To the Editor:
The recent report by Najean and Rain for the French Polycythemia Study Group1 comparing hydroxyurea with pipobroman in the treatment of polycythemia vera (PV) makes an important contribution to the debate about the treatment of this condition. We would like to make some observations about the presentation of their results. This particularly relates to their statements about the observed actuarial risk of leukemia. The investigators chose, both in the text and the summary, to state that for both drugs “the risk of leukemia is approximately 10% at the 13th year.” This statement may be misleading if it is compared to other studies without presenting confidence intervals for their observation and also appreciating that other studies often quote an overall percentage incidence of leukemia rather than actuarial risk. In their study approximately 27% and 15% of the total number of patients involved remained under observation at the 10th and 13th year, respectively. Thus, by the 13th year the number of patients remaining was relatively small and the accuracy of the actuarial risk of leukemia at this timepoint is likely to be suspect. We suggest that since the number remaining at 10 years was considerably larger, the investigators should have focused in the text and in their abstract on the actuarial risk of leukemia at that time. From their data, this is approximately 5% and 3% for hydroxyurea and pipobroman, respectively. This compares with the actuarial risk of leukemia at 10 years in the PVSG-01 study2 of approximately 16% and 18% in the 32P and chlorambucil limbs, respectively. Also in this study the estimates of risk of leukemia beyond 10 years are unstable because of the small number of patients still at risk.
With respect to leukemia, the figure of 1.5% is often quoted as the background incidence in those patients with PV not treated by cytoreductive therapy. This 1.5% incidence relates to those patients entered into the phlebotomy-only limb of the PVSG-01 study. However, this figure of 1.5% is misleading for two reasons. First, it was not based on an “intention to treat” basis; instead, patients requiring cytoreductive therapy were censored from analysis. Therefore, the figure relates to a highly selected group of patients. Secondly, as Najean and Rain3 point out, the number of patients managed by phlebotomy alone diminished rapidly over time—to less than 50% by the 5th year and 10% by the 10th year. Thus, the median follow-up time is significantly shorter than the other limbs of the PVSG-01 trial.
We believe that the evidence provided in these unique studies must be precisely evaluated to permit informed discussions in the possible leukemogenic risk of agents used in the treatment of PV.
I have read with interest the letter of my friend Prof Pearson. I am not a biostatistician and I am not sure to have well understood his argument. Why so sharp a criticism of the actuarial method for calculating the risks of complications and of death in a therapy protocol?
If we consider, for instance, the arm hydroxy-urea of our protocol of treatment of polycythemia, we observed six cases of leukemia, occurring at 4, 6, 7, 10, 10.5, and 11 years. It would be illogical to calculate the risk of leukemia from the number of events reported to the number of patients who did enter the protocol (136), since some of them have been treated for only a few months or years. Such a calculation would show a risk of leukemia of only 4.4%. I know that some investigators have published results with this type of calculation, but in my opinion, they are wrong. The actuarial method assigns each event to the number of living cases, treated and followed at the time of each event (95 cases at the 4th year, but only 39 at the 11th year). From the present data, we find a 10.8% risk, not the previous 4.4%.
I agree with Prof Pearson that there is an uncertainty when the number of cases is low. For instance, in the pipobroman arm, we did observe two cases of myelodysplasia at 3.5 and 4.5 years (respectively, 69 and 54 patients at risk), but the following cases of leukemia did occur at the 12th and 13th years (only 12 and 8 patients are then at risk). In this case, we clearly need a longer follow-up to give precise data beyond the 10th year. However, our results show that, at least for the first 10 years, the risk of leukemia in the pipobroman branch is not higher, and perhaps even lower, than that observed in the hydroxyurea branch.
The same method has been used for analyzing the risk of other complications, vascular events, carcinomas, myelofibrosis, and death. It has been also used for the 32P branch of our protocol. Why does Prof Pearson not criticize these data?
On the contrary, I agree with Prof Pearson’s other criticism. If, in fact, some patients with a particular risk are precociously excluded from a therapy protocol, the calculation of this risk for the whole population will be wrong. That was the case in the phlebotomy branch of the P.V.S.G. protocol, where the oldest and those with vascular risk factors or thrombocytosis have been excluded and treated otherwise. So, the true risk of leukemia could be under-estimated. But this is not a criticism of the method of calculation; this a criticism of the design, or of the fulfilment, of the protocol.
