By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Pierre I. Karakiewicz, Cancer Prognostics and Health Outcomes Unit, University of Montreal Health Center (CHUM), 1058, rue St-Denis, Montréal, Québec, Canada, H2 X3J4. e-mail: email@example.com
To test the accuracy of life tables (LT), the standard tool for predicting life-expectancy (LE), but the accuracy of which is unknown in patients with prostate cancer, where the 10-year LE is a widely accepted threshold for the delivery of definitive therapy.
PATIENTS AND METHODS
We tested the accuracy of predictions of LE from LT in 9678 men treated with radical prostatectomy (RP) for prostate cancer. The predictions of LE from LT at 10 years after RP were compared to Kaplan Meier-derived 10-year survival values. Moreover, the accuracy of LT predictions was quantified in a Cox-regression using Harrell’s concordance index. To control for the effect of prostate cancer mortality, analyses were repeated in a subset of 5955 patients with no evidence of disease recurrence. Additional stratification schemes were applied to control for age and comorbidity.
At RP, the median age was 64 years, the median Charlson Comorbidity Index (CCI) was 1 and the median LT-derived LE was 16 years. The median actuarial survival was not reached (mean 12.4 years). In the whole group the LT-predicted 10-year survival was 96.8%, vs an observed of 75.3%. In men with no disease recurrence the LT-predicted survival was 97.3%, vs 81.1% observed. After age and CCI stratification, LT overestimated the 10-year survival the most in those aged 65–69 years and in patients with CCI scores of >2.
The overestimation of LE can lead to overtreatment of prostate cancer, especially in those men who die early from other causes.
A life-expectancy (LE) of >10 years is the most frequently used threshold for the delivery of definitive therapy to patients with localized prostate cancer. This ‘10-year’ rule has been adopted by several professional associations and appears in their guidelines [1,2]. It is based on the assumption that the delivery of definitive therapy to men whose tumour characteristics are too indolent to threaten their LE would be an overtreatment. Similarly, the definition of overtreatment also applies to patients whose LE is too short for the cancer to progress in a clinically meaningful way . Overtreatment might unnecessarily add to cost, complications, side-effects, early and late morbidity, and to treatment-related mortality [3–6]. In the context of definitive treatments, e.g. radical prostatectomy (RP), predicting the LE is meant to identify those men with a suboptimal LE (<10 years). These patients might be redirected towards other less definitive treatments [1,2,7]. One of the oldest and possibly most widely accessible tools for predicting LE are life tables (LT). However, to the best of our knowledge, the ability of LT to accurately predict LE has never been tested in patients treated with RP, and for practical purposes it can be considered unknown. This prompted us to test the accuracy of LT in a large population-based cohort of men treated with RP.
PATIENTS AND METHODS
In Canada, health care is under independent administrative control by its 11 provinces and territories. In the Province of Quebec, the Quebec Health Plan represents the exclusive insurer. Its database is used for billing purposes and allows virtually complete ascertainment of all health services and medications covered by the Plan that are provided across the Province. These also include all treatments for prostate cancer, e.g. RP, radiation therapy and all types of hormonal manipulation. The last consist of either medical androgen deprivation therapy with LHRH agonists, steroidal or nonsteroidal antiandrogens or bilateral orchidectomy. Moreover, the Health Plan relies on the 9th Version of the International Classification of Diseases (ICD-9) and the respective dates of all disease codes since 1 June 1983. These allow the Charlson Comorbidity Index (CCI) scores to be defined at the time of definitive therapy . The CCI can predict the risk of death according to existing comorbidities [9,10]. The D’Hoore adaptation of the CCI was used in the current analyses, as this adaptation was devised to obtain the CCI score based on ICD-9 codes . Survival was defined according to patient’s vital status in the Health Plan data file as of 30 June 2004.
The Health Plan database allowed us to identify all men diagnosed with prostate cancer (ICD-9185). We used the RP billing code (6243) to identify patients who had RP between 1 January 1989 and 31 December 2000. Each record included the date of RP, types and dates of any secondary therapy, age, ICD-9 codes of individual comorbidities and CCI score before RP. The Health Plan records contain no information on cancer stage, grade, preoperative PSA level or specific cause of death. To obtain the LT-derived LE at the time of RP, we used the 1995–1997 LT for male residents of the Province of Quebec, provided by Statistics Canada .
