- Email: [email protected]
In Reply: Re: Long-Term Survival and Mortality in Prostate Cancer Treated with Noncurative Intent
LETTERS TO THE EDITOR
297
lowed patients with properly staged disease who received no treatAlthough there is no lack of additional features in this study that ment until progression was documented or symptoms appeared. As merit criticism, the points raised suffice to show that their conclujudged from a recent pooled analysis from 6 nonrandomized studies sions are made without a sound scientific basis and may mislead of 828 men with clinically localized prostate cancer at diagnosis, the clinical colleagues. The most scientifically valid method to assess disease specific mortality rate a t 10 years was low, that is 13% of whether curative therapy is indicated for patients with localized men with-grade 1or 2 cancer and 66%with grade 3 cancer died ofthe prostate cancer and a life expectancy of 10 years or longer is to disease.' The low disease specific mortality rate in patients with conduct a randomized trial comparing curative and noncurative highly and moderately differentiated tumors means that when pros- therapy in this patient category. tate cancer is detected clinically it progresses slowly, at least the first Respectfully, 10 years after diagnosis. Per Anders Abrahamsson The survival rate after 10 years in patients with noncuratively Department of Urology treated prostate cancer, however, has been assessed in only a limited Lund University number of small, nonvalidated studies. Since curative treatment is Lund, Sweden presently a commonly recommended approach for clinically localized Hans Olov Adami and Adam Taube prostate cancer in younger men with a life expectancy of more than Departments of Cancer Epidemiology and Statistics 10 years, more information in terms of long-term survival and morUniversity of Uppsala tality in these patients treated with noncurative intention is urS-75120 Uppsala, Sweden gently needed. The high proportion of prostate cancer specific deaths among these patients reported by Aus et al, and the conclusion, if KyungMann Kim and Marvin Zelen correct, would have large clinical and practical implications. ThereDivision of Biostatistics and Department of Biostatistics Dana-Farber Cancer Institute fore, a critical assessment of the methods used is particularly imporHarvard School of Public Health tant. In a clinical cohort study, the point of departure is a group of Boston, Massachusetts 02115 individuals who had a certain diagnosis, for example prostate cancer, during a specified calendar period. These individuals are followed and with time, during which relapses, deaths or some events are regisMartin Kulldorff tered. This study design is widely used and the methodology for the Biometry Branch, DCPC subsequent statistical analysis is well established. National Cancer Institute However, as touched upon in the discussion, the approach adopted Bethesda, Maryland 20892 by Aus et al is inconsistent with the classical cohort strategy. Instead, the basis is all men who died during a certain calendar period 1. Chodak, G. W., Tisted, R. A., Gerber, G. S., Hohansson, J . E., (1988 to 1990) who were identified through the cancer registry as Adolfsson, J., Jones, G. W.,Chisholm, G. D., Moskovitz, B., Livne, P. M. and Warner, J.: Results of conservative managehaving had prostate cancer diagnosed in Goteborg sometime bement of clinical localized prostate cancer. New Engl. J. Med., tween 1958 and 1987, and for whom the relevant data were collected 330 242, 1994. retrospectively. This approach is unconventional and well estab2. Stenbeck, M. and Rosen, M.: Cancer survival in Sweden in lished statistical methods do not exist for the analysis of such data. 1961-1991. Acta Oncol., suppl. 4, 34: 63, 1995. Not only did Aus et al fail to consider the consequences of this fact when designing the study but they performed an analysis as if the data were derived from a proper cohort study. Reply by Authors. Abrahamsson et al state that our conclusions There are several problems with this approach. In this type of analysis there are no survivors. People surviving more than approx- (that is that mortality rate due to localized prostate cancer increases imately 30 years are automatically excluded, leading to an underes- when the observation time exceeded 10 years) are made without a timation of the survival curve and especially so in the later patient sound scientific basis and may mislead clinical colleagues. This strong statement warrants a comment. ages when there are fewer observations. Also, those