Scores and Score Reductions vs. Severity
Sherman Frankel
Physics Department, University of Pennsylvania
33rd and Walnut Streets, Philadelphia PA 19104
printed Mon Aug 28 16:29:56 EDT 1995
Click on any figure to view at full size
Abstract
Published data on the safety and efficacy of finasteride in the treatment of benign prostatic hyperplasia are studied and compared with published conclusions. The need for publishing correlation studies of both the benefits and side-effects, to allow for informed physician and patient decisions, is detailed.
I. Introduction
Finasteride(PROSCAR) is presently being widely advised as a drug to treat benign prostatic hyperplasia (BPH) and physicians appear to be recommending it to patients, perhaps without a sufficient understanding of its merits. Studies of the effects of finasteride have been carried out which trace the effects on several variables, urinary flow, prostate volume, and symptom scores being among them. Some summaries of the trials have been published but not the basic data relevant to clear clinical decisions. In this note we comment both on the efficacy studies and the lack of available relevant information.
II. Examination of Relevant Data
Of immediate interest is the data published in an informative paper by R. S. Rittmaster [1]. That paper describes the results of a large study of patients who were given either a placebo or 5mg of finasteride per day over a period of a year. The data reported dealt solely with the effect on the symptom score. Fig. 1 shows both a reproduction of the change in symptom score and the superposition, on the same scale, of the range in scores appropriate to mild, moderate, and severe ratings of benign prostate hyperplasia (BPH), both taken from the Rittmaster paper.
Scores and Score Reductions vs. Severity
The first point to be extracted from the data is that the shape of the increase in the change in score vs rating curves is almost identical for the placebo and finasteride, and further, that the more severe the symptoms, the more effective both drug and placebo. This suggests the conclusion that the decrease in score for both groups may have a dominant psychological rather than a chemical basis. A second point is that the difference in the effect of placebo and drug is independent of the severity. The third point is that the improvement relative to the score is almost negligible. A decrease of 2-3 units in a mid-range score of about 32 for severe symptoms, after a year of treatment, can be read from the data. The vagueness of such scoring tests and their repeatability over short time intervals suggests that there is easily an uncertainty of a few units in the score itself so that a few units of bias could easily be introduced by patient expectations. Since the improvement in the placebo group averages is 4 units compared with about 6.4 units in the finasteride group, it might seem that the placebo would be favored over the drug since it is almost as effective, has zero side effects, is inexpensive, and can be obtained without a doctor's prescription.
These few data suggested an examination of the detailed results reported by E. Stoner et al.[2] which summarizes two independent studies of several variables, the North American (NA) study and the International (I) study. A wealth of data was accumulated for maximum urinary flow, prostate volume, and symptom score. However, all that is reported from the study are means or medians of the data summed over the participants. No distributions, correlated distributions or n-dimensional correlations are reported from the rich data accumulated in the studies.
We first examine the comparisons of reported results of the studies with placebo and drug which took place over a year:
Symptom Score: The mean symptom score, averaged over the two studies which have very similar results, is 19 for the placebo (P) group and 19.4 for the finasteride (F) group. But the change in score for the F group was at most 3 units and for the P group 2 units, the difference being 1 unit. Thus the more detailed study agrees with summary data given in the Rittmaster paper. The improvement in score appears to be minor. Although both studies appear to show a small difference between the P and F groups, the difference among the two studies leaves open the possibility that systematic differences between the P and F groups could account for the score difference. The data shows further puzzling effects. The symptom score for the F sample does not change from the second to the ninth month of treatment and then drops suddenly. However the score for the placebo rises about the same amount over the year.
Urinary Flow: The mean maximum urinary flow was measured and was 9.0 for the P groups and 9.2 for the F groups, again using the NA-I average. At the year's end the average of the two studies showed a P-F difference of 1 unit, so the fractional improvement in flow rate was small. (The error bars are not defined in the paper but the uncertainties appear to be about +/- 0.5 units. The reproducibility of the flow measurement is not reported.)
Prostate Volume: Studies of the shrinkage of the prostate volume, based on MRI studies, show about a 16% reduction averaged over the two studies, for the difference between the P and F groups. Once again, as in the case of symptom score, there is an unexplained decrease in volume for the placebo group of about 5%, but this cannot be a psychological effect as in the case of the score reduction. What is unexplained is the fact that about 20% of the P group show greater than a 20% decrease in volume. While this decrease is indeed higher for the F group (40%), these facts leave open the possibility that there is something being missed in the prostate shrinkage study.
To the data analyst it is also puzzling why the functional form of the monthly change in both the symptom score and urinary flow are the same for drug and placebo.