Overall, 9678 patients who had RP were identified and included in the present analyses. We repeated all analyses in a subgroup of men who, according to the Health Plan records, had no evidence of disease recurrence, based on lack of exposure to any secondary therapy after RP. In these men the cause of death was virtually certainly unrelated to prostate cancer. Secondary therapy was defined as either radiotherapy after RP or hormonal manipulation at any time before or after RP. This restriction criterion resulted in the inclusion of 5955 men who were exclusively treated with RP, as monotherapy. The rationale for repeating the analyses in men who virtually certainly did not die from prostate cancer was based on the consideration that prostate cancer mortality might spuriously undermine the ability of LT, as these are meant to provide a population average, where prostate cancer-specific mortality is not taken into consideration. The study was approved by the Provincial Review Board.
Statistical assessments consisted of Kaplan-Meier survival analyses and Cox-regression analyses addressing overall survival. The observed (Kaplan-Meier) and LT-predicted survival was plotted for the entire group of 9678 men, and for the subgroup of 5955 with no evidence of disease recurrence. In Cox-regression analyses, the accuracy of the LT-predicted LE in the whole group and the subgroup was quantified using Harrell’s concordance index . This index is expressed as a value of 50–100%, where 100% indicates perfect predictions and 50% is equivalent to the toss of a coin . Overfit bias was corrected with 200 bootstrap re-samples.
To assess the effect of age, the survival analyses were further stratified according to age categories, i.e. <60, 60–64, 65–69 and ≥70 years. To assess the effect of comorbidities, the analyses were stratified according to CCI score categories, i.e. 0, 1, 2 and ≥3. The goal of the stratified analyses was to explore whether potential discrepancies between LT-predicted and observed (Kaplan-Meier) survival were particularly related to either age or comorbidity strata. All tests were two-sided with a significance level set at 0.05.
The characteristics of all 9678 evaluable patients and of the restricted subgroup of 5955 are shown in Table 1.Figures 1 and 2 show the overall survival curves, plotted according to the Kaplan-Meier method, and to LT-predicted survival, respectively, for all 9678 men and for the subgroup of 5955 men. In the 9678 men (Fig. 1), the median actuarial survival was not reached (mean 12.4 years) and the 5-, 10- and 15-year actuarial survival probabilities were, respectively, 90.8%, 75.3% and 52.3%. The respective LT-predicted 5-, 10- and 15-year survival probabilities were 99.9%, 96.8% and 54.2%. The LT-predicted survival overestimated the observed survival by 9%, 22% and 2%, respectively, at 5, 10 and 15 years (log-rank P < 0.001). The accuracy of LT-predicted LE was 62.4% in the overall group.
Table 1. Descriptive characteristics of 9678 men treated with RP and of 5955 who had no evidence of prostate cancer recurrence after RP
NR, not reached.
Age at treatment, years
Median (mean, range)
65 (64.6, 45–89)
64 (64.2, 55–89)
LT-derived LE, years
Median (mean, range)
15 (15.7, 3.9–32.0)
16 (16.2, 3.9–23.2)
Median (mean, range)
7.2 (7.5, 0.1–15.5)
7.0 (7.4, 0.1–15.5)
Deaths, n (%)
Actuarial survival, years
Overall survival probabilities, %, at
In the 5955 men who were unexposed to any therapy other than RP, and who were presumed to be free of prostate cancer recurrence (Fig. 2), the median actuarial survival was not reached (mean 13.0 years), and the 5-, 10- and 15-year actuarial survival probabilities were, respectively, 92.0%, 81.1% and 61.9%. The respective LT-predicted 5-, 10- and 15-year survival probabilities were 99.9%, 97.3% and 59.4%. The differences between LT-predicted and observed overall survival were +8%, +16% and −3% at, respectively, 5, 10 and 15 years (log-rank P < 0.001). The accuracy of LT-predicted LE was 64.8% in the subgroup of 5955 men.
Figure 3 shows the overall survival curves plotted according to the Kaplan Meier method and to LT-predicted survival, respectively, for the age categories, for all 9678 patients. At 10 years the LT- predicted LE overestimated the observed (Kaplan-Meier) survival for the age categories of <60, 60–64, 65–69 and ≥70 years by, respectively, 13%, 19%, 27% and 22% (Table 2).