with short SIXIn the study we have chosen primarily to present mortality as viva1 are recent cases while those with long survival were diagnosed long ago. It can be demonstrated that under certain conditions this proportional mortality ratio (the number of prostate cancer deaths approach will yield results comparable to those of a prospective divided by the number of all deaths) in 5-year intervals after diagcohort, that is if neither the population at risk and the incidence of nosis. Thus, only patients diagnosed within the same period were the disease nor the survival change during the period under consid- compared. In this analysis it was found that prostate cancer mortaleration (which must be longer than the longest possible survival of a ity increased with increasing observation time stage by stage and prostate cancer patient). In fact, however, the incidence of the dis- grade by grade. Even if changes in the population have occurred, it ease has nearly doubled from 1958 to 1987, the population at risk did is almost impossible to explain these data by selection bias. Lifenot remain constant neither in size nor in age distribution and the table and cause specific survival curves have only been constructed for comparison with other data and this is also clearly stated in the survival increased.2 Regarding prostate cancer deaths, whether a prostate cancer pa- article. It is also stated that these curves should be interpreted with tient will eventually die of the disease depends not only on the death caution. Nevertheless, cause specific survival closely followed the risk associated with prostate cancer but also on the patient risk results from other series of deferred treatment with up to 15 years of pattern in relation to other competing causes of death. In connection followup. A further deterioration beyond 15 years is not surprising, with their table 5, Aus et a1 assert that "Survival longer than 10 since most other series have shown a progression rate corresponding Years was noted in 65 stage MO cancer patients. The mortality rate to the mortality rate in our series. Swehsh registry studies from due to prostate cancer was higher in this group compared to the Uppsala have also shown a progressive mortality rate in prostate entire group with stage MO disease (63%versus 50%,P <0.05)."Not cancer up to 20 years after diagnosis, with an almost constant anonly is such a comparison meaningless, it is simply wrong (since the nual excess death rate between 5 and 20 years after diagnosis. These 2 groups are not independent). There are more prostate cancer late prostate cancer deaths must certainly have been early prostate specific deaths among those whose survival is beyond 10 years, not cancer at diagnosis.' Are the patients studied comparable to a prospective cohort? It because prostate cancer specific mortality is higher in this group but, rather, because other causes bring about earlier death among the is not correct to state that our patient group is inconsistent with actual patients. This fact is especially true considering that the a classical cohort strategy. The prevalence of prostate cancer in patients were 5 years older to begin with, as the authors indicated in Giiteborg as of December 31, 1990 revealed that only 6 cases were diagnosed before 1965 (all other patients alive with prostate cancer their discussion. The Kaplan-Meier curves presented by Aus et al are cause specific on this date had been diagnosed with disease after 1965). This means and the survival times for all persons who died of causes other than that the risk that patients whose cancer was diagnosed before 1958 Prostate cancer, that is the rest of the data, are treated as censored (when the cancer registry started) were missed in the actual study is observations. Thus, the shape ofthe curves is determined not only by extremely small. We do not understand the criticism that there are deaths from prostate cancer but also by deaths from other causes. As no survivors. All patients die sooner or later but this does not disa minor point, note that some of the Kaplan-Meier curves are based criminate for or against a specific cause of death. If prostate cancer on few observations, although they make a dramatic impression, for patients are followed long enough of course all must eventually die. example the curves for the stages T l a and T4 groups are based on One might, therefore, look upon the actual study population as a sample from the entire group of patients diagnosed between 1958 only 20 (with 2 deaths) and 10 patients, respectively.
