may be like taking chicken soup and an antibiotic at the same time. (The recent Lepor study claims that the mixture was no better than terazosin alone.)
Footnotes:
1. Proceedings of the American Urological Society, vol, 155 587a, May 1996 Supplement
3. Elizabeth Stoner, Urology 43 3 (1994)
4. The reader is advised to study the editorial comment of H. Lepor following that article.
5. G. J. Gormley et al. NEJM 327 17:1185-1191
Sherman Frankel