1
TIME-DEPENDENT PREDICTIVE ACCURACY OF PROSTATE CANCER SPECIFIC MORTALITY OF THE CONTEMPORARY vs.ORIGINAL GLEASON SCORE
Maule Milena1, Baldi Ileana2, Delsedime Luisa3, Grasso Chiara1, Fiano Valentina1, Zugna Daniela1, Merletti Franco1, Richiardi Lorenzo1
1
Cancer Epidemiology Unit, University of Turin, Italy
2
Department of Environmental Medicine and Public Health, University of Padua, Italy
3
Department of Pathology, San Giovanni Battista Hospital, Turin, Italy
Introduction. Prostate cancer is the most frequent male cancer in the United States (US) and Western Europe1. Classification of newly diagnosed cancers to assess prognosis and decide treatment is particularly important for this type of cancer because of the large variability in its progression. Gleason score is the most important prognostic indicator. A number of studies have observed an upward shift in Gleason score from the early 1990s to the early 2000s2 and the criteria to assign the Gleason score have been further revised in 20053. It is expected that the contemporary Gleason score has an enhanced utility to guide treatment decisions. Whereas it has been described how such a shift has produced a spurious improvement in survival for most recent cases due to stage migration2, to our knowledge the desired improvement in predictive ability has not been assessed.
Objectives. The aims of this study were to evaluate to what extent the upgrade in Gleason score has artificially improved stage-specific survival and to assess if the contemporary Gleason score is more accurate in predicting prostate cancer mortality than the original Gleason score.
Methods. The diagnostic slides of 243 patients diagnosed with prostate cancer between 1993 and 1996 were traced and re-evaluated in 2010 by a pathologist who assigned a new Gleason score based on the classification criteria currently in use. The effect of stage migration of patients due to the upward shift in Gleason score was assessed by estimating the cumulative probability of death from prostate cancer taking into account death from other causes as a competing risk event4. Time-dependent measures of the accuracy of the original and contemporary Gleason scores in predicting the probability of mortality due to prostate cancer were assessed by computing the areas under (AUC) the receiver operating characteristics (ROC) curves at different follow-up times, in a setting with censored survival and competing risks5. The AUCs were compared through a Wilcoxon rank sum test for dependent samples6.
Results. The median age of the 243 prostate cancer patients at diagnosis was 70.6 years (standard deviation: 8.7), and the median survival time was 6.3 (inter-quartile range: 2.6-14). After 14 years of follow-up, 76 patients had died of prostate cancer and 101 died of other causes. The mean Gleason score increased from 6.1, 95%CI 5.8-6.4, (original readings) to 7.2, 95%CI: 7.1-7.3, (contemporary readings; p-value<0.0001). The decline in the proportion of low-grade prostate cancers after re-classification results in an apparent reduction of stage-specific probability of death (not shown). ROC curves show the accuracy of the original and contemporary Gleason scores in predicting the probability of death due to prostate cancer at different follow-up times in Figure 1. AUCs are moderate for both the original and the contemporary Gleason score readings, ranging from a minimum of 64% (contemporary score, 5-year mortality) to a maximum of 77% (original score, 1-year mortality). AUCs are higher for the original than the contemporary score both after 1 and 5 years since cancer diagnosis, whereas after 10 years of follow-up the contemporary score outperforms the original one (p-value for test for difference of AUCs < 0.0001).
Figure 1. Time dependent predictive accuracy of Gleason scores measured as the area under curve (AUC) of the receiver operating characteristics (ROC) curves at 1, 5 and 10 years after prostate cancer diagnosis.
2
Conclusions. Contemporary Gleason scores are higher than the original ones, artificially improving survival, but do not have better predictive accuracy.Bibliography
1. Ferlay J, Shin HR, Bray F, Forman D, Mathers C, Parkin DM. GLOBOCAN 2008, Cancer Incidence and Mortality Worldwide: IARC CancerBase No. 10 [Internet]. Lyon, France: International Agency for Research on Cancer. Available from: http://globocan.iarc.fr, 2010.
2. Albertsen PC, Hanley JA, Barrows GH, Penson DF, Kowalczyk PD, Sanders MM, Fine J. Prostate cancer and the Will Rogers phenomenon. J Natl Cancer Inst 2005;97:1248-53.
3. Epstein JI, Allsbrook WC, Jr., Amin MB, Egevad LL. The 2005 International Society of Urological Pathology (ISUP) Consensus Conference on Gleason Grading of Prostatic Carcinoma. Am J Surg Pathol 2005;29:1228-42.
4. Gooley TA, Leisenring W, Crowley J, Storer BE. Estimation of failure probabilities in the presence of competing risks: new representations of old estimators. Stat Med 1999;18:695-706.
5. Saha P, Heagerty PJ. Time-dependent predictive accuracy in the presence of competing risks. Biometrics 2010;66:999-1011.
6. Haibe-Kains B, Desmedt C, Sotiriou C, Bontempi G. A comparative study of survival models for breast cancer prognostication based on microarray data: does a single gene beat them all? Bioinformatics 2008;24:2200-8.