298
LETTERS TO THE EDITOR
and 1990. If no changes in incidence, population size or mean survival occur this sample would yield the same results as a prospective cohort. As Abrahamsson et al note, such changes have occurred and the question is how these changes affect the results. Since we know all of the changes that occurred during the actual period, we put the data into a computer model to study the effect of these changes were studied. Interestingly. these changes had almost no impact on mortality rate and mean survival was influenced by less than 10%.We therefore believe that this study is representative of a population of unselected patients with all stages of prostate cancer followed from diagnosis to death. We are also tempted to say that this study is a t least as good a s most prospective studies, since it avoids frequent problems with selection and patients lost to followup. often encountered in prospective studies. The conclusion is supported by the results from a recent study based on register data in which all consecutive patients in Goteborg diagnosed with prostate cancer from 1960 to 1980 and who survived for more than 10 years were examined (prospective series). The proportional mortality ratio in these 490 patients was 621, in perfect harmony with the present series.2 We agree that the comparison between all stage MO cancer patients with the 65 who survived for longer than 10 years is incorrect from a statistical viewpoint, since the groups are dependent on each other. It would have been more correct to compare stage MO cancer patients with more or less than 10 years of survival, which renders a larger difference (46% versus 63%). This finding. in combination with the findings in tables 2 and 5 in our article, show again that the proportional mortality ratio is higher in patients with longer followup (survival)than in those with shorter followup. We also agree that a randomized trial with no selection bias would provide excellent data regarding the effects of curative treatment. However. we have more than 30 years of experience in Sweden with palliative treatment for localized prostate cancer. Why not carefully study these patients, which might help us to provide some fruitful clinical recommendations for prostate cancer patients to date? 1. Norlen, B. J.: Survival and mortality in prostatic cancer. A study based on the Swedish Cancer Register. Acta Oncol., 3 0 141, 1991. 2. Hugosson. J., Aus, G., Bergdahl, C. and Bergdahl, S.: Prostate cancer mortality in patients surviving more than 10 years after diagnosis. J. Urol., accepted for publication. Note by Editor. This Letter to the Editor was written at my re uest inasmuch as it was too late for an Editorial Comment to%, appended to this article at the time of its publication. Also, we have received 2 subsequent Letters to the Editor concerning the article, which will be published in future issues of the Journal.
ERRATA
INTERSTITIAL CYSTITIS Volume 153, Number 4, page 1322, Table 1: Codon 3 is GYT in sequence for primer 5, codon 7 is GAA and codon 10 is CT in sequence for primer 3, codon 9 is GTG in sequence for primer 4 and codon 1 is CCG in sequence for primer 240.
PERCUTANEOUS AORTIC STENT PLACEMENT Volume 153, Number 5, page 1632, paragraph 1, sentence 2 should read: ". . .6 mm. Impragraft (Impra, Inc., Tempe, Arizona." and legends for figures 3 and 4 should read: ''. . .polyethylene terephtalate grafts. . .";page 1633, paragraph 2, sentence 3 should read: ". . .material is usually thin polyethylene terephtalate. . .".
URODYNAMIC EVALUATION OF INTRAVESICAL OBSTRUCTION Volume 154, Number 2, pages 582-585: The correct figures 1 to 9 will be supplied by the authors upon request.
297
lowed patients with properly staged disease who received no treatAlthough there is no lack of additional features in this study that ment until progression was documented or symptoms appeared. As merit criticism, the points raised suffice to show that their conclujudged from a recent pooled analysis from 6 nonrandomized studies sions are made without a sound scientific basis and may mislead of 828 men with clinically localized prostate cancer at diagnosis, the clinical colleagues. The most scientifically valid method to assess disease specific mortality rate a t 10 years was low, that is 13% of whether curative therapy is indicated for patients with localized men with-grade 1or 2 cancer and 66%with grade 3 cancer died ofthe prostate cancer and a life expectancy of 10 years or longer is to disease.' The low disease specific mortality rate in patients with conduct a randomized trial comparing curative and noncurative highly and moderately differentiated tumors means that when pros- therapy in this patient category. tate cancer is detected clinically it progresses slowly, at least the first Respectfully, 10 years after diagnosis. Per Anders Abrahamsson The survival rate after 10 years in patients with noncuratively Department of Urology treated prostate cancer, however, has been assessed in only a limited Lund University number of small, nonvalidated studies. Since curative treatment is Lund, Sweden presently a commonly recommended approach for clinically localized Hans Olov Adami and Adam Taube prostate cancer in younger men with a life expectancy of more than Departments of Cancer Epidemiology and Statistics 10 years, more information in terms of long-term survival and morUniversity of Uppsala tality in these patients treated with noncurative intention is urS-75120 Uppsala, Sweden gently needed. The high proportion of prostate cancer specific deaths among these patients reported by Aus et al, and the conclusion, if KyungMann Kim and Marvin Zelen correct, would have large clinical and practical implications. ThereDivision of Biostatistics and Department of Biostatistics Dana-Farber Cancer Institute fore, a critical assessment of the methods used is particularly imporHarvard School of Public Health tant. In a clinical cohort study, the point of departure is a group of Boston, Massachusetts 02115 individuals who had a certain diagnosis, for example prostate cancer, during a specified calendar period. These individuals are followed and with time, during which relapses, deaths or some events are regisMartin Kulldorff tered. This study design is widely used and the methodology for the Biometry Branch, DCPC subsequent statistical