Estimated impact of the Prostate Cancer Prevention Trial on population mortality

Abstract

BACKGROUND

The potential public health impact of the recently completed Prostate Cancer Prevention Trial (PCPT) is debated. The results indicated that the period prevalence of prostate cancer was reduced by 24.8% due to finasteride, whereas an increase in the rate of high-grade tumors (Gleason score 8–10) among men who were diagnosed with cancer also was found (5.0% in the PCPT placebo arm vs. 11.9% in the PCPT finasteride arm). Whether the increased Gleason score was valid or was a histologic artifact is under investigation.

METHODS

The authors estimated the number of person-years saved assuming a 24.8% reduction in the incidence of prostate cancer for 5 years among United States males age ≥ 55 years. Scenarios for different proportions of patients with high-grade Gleason scores also were considered.

RESULTS

With a 24.8% reduction in the number of men with newly diagnosed prostate cancer, the authors estimated that 316,760 person-years would be saved due to finasteride in the United States. An absolute increase of 6.9% in the proportion of men with high-grade tumors in the United States cancer population (corresponding to the difference between the rates on the placebo and finasteride arms of the PCPT) would reduce the number of person-years saved to 262,567. For each absolute increase of 5% in the proportion of patients with high-grade tumors, the number of person-years saved would be reduced by approximately 39,000.

CONCLUSIONS

The results of the PCPT may have a major impact on population mortality from prostate cancer if they are applied clinically. The potential detrimental effects of an increased rate of patients who have prostate cancer with high-grade Gleason scores would be outweighed by a reduction in incidence. Cancer 2005. © 2005 American Cancer Society.

The recent results from the Prostate Cancer Prevention Trial (PCPT) represent a milestone in cancer research.1 It has now been established that prostate cancer can be prevented through chemoprevention. The results of the PCPT showed that the administration of a daily dose of 5 mg of finasteride significantly reduced the 7-year period prevalence of prostate cancer. On the finasteride arm of the study, cancer was detected in 24.8% fewer men than on the placebo arm (P < 0.001). The PCPT joins the Breast Cancer Prevention Trial in demonstrating the potential for cancer prevention.2

The public health impact of the PCPT remains a source of debate. Although finasteride reduced the period prevalence of prostate cancer by 24.8%, an increase in the percent of men who had prostate cancer with high-grade characteristics in the finasteride arm (11.9%), compared with the placebo arm (5.0%), also was found. Some investigators have argued that this increase in high-grade tumors may nullify the benefits of reducing the period prevalence.3, 4 In the absence of long-term survival data for PCPT participants, the potential impact of finasteride on population mortality is unknown.

Estimating the impact of a clinical trial result on population mortality is a topic that has received little attention. Recently, the Southwest Oncology Group presented a model that uses clinical trial results and National Cancer Registry data from the Surveillance, Epidemiology, and End Results (SEER) survey to estimate how positive Phase III therapeutic or prevention trial results may impact on population mortality.5, 6 It was estimated that the PCPT had an impact on population mortality similar to eight positive therapeutic trials, illustrating the public health potential of cancer prevention. However, this estimate was based on the observed reduction in incidence only and did not attempt to account for possible mitigating factors. In the current analysis, we further investigated the potential impact of finasteride on population mortality, extending the prior model to allow for the adjustment of other factors, in particular, the increase in high-grade tumors.

MATERIALS AND METHODS

Data

Participants on the PCPT were randomized to receive either 5 mg finasteride daily or placebo for 7 years. The main study results from the PCPT have been reported previously.1 The following results from the PCPT serve as data for this analysis: Among the 4368 men on the finasteride arm with data available for the PCPT final analysis, 803 man (18.4%) had prostate cancer detected; in the placebo group, 1147 of 4692 men (24.4%) had prostate cancer detected. The relative difference, which represents the primary endpoint of the study, was a 24.8% reduction (P < 0.001) in the period prevalence of prostate cancer at 7 years in the finasteride arm. It is noteworthy that the percent reduction in prostate cancers detected for cause (that is, due to an elevated prostate-specific antigen or abnormal digital rectal examination) or after another interim procedure was of similar magnitude (23.8%).

For the current analysis, high-grade Gleason scores are defined as Gleason scores between 8 and 10 to match the coding system for the SEER registry. This analysis is possible because of the similar number of Gleason 7 tumors in both study groups: 190 Gleason 7 tumors in the finasteride arm and 184 Gleason 7 tumors in the placebo arm. Overall, most graded tumors were low or intermediate grade (Gleason score, 2–7). However, a higher proportion (and a higher rate) of tumors on the finasteride arm had Gleason scores of 8, 9, or 10 (90 of 757 graded tumors; 11.9%) compared with the placebo arm (52 of 1068 graded tumors; 5.0%). Most men with prostate cancer (98%) had clinically localized disease.

