Back to 78th Annual Meeting
Back to Program Outline

Use of a Predictive Model for Discerning Clinically Significant Versus Indolent Prostate Cancer Prior to Prostate Biopsy
Stephen B. Williams, M.D.1, Meredith M. Regan, Sc.D.2, John Wei, M.D.3, Michael Kearney, M.D.4, Williams C. DeWolf, M.D.4, Jeffrey Tang, M.D.5, Gerry Bueti, B.S.4, Mark A. Rubin, M.D.5, Elizabeth Genega, M.D.4, Andrew Eyre, B.S.4, Martin G. Sanda, M.D.6.
1Brigham and Women's Hospital, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, USA, 2Dana Farber Cancer Institute, Harvard Medical School, Boston, MA, USA, 3University of Michigan, Ann Arbor, MI, USA, 4Beth Israel Deaconess Medical Center, Boston, MA, USA, 5Weill-Cornell Medical College, New York, NY, USA, 6Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, USA.

Background:
Predictive models have been developed to identify patients at risk for prostate cancer. There is an ever increasing need to discern clinically significant prostate cancer from indolent cancers, especially in light of recent PSA screening publications. Prediction tools to discern risk of aggressive vs. indolent cancer before a patient undergoes prostate biopsy are lacking. We sought to develop a predictive model to discern clinically significant prostate cancers from indolent prostate cancer based on pre-biopsy parameters.
Methods:
Patients undergoing prostate biopsy at 5 U.S. clinical sites were prospectively enrolled in the Harvard-Michigan NCI/Early Detection Research Network. Eligibility for analysis was limited to men with no prior prostate biopsy. Indolent prostate cancer was defined by Epstein’s histopathologic criteria (Gleason 6 or less and <50% of cores positive for cancer); all other cases were deemed clinically significant. A logistic regression model was developed and sensitivity, specificity, area underneath the receiver operating characteristic curve (AUC) and predicted probabilities were calculated.
Results:
635 men with no prior prostate biopsy who presented for initial biopsy were identified. Of these, 361 (57%) patients had no cancer while 106 (17%) and 168 (26%) patients were identified with indolent and clinically significant cancer, respectively. Age, family history, abnormal digital rectal exam (DRE) and log of prostate specific antigen (PSA) density (logPSAD) were significant predictors for clinically significant cancer (all p<0.001) and comprise the multivariable model. PSA and prostate size were each significant, however, the most predictive form of these variables was logPSAD. Predicted probabilities of the presence of clinically-significant prostate cancer on first prostate biopsy based on pre-biopsy clinical factors were calculated. AUC for the multivariable model was better than PSA alone in predicting clinically significant tumor (AUC=0.819 vs. 0.726). At a sensitivity level of 90%, the multivariable model improved specificity from 31% to 44% versus PSA alone.
Conclusions:
The model improved the specificity of PSA alone in detecting clinically significant prostate cancer and may be useful in counseling patients prior to prostate biopsy. Predicted probabilities provided may be helpful in identifying patients at high risk for clinically significant prostate cancer.
Predicted probabilities of the presence of clinically-significant prostate cancer on first prostate biopsy, for a range of demographic and pre-biopsy clinical factors. The values are presented as percentages with 95% CI.

Normal DREAbnormal DRE
AgeFamily HistoryBMI (kg/m2)PSA Density (ng/mL/cc)PSA Density (ng/mL/cc)
0.0660.1020.1570.0660.1020.157
55 yrsNo254.3
(2.7, 6.9)
8.3
(5.6, 12.2)
15.3
(10.8, 21.2)
13.2
(8.0, 21.2)
23.3
(15.2, 34.0)
37.6
(26.3, 50.6)
305.5
(3.6, 8.4)
10.4
(7.3, 14.6)
18.8
(13.8, 25.0)
16.3
(10.1, 25.2)
28.0
(18.8, 39.5)
43.6
(31.3, 56.7)
Yes259.7
(6.2, 14.9)
17.6
(12.1, 24.9)
29.8
(21.5, 39.7)
26.4
(16.5, 39.4)
41.8
(28.8, 56.0)
58.7
(44.1, 72.0)
3012.1
(8.1, 17.6)
21.5
(15.6, 29.0)
35.2
(26.6, 45.0)
31.5
(20.5, 45.0)
47.9
(34.3, 61.7)
64.6
(50.3, 76.6)
60 yrsNo255.7
(3.8, 8.7)
10.8
(7.8, 14.9)
19.4
(14.6, 25.3)
16.9
(10.8, 25.5)
28.9
(20.1, 39.6)
44.6
(33.3, 56.6)
307.2
(5.0, 10.4)
13.4
(10.1, 17.7)
23.6
(18.4, 29.6)
20.7
(13.6, 30.1)
34.2
(24.5, 45.5)
50.8
(38.8, 62.6)
Yes2512.5
(8.2, 18.6)
22.2
(15.8, 30.3)
36.2
(27.1, 46.4)
32.4
(21.3, 46.0)
48.9
(35.5, 62.6)
65.5
(51.7, 77.2)
3015.5
(10.7, 21.9)
26.8
(20.0, 34.9)
42.1
(32.8, 51.9)
38.0
(26.0, 51.8)
55.1
(41.6, 68.0)
70.9
(57.7, 81.3)
65 yrsNo257.5
(5.1, 11.0)
13.9
(10.3, 18.6)
24.3
(18.9, 30.7)
21.4
(14.2, 31.0)
35.2
(25.5, 46.2)
51.9
(40.4, 63.1)
309.4
(6.6, 13.2)
17.2
(13.2, 22.0)
29.2
(23.4, 35.6)
25.8
(17.6, 36.2)
41.0
(30.5, 52.3)
58.0
(46.2, 68.9)
Yes2516.1
(10.6, 23.6)
27.6
(19.8, 37.2)
43.1
(32.8, 54.1)
39.0
(26.3, 53.5)
56.1
(42.1, 69.3)
71.7
(58.5, 82.0)
3019.7
(13.6, 27.6)
32.8
(24.6, 42.3)
49.3
(38.9, 59.7)
45.0
(31.7, 59.2)
62.1
(48.3, 74.2)
76.5
(64.2, 85.4)

Values of age are approximately the 25th, 50th and 75th percentiles of the distribution; values of BMI are approximately the 25th and 75th percentiles of the distribution. Values of PSA density are the 25th, 50th and 75th percentiles of the distribution.
Abbreviations: BMI=body mass index; PSA=prostate specific antigen; DRE=digital rectal exam; CI=confidence interval
Back to 78th Annual Meeting
Back to Program Outline