analysis is well established. National Cancer Institute However, as touched upon in the discussion, the approach adopted Bethesda, Maryland 20892 by Aus et al is inconsistent with the classical cohort strategy. Instead, the basis is all men who died during a certain calendar period 1. Chodak, G. W., Tisted, R. A., Gerber, G. S., Hohansson, J . E., (1988 to 1990) who were identified through the cancer registry as Adolfsson, J., Jones, G. W.,Chisholm, G. D., Moskovitz, B., Livne, P. M. and Warner, J.: Results of conservative managehaving had prostate cancer diagnosed in Goteborg sometime bement of clinical localized prostate cancer. New Engl. J. Med., tween 1958 and 1987, and for whom the relevant data were collected 330 242, 1994. retrospectively. This approach is unconventional and well estab2. Stenbeck, M. and Rosen, M.: Cancer survival in Sweden in lished statistical methods do not exist for the analysis of such data. 1961-1991. Acta Oncol., suppl. 4, 34: 63, 1995. Not only did Aus et al fail to consider the consequences of this fact when designing the study but they performed an analysis as if the data were derived from a proper cohort study. Reply by Authors. Abrahamsson et al state that our conclusions There are several problems with this approach. In this type of analysis there are no survivors. People surviving more than approx- (that is that mortality rate due to localized prostate cancer increases imately 30 years are automatically excluded, leading to an underes- when the observation time exceeded 10 years) are made without a timation of the survival curve and especially so in the later patient sound scientific basis and may mislead clinical colleagues. This strong statement warrants a comment. ages when there are fewer observations. Also, those with short SIXIn the study we have chosen primarily to present mortality as viva1 are recent cases while those with long survival were diagnosed long ago. It can be demonstrated that under certain conditions this proportional mortality ratio (the number of prostate cancer deaths approach will yield results comparable to those of a prospective divided by the number of all deaths) in 5-year intervals after diagcohort, that is if neither the population at risk and the incidence of nosis. Thus, only patients diagnosed within the same period were the disease nor the survival change during the period under consid- compared. In this analysis it was found that prostate cancer mortaleration (which must be longer than the longest possible survival of a ity increased with increasing observation time stage by stage and prostate cancer patient). In fact, however, the incidence of the dis- grade by grade. Even if changes in the population have occurred, it ease has nearly doubled from 1958 to 1987, the population at risk did is almost impossible to explain these data by selection bias. Lifenot remain constant neither in size nor in age distribution and the table and cause specific survival curves have only been constructed for comparison with other data and this is also clearly stated in the survival increased.2 Regarding prostate cancer deaths, whether a prostate cancer pa- article. It is also stated that these curves should be interpreted with tient will eventually die of the disease depends not only on the death caution. Nevertheless, cause specific survival closely followed the risk associated with prostate cancer but also on the patient risk results from other series of deferred treatment with up to 15 years of pattern in relation to other competing causes of death. In connection followup. A further deterioration beyond 15 years is not surprising, with their table 5, Aus et a1 assert that "Survival longer than 10 since most other series have shown a progression rate corresponding Years was noted in 65 stage MO cancer patients. The mortality rate to the mortality rate in our series. Swehsh registry studies from due to prostate cancer was higher in this group compared to the Uppsala have also shown a progressive mortality rate in prostate entire group with stage MO disease (63%versus 50%,P <0.05)."Not cancer up to 20 years after diagnosis, with an almost constant anonly is such a comparison meaningless, it is simply wrong (since the nual excess death rate between 5 and 20 years after diagnosis. These 2 groups are not independent). There are more prostate cancer late prostate cancer deaths must certainly have been early prostate specific deaths among those whose survival is beyond 10 years, not cancer at diagnosis.' Are the patients studied comparable to a prospective cohort? It because prostate cancer specific mortality is higher in this group but, rather, because other causes bring about earlier death among the is not correct to state that our patient group is inconsistent with actual patients. This fact is especially true considering that the a classical cohort strategy. The prevalence of prostate cancer in patients were 5 years older to begin with, as the authors indicated in Giiteborg as of December 31, 1990 revealed that only 6 cases were diagnosed before 1965 (all other patients alive with prostate cancer their discussion. The Kaplan-Meier curves presented by Aus et al are cause specific on this date had been diagnosed with disease after 1965). This means and the survival times for all persons who died of causes other than that the risk that patients whose cancer was diagnosed before 1958 Prostate cancer, that is the rest of the data, are treated as censored (when the cancer registry started) were missed in the actual study is observations. Thus, the shape ofthe curves is determined not only by extremely small. We do not understand the criticism that there are deaths from prostate cancer but also by deaths from other causes. As no survivors. All patients die sooner or later but this does not disa minor point, note that some of the Kaplan-Meier curves are based criminate for or against a specific cause of death. If prostate cancer on few observations, although they make a dramatic impression, for patients are followed long enough of course all must eventually die. example the curves for the stages T l a and T4 groups are based on One might, therefore, look upon the actual study population as a sample from the entire group of patients diagnosed between 1958 only 20 (with 2 deaths) and 10 patients, respectively.