The cancer population data for this analysis are from the nine standard areas of the SEER registry.7 To match the accrual period of the PCPT, we considered only men who were diagnosed with prostate cancer during the period from 1993 to 1997. In addition, to match the eligibility criteria of the PCPT, we restricted our SEER dataset to men age ≥ 55 years. Estimates of normal survival (in which the proportion of individuals dying during any interval is based on death due to any cause) for the United States population were derived from the life tables of the National Center for Health Statistics (NCHS).8

Statistical Methods

Estimating person-years saved.

The modeling method for estimating the number of person-years saved (PYS) due to a positive prevention trial based on a single variable (percent of cases prevented) has been summarized briefly.5, 6 The method is extended here, allowing for adjustment of multiple variables. In particular, we wanted to adjust for potential induced changes in the proportion of high Gleason scores (grade 8–10) among men who get cancer.

For a given historic cohort of individuals who developed cancer—in this instance, men with prostate cancer in the SEER registry from 1993 to 1997—we can estimate how many person-years would have been saved over 10 years assuming a proportion of cases had been prevented. PYS would be computed as the difference between person-years lived (PYL) with prevention and PYL without prevention.

• (1)

In general, the number of PYL for a cohort is the product of the size of the cohort (N) and the average number of years lived (AVG) for a single member of the cohort: PYL = N * AVG. The average number of years lived for a single member of the cohort is calculated as the area beneath the survival curve for the cohort to account for censoring (illustrated in Fig. 1).

Figure 1. Person-years saved in prevention setting. N_D: the number of individuals with disease; AVG_DISEASE: the average number of years lived for a patient in the disease cohort; AVG_NORMAL: the average number of years lived for an individual in the normal cohort.

This method condition on the number of observed cases with disease, represented by N_D. Thus, PYL_{WITHOUT PREVENTION} for the disease cohort is

• (2)

where AVG_DISEASE is the average number of years lived for a patient in the disease cohort. PYL_{WITH PREVENTION} is comprised of two components: a certain proportion of the patients (X%) will not get the disease. This subset of patients, of size N_D * X%, will have normal survival, with the average number of years lived for a member of this subset represented by AVG_NORMAL (see Fig. 1). The remaining N_D * (1 − X%) patients will have reduced survival due to disease: AVG_DISEASE. The sum of these 2 components represents PYL for the cohort with prevention:

• (3)

Equations 2 and 3 represent the components of Equation 1. Simple substitution produces:

• (4)

Note that Equation 4 reduces to a very simple equation as follows: PYS = N*X%*(AVG_NORMAL − AVG_DISEASE). This equation is relevant only to the “univariate” case, in which no other factors are adjusted for, such as Gleason score.

Suppose that, like in the PCPT, the prevention regimen induces a change in the characteristics of the cohort of individuals who develop disease, leading to a change in the overall survival for the cohort. Consider again Equation 4 above. We can regard AVG_DISEASE as a weighted sum of the subgroups of a clinical variable. For instance, a proportion of all individuals with prostate cancer will have a high-grade (HG) Gleason score (Pr_HG), and the rest will have a low-grade or intermediate-grade (LIG) Gleason score (Pr_LIG); the two groups also will have different survival (AVG_DISEASE-HG and AVG_DISEASE-LIG, respectively). Thus,

• (5)

To adjust for Gleason score or for any clinical variable, we reweight the proportion of cases in the high-grade versus low-grade/intermediate-grade groups. For instance, if we assume that administration of finasteride induces an absolute increase of 5% in the proportion of patients with high-grade Gleason score, then Equation 5 becomes AVG_DISEASE = (Pr_HG + 0.05) * AVG_DISEASE-HG + (Pr_LIG − 0.05) * AVG_DISEASE-LIG. Substituting the recalculated AVG_DISEASE from Equation 5 into AVG_DISEASE from the PYL_{WITH PREVENTION} component of Equation 4 gives us PYS adjusted for a change in the distribution of Gleason scores.

In practice, as discussed above, we compute PYS for the SEER cohort. Because SEER represents 9.5% of the United States population, we divide our estimate of PYS for SEER by 0.095 to get the total number of PYS in the United States. Our estimate of PYS is measured out to 10 years from diagnosis.

Model parameters.

