Sherman Frankel - June, 1996
Physics Department, University of Pennsylvania, Philadelphia
The importance of reporting correlations among side effects is examined using studies of BPH drugs as examples. The special relation of placebo effects in studying side-effects of drugs, as opposed to their efficacies, is introduced, and the need for special new controls is presented.
Key words: benign prostatic hyperplasia, Proscar, Hytrin, Cardura, placebo effect, correlation studies
INTRODUCTION
Recently, from an examination of the data on the drug finasteride (Proscar), (1) I pointed out that the placebo was almost as good as the drug in the treatment of Benign Prostate Hyperplasia (BPH). This conclusion has recently been supported in public announcements of the results of a comparison test between Proscar and Hytrin. (2)
More generally my study hoped to alert researchers to the need for publishing the full results of double-blind tests of drug and placebo. In particular it claimed that key correlation data accumulated in the finasteride studies on efficacy were available but were not reported, and that patients and physicians needed that information to make intelligent judgements on use.
In this paper I turn to another correlation subject, namely the correlations between side-effects. The study is applied to the drugs Proscar, Hytrin, and Cardura. (3)
However, another part of the study, which was stimulated by this simple investigation, is to elucidate the different role that placebo studies play in side-effect studies as compared with efficacy studies. I shall examine this matter first since it will allow the reader to understand why I have chosen to emphasize the drug side effect data independent of the placebo side effect data in my later discussion.
PLACEBO EFFECTS IN SIDE-EFFECT STUDIES
In double blind studies it is customary to gather data on the incidence of certain undesirable consequences (side-effects) for both drug and placebo. If the percentage of patients getting a particular consequence is statistically and systematically higher for the drug group than the placebo group, one says that the drug can produce a side effect. However the patient really wants to know whether taking the drug produces a side effect. These are not the same thing. It is well known that the ``placebo effect'' is the effect where the mere taking of a drug or placebo biases the answers to the question that is being asked. The name ``placebo effect'' is given when symptoms improve, whether one takes a placebo or whether the symptoms improve for a drug that actually has no ability to affect that symptom. It is also well known that the placebo effect is larger the stronger the symptom. The same ``psychological'' effects now come into play in the study of side-effects. If a patient knows what the plausible side effects may be, taking either drug or placebo will enhance the side effects probabilities. Therefore side effects should be treated differently than improvements in scores or other variables. One can imagine measuring side effects using a third group which takes neither drug nor placebo over the course of the double-blind study of drug and placebo. A comparison of the drug users with that response group measures a different quantity. That is the quantity which answers the question the user needs to know and which includes the placebo effect in the side effect data: Will I get undesirable effects if I take the drug? Drug companies are interested in another question, which is needed for FDA approval: Does the drug have more side effects than the placebo?
SIDE EFFECT STUDIES IN DOUBLE-BLIND STUDIES 0F PROSCAR, HYTRIN, AND CARDURA
With this in mind we examine the list of side effects that appear in the inserts which companies are required to place in drug packages, as well as data appearing in journals. The simplest question is: What is the chance of getting any one of the side-effects if the side effects are uncorrelated . One cannot address more relevant questions because the researchers and companies do not report the correlation data they have and, in many cases, believe that such data is useless. Nevertheless, this simple question throws some light on side-effect probabilities.
For example, in the insert accompanying Hytrin one can notice the relevant section, labelled ``Adverse Reactions''. This section gives results on adverse reactions to the body as a whole, the cardiovascular system, the digestive system, metabolic and nutritional disorders, the musculoskeletel system, the nervous system, the respiratory system, special senses (vision) and the urogenital system. The Hytrin insert reports the adverse reactions for a sample of users suffering from BHP and sample of users suffering from hypertension. The table of side effects gives the percentage of persons, p, getting a side effect. The chance of not getting that side effect is just (1- 100p). To calculate the chance of getting one of the side effects when there are many independent side effects, one multiplies all the chances of not getting the side effect together and subtracts that number from 1. We denote that measure as P.
