LETTERS Antiparklnson Efficacy of Depren yl David Oakes, PhD, for DATATOP Steering Committee on behalf of Parkinson Study Group Schulzer and colleagues {l) show misguided ingenuity in reconstructing raw data from published DATATOP KaplanMeier curves 121. There are problems with their approach and conclusions. Their method introduced minor numerical errors. Also, if correlations between data points on cumulative hazard curves were ignored in deriving standard errors for the fitted parameters, these standard errors will have been substantially underestimated, and statistical significance of the fitted parameters overestimated. We refitted the parametric models in [I] to the actual DATATOP data analyzed in {2). Our results differed, particularly for p values of the fitted coefficients, but they lend some support to rejection of the proportional hazards model in favor of convergence of the deprenyliplacebo hazard ratio to unity. Differences from our results [2) are due to use of a different statistical test, and our stratification by investigator. Our more recent results including further follow-up 131 do suggest that the deprenyliplacebo hazard ratio may converge to unity, so that we have no substantive disagreement regarding the statistical behavior of the hazard ratio. W e strongly disagree with the interpretation in 11. A hazard function is the risk of reaching the endpoint at a specific time among subjects who have not previously reached the endpoint, essentially the slope of a cumulative incidence curve. Even if the deprenyl/placebo hazard ratio did become unity after 1 year, deprenyl would not lose its benefit after 1 year. Rather, the benefit of deprenyl seen in the first year would be maintained, though not enhanced, in subsequent years. Our data are much closer to pattern C of the schematic curves given by Schulzer and colleagues {I], two lines that separate for a while and are then parallel, than, as they claim, pattern D, two lines that separate and then come together. A more subtle point concerns the effect of heterogeneity among subjects. After 1 year 294 deprenyl subjects were still being followed without having reached endpoint, but only 21 1 nondeprenyl subjects were being followed without having reached endpoint. If deprenyl had no real effect after 1 year, the subsequent risk of endpointing among the 294 would exceed, not equal, the risk among the 2 11, because the former group would be less highly selected for survivorship. Schulzer and colleagues [l] imply that the effect of deprenyl is either totally symptomatic or totally protective. This oversimplifies the issue. W e have detected a symptomatic effect. W e have not definitively shown, but cannot exclude, that deprenyl also has a protective effect. Our continued follow-up of DATATOP subjects, both before and after initiation of levodopa therapy, may help to elucidate the effects of deprenyl, though definitive interpretation is hampered by the lack of a gold standard for measurement of neuronal dysfunction in Parkinson’s disease and also by the difficulty of inferring mechanisms as opposed to treatment outcomes on the basis of clinical trial data. University of Rochester School of Mediczne and Dentishy Rochater, N Y References 1. Schulzer M, Mak E, Calnc DB. The antiparkinson efficacy of deprenyl derives from transienr improvement that is likely to be sympromatic. Ann Neurol 1992;32:795-798 2. The Parkinson Study Group. Effect of deprenyl on the progression 06 disability i n early Parkinson’s disease. N Engl J Med 1989;321:1364-1171 3. The Parkinson Study Group. Effects of tocopherol and deprenyl on the progression of disability in early Parkinson’s disease. N Engl J Mcd 1993;328:176-183 Reply Michael Schulzer, MD, PhD, E. Mak, BASc, and D. B. Calne, D M We are pleased that upon reexamining his data, to which we had been denied access, Dr Oakes agrees with our findings in El] regarding the convergence of the hazard ratio to unity. W e thank him for his cordial characterization of our reconstruction of the raw data as a work of “ingenuity.” We are somewhat puzzled by D r Oakes’s statement that his data are “much closer to pattern C” of our schematic curves, “two lines which separate for a while and then are parallel.” If plac(t) and dep(t) represent, respectively, hazard functions corresponding to the placebo and to deprenyl, a parallel configuration at around 1 year means that, for t 1 year, plac(t) dep(t) + k, where k is constant. In that case, plac(t)/dep(t) 1 k/dep(t) and for this hazard ratio t o converge to unity, k must converge to zero and/or dep(t) must converge to infinity. The first case implies that the hazard functions come together at around 1 year, as we found in [l]. The second case implies that deprenyl hazard function increases greatly at 1 year, an even worse outcome for the argument for deprenyl efficacy. The schematic curves that we displayed in [l], and to which Dr Oakes refers, outline a number of scenarios. W e were quite cognizant of the intermediate models between a totally symptomatic and a totally neuroprotective effect. However, we found no particular evidence for their support. W e object to D r Oakes’s labeling of our conclusions as misguided. We were well guided, inter alia, by his own book {2], to which we gave due credit [l]. In it he states: “There are a number of reasons why consideration of the hazard function may be a good idea. . . . Comparisons of groups are . . . most incisively made via the hazard” (pp 15-16). Again, on pp 76-77 he examines the case in which “the hazard for individuals in group 1 eventually revem to that for group 0” (thus the hazard ratio converges to unity). The title of that section (section 5.7) is, appropriately, “Treatments with a Transient Effect,” in accord with our conclusion Ell. W e remain mystified by the total omission from D r Oakes’s report 131 of any reference to the incisive results of the analysis of the hazard ratio and to their fundamental implication. The conclusions from this analysis speak for themselves. Neurogenerative Disorders Centre University Hospital, Vancouver, BC, Canada + References 1. Schulzer M, Mak E, Calne DB. The antiparkinson efficacy of deprenyl derives from transient improvement that is likely to be symptomatic. Ann Neurol 1992;32:795-798 2. Cox UR, Oakes D. Analysis of survival data. London: Chapman and Hall, 1984 3. The Parkinson Study Group. Effect of deprenyl o n the progression of disability in early Parkinson’s disease. N Engl J Med 198‘);321:1364-1371

1/--страниц