The percent reduction in the number of new diagnoses is obtained from the reduction in period prevalence in the clinical trial, so X% in the equations presented above = 24.8%. It is worth noting that a reduction in the period prevalence represents a conservative estimate of the reduction in number of new cases, because a nonzero fraction of the participants would have entered the PCPT with existing (but undetected) prostate cancer. In addition, due to the 25% reduction in prostate gland volume achieved with finasteride and the same number of core biopsies performed in both study groups, the true reduction in prevalence, in all probability, is greater due to the proportionate oversampling of the prostate in men in the finasteride group. We also know from the PCPT that a population administered finasteride may experience an increase in the proportion of men with high-grade Gleason scores. The study showed an absolute difference of 6.9% between the finasteride arm (11.9%) and the placebo arm (5.0%). However, the difference between the proportion of high-grade tumors in the placebo arm of the PCPT (5.0%) versus the general cancer population according to SEER (19.2%) is sufficiently large to make it difficult to extrapolate how finasteride may affect the general cancer population; as such, in addition to estimation based on the difference of 6.9%, we also provide a range of estimates. Thus, the results represent a conditional sensitivity analysis, with the proportion of high-grade tumors allowed to vary and with the percent reduction in prostate cancers conditioned on the observed rate of 24.8%.

Assumptions.

Like any model, the PYS method requires some assumptions both to simplify the modeling procedure and to represent factors that may be unknown. The assumptions were as follows: 1) We assume a percent reduction in the incidence of prostate cancer in the population equal to the percent reduction in period prevalence found in the PCPT (24.8%). Although this was a heavily screened population, the relevant issue is not the absolute number of prostate cancers detected but the relative difference (percent reduction) in detected prostate cancers due to finasteride. 2) We assume that the 24.8% of cases of prostate cancer in the population that would have been prevented with the administration of finasteride truly are prevented from occurring and not simply delayed until later onset. 3) We assume that the SEER data base, which represents a clinically diagnosed cancer population, is a valid population for use in this modeling procedure. Because the percent reduction in prostate cancers detected by biopsies performed for cause was of similar magnitude (23.8%) to the overall percent reduction in period prevalence (24.8%), we conclude that projections using SEER data are not unreasonable. 4) We assume that the survival of SEER prostate cancer patients is representative of the outcome for prostate cancer patients in the general United States population. 5) We assume that normal survival (as indicated in Fig. 2) for the general United States population, derived from the NCHS, is comparable to normal survival in the SEER registry areas. This is crucial, because the PYS estimate is based on the differences between normal survival and survival for those persons (from SEER) with local/regional or distant disease. It is particularly noteworthy that, as illustrated in Figure 2A, there is an added risk of death due to local/regional prostate cancer, especially beyond 5 years from diagnosis. This added risk of death due to indolent prostate cancer has been noted previously.9 Even among the low-grade (Gleason score, 2–7) tumors in patients within local/regional disease, the risk of death exceeds that of the normal population (Fig. 2B). Thus, low-grade tumors are relevant clinically and will contribute to PYS if prevented.

Figure 2. (A) Ten year overall survival for the Surveillance, Epidemiology, and End Results (SEER) population with local/regional prostate cancer; versus the SEER population with distant prostate cancer; versus a population with normal survival from the National Center for Health Statistics (NCHS). (B) Ten year overall survival for the SEER population with Gleason score 2–7 prostate cancer (local/regional stage only); versus the SEER population with Gleason score 8–10 prostate cancer (local/regional stage only); versus a population with normal survival from the NCHS.

RESULTS

Our analysis shows that if finasteride had prevented 24.8% of the prostate cancers (any stage) among the male population age > 55 years from 1993 to 1997, then 316,760 person-years would have been saved over a 10-year period (Table 1). According to the method outlined above, we can adjust for possible changes to the distribution in Gleason scores that finasteride may induce in a population. The SEER registry indicates that 19.2% of new prostate cancers from 1993 to 1997 were high-grade tumors (Gleason score, 8–10), with 70.7% of tumors Gleason score 2–7, 9.6% unknown, and 0.6% undifferentiated. Among graded tumors on the PCPT, there was an increase in high-grade Gleason scores (8–10) due to finasteride, from 5.0% on the placebo arm to 11.9% on the finasteride arm: an absolute difference of 6.9%. A similar 6.9% absolute increase in the United States prostate cancer population, from 19.2% to 26.1%, would result in 262,576 PYS (any stage). But the actual percentage of men in a population who are administered finasteride who may present with higher Gleason score tumors is unknown; thus, Table 1 shows a range of possible values, indicating that every increase in 5% in the proportion of high-grade tumors in the general cancer population due to finasteride reduces PYS by about 39,000. However, even if we hypothesize more than a doubling of the proportion of high-grade tumors in the cancer population (from 19.2% to 39.2%), 159,680 person-years still would have been saved, representing a positive benefit to society.

