DOI:10.1093/jnci/djt319 Advance Access publication November 22, 2013

© The Author 2013. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: [email protected].

Review

Androgen Receptor Expression and Outcomes in Early Breast Cancer: A Systematic Review and Meta-Analysis Francisco E. Vera-Badillo, Arnoud J. Templeton, Paulo de Gouveia, Ivan Diaz-Padilla, Philippe L. Bedard, Mustafa Al-Mubarak, Bostjan Seruga, Ian F. Tannock, Alberto Ocana, Eitan Amir Manuscript received June 6, 2013; revised September 16, 2013; accepted September 18, 2013. Correspondence to: Alberto Ocana, MD, PhD, Medical Oncology Department and Translational Research Unit, Albacete University Hospital, Edificio de Investigación, Calle Francisco Javier de Moya, 02006 Albacete, Spain (e-mail: [email protected]).

Background

The androgen receptor (AR) is expressed frequently in breast cancer, but its prognostic significance is unclear. Preclinical data suggest that expression of AR may modify clinical outcomes in early breast cancer with improved prognosis in estrogen receptor (ER)–positive disease and poorer prognosis in ER-negative disease.



Methods

A systematic review of electronic databases was conducted to identify studies published between 1946 and July 2012 and to explore the association between AR expression and overall survival (OS) and disease-free survival (DFS) in women diagnosed with early breast cancer. The odds ratios (OR) for OS and DFS at 3 and 5 years were calculated and then weighted and pooled in a meta-analysis with Mantel–Haenszel random-effect modeling. All statistical tests were two-sided.



Results

Nineteen studies with a total of 7693 women were included. AR expression was documented in 60.5% of patients. ER-positive tumors were more likely to express AR- than ER-negative tumors (74.8% vs 31.8%, χ2 P < .001). Compared with tumors without AR expression, those expressing AR were associated with improved OS at both 3 and 5 years (OR = 0.47, 95% confidence interval [CI] = 0.39 to 0.58, P 5 fmol/mg Nuclear R1881 New England ≥10 fmol/mg Nuclear R1881 ≥43 fmol/mg 2470 arrayer Aushon log mean centered Biosystems value ≥ 0.0852

AR411 DAKO (mouse) AR27 Novocastra (mouse) AR 441 DAKO F39.3 (mouse)

NR NR

NR

Nuclear NR

Nuclear

NR

Nuclear

Nuclear

Nuclear

Cytoplasmatic and/ or membranous NR NR NR

Nuclear

Nuclear Nuclear

NR Nuclear

Location

Definition of AR+

137 (51) 174 (50)

24 (39)

212 (43) 398 (47)

116 (54)

51 (58)

41 (56)

128 (57)

38 (28)

609 (71) 83 (75) 358 (53)

237 (72)

541 (58)

1155 (79) 128 (84)

16 (64) 212 (53)

AR+, No. (%)

* AR+ = androgen receptor positive; ER+ = estrogen receptor positive; ER− = estrogen receptor negative; IHC = immunohistochemistry; NR = not reported; RIA = radioimmunoassay; RPPA = reverse-phase protein array; TNBC = triple negative breast cancer.

61

Langer et al. 1990 (52)

488 54.3 (32–79) 840 NR

Agrawal et al. 2008 (22) Bryan et al. 1984 (51)

55 (24–87)

215

Peters et al. 2009 (13)

88 54.9 (26–91)

Agoff et al. 2003 (50)

54 (30–94)

73

Peters et al. 2012 (49)

49 (25–80)

NR NR NR

327 52.5 (NR)

ER+/ER−

ER+/ER− ER+/ER−

Yu et al. 2011 (43)

61 (39–75) 55 (29–88)

931 49.3 (NR)

1467 153

Hu et al. 2011 (4) Kuenen-Boumeester et al. 1996 (42) Park et al. 2010 (24)

IHC ≥10% IHC ≥10%

Carreno et al. 2007 (41) Honma et al. 2012 (23)

ER+/ER− ER+ only

25 403

References

61 (27–92) 56 (32–81)

Age, median Method and Follow up, Technique for No. (range) Population definition of ER+ months AR assessment

Table 1.  Characteristics of studies included in the meta-analysis*

A Study Bryan 1984 Castellano 2010 Collett 1996 Gonzalez 2008 Gonzalez-Angulo 2009 Honma 2012 Hu 2011 Langer 1990 Loibl 2011 Luo 2010 Park 2010 Peters 2009 Peters 2012 Yu 2011 Total (95% CI)

Weight , %

Absolute risk difference (95% CI)

Odds ratio (95% CI)