Table 2. Differences between LT-derived and observed (Kaplan-Meier) survival in all patients treated with RP; results are stratified according to age and CCI categories
Figure 4 shows the overall survival curves plotted according to the Kaplan- Meier method and to LT-predicted survival, respectively, for the CCI categories for all 9678 patients. At 10 years the LT-predicted LE overestimated the observed survival for CCIs of 0, 1, 2 and ≥3 by, respectively, 17%, 22%, 26% and 32% (Table 2).
Prostate cancer represents an important health burden [13–15], and consequently the related health expenditures represent an important part of the total health budget . The delivery of definitive therapy to men whose cancer characteristics are too indolent to threaten their LE represents overtreatment [3,16,17]. Similarly, treatment of a patient who has a clinically significant prostate cancer but whose LE is too short due to coexistent medical conditions also represents overtreatment . Here the cancer will not have enough time to progress to threaten the man’s LE, due to competing risks of death. In both scenarios, overtreatment might unnecessarily add to cost, complications, side-effects, early and late morbidity, and treatment-related mortality [3,5,18]. Thus, from the societal and individual perspectives, overtreatment should be avoided by preventing the definitive treatment of individuals whose disease is either too indolent or whose LE is insufficient. For definitive treatment for prostate cancer, this insufficiency in LE is defined as a LE of <10 years [1,2]. In circumstances where there is a risk of overtreatment, expectant management with delayed non-definitive intervention might represent a valid alternative to definitive therapy .
Therefore, considerations of LE are paramount in clinical decision-making, especially when the option of expectant management is considered. What are the options for estimating a patient’s LE? Age and comorbidity represent the main predictors that have been examined previously [8,9]. LT provide age, gender and average comorbidity-adjusted estimates of LE; they are usually specific to an administrative region and therefore reflect genetic, environmental and health-economic variables that are operational within that particular territory . Consequently, specific LT should be used for a given population. Despite the central role of LT in predicting LE, we are unaware of any studies that have quantified the ability of LT to predict LE, either in general or in patients with prostate cancer.
Besides LT, several investigators assessed the ability of comorbidity to predict LE. Various comorbidity indices were devised to approximate LE [8,19–22]. Of these, the CCI is one of the most frequently cited in medical publications [9,23–25]. Unfortunately, the accuracy of the available comorbidity indices has never been quantified in patients with prostate cancer. Albertsen et al. tested three different comorbidity indices in such patients, i.e. the Kaplan-Feinstein Index, the CCI and the Index of Coexisting Disease (ICED). All indices were highly significant predictors of death for patients dying from causes other than prostate cancer. However, no accuracy for these indices was provided . Moreover, Boulos et al. compared five different comorbidity indices in patients with prostate cancer, the Chronic Disease Score, the ICED, the Cumulative Illness Rating Scale, the Kaplan-Feinstein Index and the CCI, and found that the Chronic Disease Score and the ICED were the best age-independent predictors of observed survival. Unfortunately, neither of the investigators provided accuracy estimates for any of the tested comorbidity indices.
A few groups have looked beyond the contribution of age and comorbidity indices [24,26,27]; they combined in different ways the characteristics of the patient and the cancer to predict the man’s LE after treatment Tewari et al. relied on 1611 patients, using their age, race, CCI, PSA level, biopsy Gleason sum and treatment type (watchful waiting vs RP vs radiotherapy), and reported 69% accuracy in predicting LE. Albertsen et al. relied on 451 patients, and using age, comorbidity and biopsy Gleason sum, reported an accuracy for the LE predictions of 71%, in an external validation. Finally, Cowen et al. relied on 11 predictors that ranged from age through various medical conditions to biopsy Gleason sum, and reported 73% accuracy.
To the best of our knowledge, the present is the first to test the accuracy of LT; our results show that the LT overestimates the projected LE, as shown by a concordance index of 62.4–64.8%. Compared with multivariable models such as those described above, these accuracy estimates are inferior, regardless of the restriction or stratification criteria that were used. Consequently, LT estimates cannot be proposed for predicting LE in patients treated for localized prostate cancer with RP.