Table 1. Person-Years Saved at 10 Years Adjusted for Possible Induced Changes in Gleason Score among Individuals who Took Finasteride

Prevention of…	Induced change in Gleason score 8–10 in cancer populationa
	− 5%	No change	+ 5%	+ 6.9%b	+ 10%	+ 15%	+ 20%
a According to Surveillance, Epidemiology, and End Results survey data from 1993 to 1997, the percent of patients with high- grade Gleason scores (8–10) for any stage of disease was 19.2%, and the percent of patients with local/regional disease only was 18.2%. b This value (6.9%) is the absolute difference in the percent of high-grade Gleason scores (8–10) between the finasteride arm (11.9%) and the placebo arm (5.0%) the Prostate Cancer Prevention Trial.
Any stage disease	356,030	316,760	277,490	262,567	238,220	198,950	159,680
Local/regional disease only	230,919	200,810	170,699	159,260	140,593	110,484	80,375

Because most prostate cancers detected in the PCPT were diagnosed as clinically localized disease, we also estimated how many PYS would be achieved in the prevention of local/regional (as opposed to any stage) disease only. Assuming no change in Gleason distribution in the population, PYS at 10 years would be 200,810. Estimates of PYS in the local/regional setting only, adjusted for changes in Gleason score, also are shown in Table 1. In this case, every increase of 5% in the proportion of high-grade tumors in the general cancer population due to finasteride reduces PYS by about 30,000.

DISCUSSION

Extending the impact of a chemopreventive agent from a clinical trial to a population mortality setting necessarily requires modeling, because the chemopreventive agent must be tested in a controlled sample of participants, not the entire population. Additional modeling assumptions are required when an endpoint other than survival is used. In the PCPT, a survival endpoint was considered at the time the study was designed; however, due to the size and time requirements, it was deemed that this endpoint was not feasible, and biopsy-proven prostate cancer was chosen instead.10 The PYS model allows a measurable and intuitive extension of this endpoint to mortality by conditioning on the observed number of prostate cancers and does not try to model hypothetical future demographics and prostate cancer incidence rates. This approach simplifies formulation of our model, allowing the use of observed rates and estimated survival curves from the appropriately age-restricted and stage-restricted population groups. A prospective approach presumably would need additional model components but would allow the ongoing monitoring of the impact of the prevention strategy with changes in populations over time. In addition, all-cause mortality, and not cause-specific prostate cancer death, was used as the primary outcome of this analysis. Although understanding the impact of finasteride on prostate cancer-specific death would be a laudable objective, the use of all-cause mortality eliminates the potential for significant errors in reporting of cause of death information known to occur in long-term clinical and population data bases. Finally, we limited our assessment to 10 years to focus on the most relevant component of patient survival.

Discussion about the results of the PCPT has focused on the higher rate of high-grade tumors on the finasteride arm versus the placebo arm of the study. Two particular questions regarding the potential clinical impact of finasteride are debated. First, assuming that the increase in high-grade tumors due to finasteride represents what would happen in the larger population, would the increase in high-grade tumors outweigh the benefits of reduced incidence? Second, are the high-grade tumors meaningful clinically?

The results of this analysis attempt to answer the first question. In the absence of any change to the distribution of high-grade tumors, we estimated that 316,760 person-years would be saved over 10 years. Moreover, the detrimental effect of the development of more high-grade tumors, even if it is meaningful clinically, is unlikely to outweigh the beneficial effect of the reduction in new cases of prostate cancer. Figure 3 shows how total PYS at 10 years would be affected by a change in the proportion of tumors with high Gleason scores. An increase in the proportion of high-grade tumors in the general cancer population of 6.9% (equal to the absolute difference between the rates of high-grade tumors in the finasteride and placebo arms of the PCPT) would reduce the number of PYS by 54,192–262,567. Even for substantially larger increases in the percentage of patients with high-grade Gleason scores, the benefits of reduced incidence outweigh the detrimental effect of higher Gleason scores. In fact, the proportion with high-grade Gleason scores would have to nearly triple (to about 60%) in the general prostate cancer population before the negative impact of high-grade tumors equaled the benefits of reduced incidence.

Figure 3. Person-years saved by changes in Gleason score.

It is not clear that the increase in the rate of high-grade Gleason tumors on the finasteride arm is meaningful clinically. It has been hypothesized that finasteride alters the appearance of tumors, a phenomenon that would render the Gleason grading system unreliable in this setting.11 Civantos et al. noted histopathologic changes in prostate cancer tumors among finasteride-treated patients and concluded that, unless a modified Gleason system was used, no Gleason scoring should be carried out.12 This recommendation was adopted by a consensus conference of the World Health Organization, the National Cancer Institute, and the American Cancer Society.13

The issue of Gleason scoring in the presence of hormonal therapy, such as finasteride, is complex; pathologic studies to investigate this issue further within the PCPT currently are underway. However, even if it is found that finasteride potentiates the growth of high-grade tumors, the current analysis shows that the potential detrimental effects of an increased rate of tumors with high-grade Gleason score would be substantially outweighed by a reduction in incidence. A final, oft-forgotten issue is the distinct benefit for the 25% of men who would not be affected by prostate cancer due to finasteride administration. The reduction in cost and morbidity as well as the psychosocial benefits of not having a cancer diagnosis potentially are as great as the life-years saved by this preventive intervention.