There are also interesting data that appeared earlier by the same study group. See G. J. Gormley et al.[3] Several figures from that study are reproduced for the following discussion. The symptom scores (Fig. 2), show the same improvement (1.25 units) for the placebo group, 1 mg finasteride group, and 5 mg finasteride group at the end of the first month. They separate by only .9 units at the end of the second. How can this be explained? Is there a reason for the sudden separation over a one month period with no further improvement until about the 9th month? After that period there is a sudden separation of the 5 mg sample alone but no further change for the 1 mg group. What is the effect of dropout near the end of the year?
The urinary flow data (Fig 3) again show almost no variation with time until about the 10th month when there is some further separation of the rates.
Changes in the serum dihydrotestosterone and serum prostate-specific antigen(PSA) (Gormley et al., Fig. 1) and the prostate volume (Fig. 4) show clear differences between placebo and 5 mg samples. However, what is most striking is that there is no difference at all between the the 1 and 5 mg samples in all these variables. No data were reported to determine whether another reduction of the dose by a factor of five would give the same results as 1 mg.
Rittmaster (loc. cit.) has pointed out that ``the availability of finasteride creates a substantial problem for the health care system''. In view of the fact that there are more than 8 million males in the USA who are of the age to show prostate size increase, many of whom are covered by Medicare, Medicaid, and other Health Care insurance, use of the high 5mg dose with higher cost (about $800/year) to control this variable is not justified by these data.
III. Correlations Among Variables and Side Effects
Extensive data have been accumulated on the efficacy and safety of finasteride. Unfortunately all that is reported are averages over all study participants. What we would like to emphasize is that there are standard data analysis techniques which are not pursued in the report of the data and which are crucial to understanding their significance. The physician and patient need to know more than the averages to come to sensible conclusions about treatment.
Finasteride is argued to simultaneously improve the symptom scores, urinary flow, and decrease the prostate volume. An important question is whether they are indeed correlated. (Commercial advertisements by the study sponsors suggest that they are.) This question can be handled in the following way: For each participant consider three variables, prostate size reduction, flow rate increase, and symptom score. Let x denote the change in prostate size from the beginning to the end of the study and let y and z denote the change for flow rate and symptom score respectively. A three dimensional scatter plot of these changes (each participants x,y,z value being a point on the plot) would show whether or not the points tend to cluster in some region. If the points were clustered one could determine how many of the variables improved as the result of taking the drug. Moreover one could determine if more than one variable improved simultaneously. (Or if one improved while the other got worse.)
Perhaps simpler to understand would be plots of what are called ``conditional distributions'': For example, one would plot the change in flow rate for the group that shows a large prostate size reduction and also for the group that shows no prostate size reduction.
One could also plot the distributions in one variable. One could tell then whether the average improvement came about because very few participants had high improvement.
Such correlation studies may be more important in safety studies: It is claimed in the paper that the safety profile is excellent but once again no correlations are reported. For example, decreased libido, decreased ejaculation volume, impotence, orgasmic dysfunction, and breast enlargement may show strong correlations which are the important quantities for decision making. One might risk one side effect but be wary of taking the drug if the downside was that one would get most or all of the side effects, even if the chance of side effects was not large.
As an illustration, suppose a drug is reported to have 10 possible independent side effects, each with an occurrence probability of 1%. The probability of having at least one side effect will be almost 10% (Actually 9.56%). The probability of getting all the side effects is entirely negligible. But with perfect correlation the probability is not 1% that one will get one of the side effects but rather that one will get all of the side effects 1% of the time. Knowing the correlations is a necessary input to any ``figure of merit'' for decision making on drug use.
Jens E. Altwein [4] summarizes the Seville Conference on BPH with the remark: ``The urologist must ... explain the relative harm and benefits of each intervention''. The publication of existing data and reporting of correlations might aid in this process.
IV. Conclusions
From examination of the reported data, the efficacy and safety of finasteride are open to some question. The need for reporting the basic available data, or at least reporting correlations between the tested variables, presenting conditional distributions or even simple distributions, is emphasized. Reporting of such basic data for any drug evaluation which has many measurable variables should be a pre-requisite for publication of study results in a scientific journal. Without them the physician or educated patient cannot easily make an informed judgement on possible use.
Acknowledgements: The author wishes to thank Drs. S. Day, R. S. Rittmaster, S. N. Rous, P. Hanno, E. Stoner, and R. Vagelos for useful discussions and comments, and Profs. A. Wein, P. Shaman, W. Frati, and J. M. Sprague for a critical reading of the manuscript.