298
LETTERS TO THE EDITOR
and 1990. If no changes in incidence, population size or mean survival occur this sample would yield the same results as a prospective cohort. As Abrahamsson et al note, such changes have occurred and the question is how these changes affect the results. Since we know all of the changes that occurred during the actual period, we put the data into a computer model to study the effect of these changes were studied. Interestingly. these changes had almost no impact on mortality rate and mean survival was influenced by less than 10%.We therefore believe that this study is representative of a population of unselected patients with all stages of prostate cancer followed from diagnosis to death. We are also tempted to say that this study is a t least as good a s most prospective studies, since it avoids frequent problems with selection and patients lost to followup. often encountered in prospective studies. The conclusion is supported by the results from a recent study based on register data in which all consecutive patients in Goteborg diagnosed with prostate cancer from 1960 to 1980 and who survived for more than 10 years were examined (prospective series). The proportional mortality ratio in these 490 patients was 621, in perfect harmony with the present series.2 We agree that the comparison between all stage MO cancer patients with the 65 who survived for longer than 10 years is incorrect from a statistical viewpoint, since the groups are dependent on each other. It would have been more correct to compare stage MO cancer patients with more or less than 10 years of survival, which renders a larger difference (46% versus 63%). This finding. in combination with the findings in tables 2 and 5 in our article, show again that the proportional mortality ratio is higher in patients with longer followup (survival)than in those with shorter followup. We also agree that a randomized trial with no selection bias would provide excellent data regarding the effects of curative treatment. However. we have more than 30 years of experience in Sweden with palliative treatment for localized prostate cancer. Why not carefully study these patients, which might help us to provide some fruitful clinical recommendations for prostate cancer patients to date? 1. Norlen, B. J.: Survival and mortality in prostatic cancer. A study based on the Swedish Cancer Register. Acta Oncol., 3 0 141, 1991. 2. Hugosson. J., Aus, G., Bergdahl, C. and Bergdahl, S.: Prostate cancer mortality in patients surviving more than 10 years after diagnosis. J. Urol., accepted for publication. Note by Editor. This Letter to the Editor was written at my re uest inasmuch as it was too late for an Editorial Comment to%, appended to this article at the time of its publication. Also, we have received 2 subsequent Letters to the Editor concerning the article, which will be published in future issues of the Journal.
ERRATA
INTERSTITIAL CYSTITIS Volume 153, Number 4, page 1322, Table 1: Codon 3 is GYT in sequence for primer 5, codon 7 is GAA and codon 10 is CT in sequence for primer 3, codon 9 is GTG in sequence for primer 4 and codon 1 is CCG in sequence for primer 240.
PERCUTANEOUS AORTIC STENT PLACEMENT Volume 153, Number 5, page 1632, paragraph 1, sentence 2 should read: ". . .6 mm. Impragraft (Impra, Inc., Tempe, Arizona." and legends for figures 3 and 4 should read: ''. . .polyethylene terephtalate grafts. . .";page 1633, paragraph 2, sentence 3 should read: ". . .material is usually thin polyethylene terephtalate. . .".
URODYNAMIC EVALUATION OF INTRAVESICAL OBSTRUCTION Volume 154, Number 2, pages 582-585: The correct figures 1 to 9 will be supplied by the authors upon request.