Notably, LT invariably overestimated the observed survival, regardless of whether all 9678 men or the restricted subgroup of 5955 men was examined. The rationale behind restricting to men with no evidence of disease relapse was based on the possibility that LT might not perform well in a population that could have a cancer-specific mortality that exceeds that of the general population, from which the LT are derived. Our hypothesis was correct, as LT gave better predictions in the subgroup of men with no disease relapse. However, the improvement relative to the overall group (64.8 vs 62.4%) was minor.
We also examined the effect of age on the performance of LT. As shown in Table 2 and Fig. 3, our results indicated that the most substantial overestimate of LE might be anticipated in older patients. When the 10-year LE predictions were examined, the 65–69 and ≥70 year categories, in that order, were exposed to the most severe 10-year LE overestimate (27–22%). Nonetheless, the extent of overestimation was still substantial in younger men, at 13–19% in men aged <60 and 60–64 years, respectively. Moreover, notably the LT substantially underestimated the survival of patients aged >70 years when survival beyond 12 years after RP was considered (Fig. 3d).
Finally, we examined the effect of comorbidity on the performance of LT, by stratifying between men with a CCI score of 0 vs 1 vs 2 vs ≥3 or higher. As shown in Table 2, our results indicated that the most substantial overestimate of LE might be anticipated in patients with the highest CCI score. The 10-year LE predictions were overestimated the most (26–32%) for men with a CCI score of 2 and ≥3. Nonetheless, the extent of overestimation was still substantial in patients with a CCI score of 0 or 1 (17–22%, respectively). Taken together, the age and comorbidity stratified analyses indicate that there are no safe age or comorbidity strata where the LT accurately predict a patient’s LE.
Despite the excessive enthusiasm for using LT to predict LE in this patient population, notably 75.3% of men from the overall population of 9678 patients and 81.1% of men from the restricted group of 5955 survived beyond 10 years. This implies that the practising urologists were 75% accurate in predicting the 10-year LE in the overall group and 81% accurate in predicting it in the subgroup with no evidence of disease recurrence. This finding implies that currently the clinical judgement, which reflects the complex decision-making process between the patient and the treating physician, and that encompasses several objective and subjective factors, represents the most valuable prediction of LE, as these results exceed the accuracy of LT (maximum accuracy 65%) and the accuracy of available multivariable models (maximum accuracy 73%). The data type assessed in this study makes it impossible to examine the potential causes of the significant rate of LE overestimation. Specifically, the administrative data do not allow us to examine or identify a specific variable that would predispose patients with localized prostate cancer to an earlier death.
The implications of our findings are important, as LT-predicted LE appears to overestimate the observed survival of patients treated with RP. This finding might indicate that, based on an inferior LE, in some men definitive surgical management might have not been necessary. This suggests that the use of LT might result in the overtreatment of localized prostate cancer. This finding requires further attention to assess the effect that excessively optimistic LE predictions might have on screening, early detection and treatment assignment.
Our study has some limitations; the lack of cause-specific mortality is one. Instead of restricting to patients with no clinical evidence of cancer recurrence we could have excluded those who died from prostate cancer, if that information was available. Moreover, we are well aware that treatment selection is not only based on LE and/or cancer characteristics. Quality-of-life considerations, patient and physician preferences and treatment availability all add to the complexity of treatment selection. Therefore, LE represents only one component in the complex process of making decisions about treatment for prostate cancer. Finally, our data only allowed us to examine LE in patients with prostate cancer. It is possible that the ability of LT to predict LE is better in patients without prostate cancer or in patients with other disease types. Further studies might be warranted to examine the performance of LT in other patient populations.
In conclusion, our findings indicate that LT are not an accurate predictor of LE in men treated with RP for localized prostate cancer. They invariably overestimate the LE, regardless of cancer control, age and comorbidity characteristics. Our data do not allow us to identify the specific causes of the excessive mortality. However, our results should be considered when screening, early detection and treatment selection are discussed, as LT-based overestimates of LE might contribute to overtreatment. Based on the central importance of LE in treatment considerations for prostate cancer, further work is urgently needed to improve our systematic, reproducible, approach to predicting LE. This area of research will become even more important as the population ages and as progressively more men present with prostate cancer in their sixth and seventh decades of life.
Pierre I. Karakiewicz is partially supported by the Fonds de la Recherche en Santé du Québec, the CHUM Foundation, the department of Surgery and les Urologues Associés du CHUM. JochenWalz was partially supported by the Grant of the Vereinigung Norddeutscher Urologen.