Table 1 shows the result of this P calculation for studies of the drugs Proscar, Hytrin, and Cardura on samples of BPS sufferers.
Table 2 shows the results from the recent Lepor study comparing Proscar and Hytrin.
Table 3 shows the calculations for hypertension sufferers.
The first entry is the calculation for the drug. The second entry is the calculation for the placebo. The third entry (d-p) is the probability obtained by first subtracting the drug and placebo percents for each side effect. The fourth entry is the difference between the drug and placebo rows.
Since, except for the Lepor study, the number of side effects measured was different for the various studies, the most pertinent figure of merit is the probability per side effect. The number of side effects studied (n) is given in line 5. The probability, divided by n, for the drug or the drug minus placebo is given in lines 6 and 7 respectively.
The Proscar data was taken from Table IV in the paper by E. Stoner. (4) The Hytrin and Cardura data were taken from the drug inserts. The Lepor comparison between Proscar and Hytrin was taken from the Lepor data. (5)
Two main consequences of these calculations can be read from these tables:
Table 1 (line 1) shows that the chances of getting only one side effect is not small and at least one person in three would get a side effect if they were uncorrelated. Lines 6) and 7) show that there is little real difference between the side effect probabilities per side effect for the three drugs.
Table 2, which is a direct comparison of Proscar and Hytrin in the Lepor study, shows an appreciable larger probability for Hytrin than for Proscar, somewhat in contrast with the Table 1 results.
Table 3 suggests that, for sufferers from hypertension, the probabilities are larger and two thirds of the patients could expect one side effect from taking Proscar if they were not independent.
The reader will note that, in this comparison, none of the correlations are negligible.
Having illustrated the results for uncorrelated side-effects we now turn to correlation probabilities.
In the case of those taking the Hytrin for hypertension, of the 22 reported side effects described in the insert, the chances for a Hytrin user getting a particular effect varied from .3% to 19%. The 19% rating (for dizziness) means that about one out of about five users will report getting dizzy after taking the drug.
The next most reported side effect was headache (16.2%)
In the more recent Lepor comparison the percentages for dizziness were even larger for Hytrin, namely 25.9 %. Asthenia, (loss of strength), was 13.8 % Since the most frequent side effects are these two, one would want the drug company (and the researchers) to say what was the chance that one would also get weak if one got dizzy, a question that drivers would want to know.. There could be a 13.8% chance a patient would get both of them. Or is it four times smaller (3.6%), if they were not correlated? In the case of Cardura the most likely reported side effects were dizziness (19%), Headache (14%) and fatigue (15%) and no correlations are reported.
CONCLUSIONS
The correlation information comes from the same data bank as is used to report mean chances of side effects. It is not ``raw data''. We conclude from this that the FDA should require reporting correlation analyses of the side-effects of any drug that is allowed on the market, so patients and physicians are properly informed of the possible dangers in usage. Further, medical journals, for the same reason, should not publish papers when the correlations, which could provide a full picture of the conclusions from the experimental studies, are not included. While it is often not possible to give precise estimates of some of the uncertainties, even qualitative estimates of correlated effects are of value.
Finally, we argue that the relevant quantities that the patient, rather than the drug company, wishes to know, is the probability of getting a side effect from taking a drug as compared with taking nothing, rather than the difference between drug use and placebo use. This would require somewhat more expenditures of funds for the tests. Public interest groups, health care organizations or governments might supply such funds if the medical profession supported it.
Footnotes:
1. Analyzing Finasteride Data, Neurourology and Urodynamics 14:619-624 (1995)
2. The study was supervised by H. Lepor, New York University. See the New York Times, August 25, 1995. See also H. Lepor, Proceedings of the American Urological Society, vol. 155 587a, May 1996 Supplement
3. I use the brand names because there has been much controversy among the companies, Merck, Abbott, and Pfiser, that both make and study them and lay readers know these names well.
4. E. Stoner, Urology 43 3 (284-290) 1994.
5. loc. cit.