26.5 4.9 4.9 4.5 12.2 3.3 8.1 1.4 11.3 6.9 9.9 2.5 2.6 0.8

-0.11 (-0.16 to -0.05) 0.00 (-0.02 to 0.03) -0.03 (-0.10 to 0.03) -0.28 (-0.49 to -0.08) -0.15 (-0.24 to -0.06) -0.03 (-0.07 to 0.01) -0.03 (-0.06 to -0.01) -0.08 (-0.24 to 0.08) -0.05 (-0.09 to -0.00) -0.09 (-0.17 to -0.00) -0.02 (-0.05 to 0.00) -0.16 (-0.24 to -0.07) -0.13 (-0.31 to 0.05) -0.22 (-0.43 to -0.01)

0.54 (0.38 to 0.75) 1.06 (0.44 to 2.56) 0.64 (0.27 to 1.56) 0.25 (0.10 to 0.64) 0.41 (0.24 to 0.70) 0.44 (0.15 to 1.30) 0.31 (0.16 to 0.61) 0.47 (0.09 to 2.55) 0.57 (0.32 to 0.99) 0.46 (0.22 to 0.96) 0.59 (0.32 to 1.08) 0.12 (0.03 to 0.42) 0.42 (0.12 to 1.43) 0.17 (0.02 to 1.55)

100.0

-0.07 (-0.10 to -0.04)

0.47 (0.39 to 0.58)

Heterogeneity: χ² = 14.00, (P = .37); I ² = 7% Test for overall effect: Z = 7.29 (P < .001)

0.01

0.1 1 10 100 Favors AR+ Favors AR-

B Study Agoff 2003 Castellano 2010 Gonzalez 2008 Gonzalez-Angulo 2009 Honma 2012 Kuenen-Boumeestar 1996 Loibl 2011 Micello 2010 Park 2010 Yu 2011 Total (95% CI)

Weight , %

Absolute risk difference (95% CI)

Odds ratio (95% CI)

2.8 13.3 2.5 17.4 7.4 5.1 21.2 3.3 20.0 7.1

-0.36 (-0.60 to -0.13) -0.05 (-0.09 to -0.01) -0.11 (-0.26 to 0.05) -0.19 (-0.29 to -0.10) -0.07 (-0.13 to -0.02) -0.22 (-0.43 to -0.01) -0.11 (-0.17 to -0.05) -0.04 (-0.11 to 0.02) -0.08 (-0.12 to -0.04) -0.02 (-0.10 to 0.06)

0.19 (0.06 to 0.65) 0.47 (0.27 to 0.81) 0.36 (0.10 to 1.28) 0.38 (0.24 to 0.62) 0.38 (0.18 to 0.80) 0.36 (0.15 to 0.87) 0.44 (0.28 to 0.68) 0.49 (0.16 to 1.47) 0.42 (0.27 to 0.65) 0.81 (0.38 to 1.73)

100.0

-0.09 (-0.12 to -0.05)

0.43 (0.35 to 0.52)

Heterogeneity: χ² = 5.04, (P = .83); I ² = 0% Test for overall effect: Z = 8.28 (P < .001)

0.01

0.1 1 10 Favors AR+ Favors AR-

100

Figure 3.  Patient outcome at 3 years by androgen receptor (AR) expression. A) Three-year overall survival. B) Three-year disease-free survival. Odds ratios in reference to AR expression compared with no AR expression for each trial are represented by the squares, the size of the square represents the weight of the trial in the meta-analysis, and the horizontal line

crossing the square represents the 95% confidence interval (CI). The diamonds represent the estimated pooled effect (labeled total). Estimates of odds ratios were weighted and pooled using the Mantel–Haenszel random-effect model. Statistical heterogeneity was assessed using Cochran’s Q and I2 statistics. All P values are two-sided.

in 5-year DFS of 20.7% (95% CI = 8.7% to 32.7%). The association of AR expression and improved 5-year DFS was once again independent of coexpression of ER, with similar magnitudes of effect observed for studies including only ER-positive patients (OR = 0.41; 95% CI = 0.20 to 0.83), those including only ER-negative patients (OR  =  0.39; 95% CI  =  0.13 to 1.16), and those with patients unselected for ER expression (OR  =  0.27; 95% CI = 0.11 to 0.71). There was no statistically significant difference between these subgroups (P = .79).

Sensitivity Analyses Removal of studies using radioimmunoassay or reverse-phase protein array did not influence results for 3-year or 5-year OS (OR = 0.43, 95% CI = 0.31 to 0.60; OR = 0.37, 95% CI = 0.25 to 0.55, respectively). Similarly, removal of the single study (13) that was an outlier with regard to definition of AR expression by IHC (75% vs 1%–10% for other studies) did not substantially affect the association between AR expression and favorable 3-year and 5-year OS compared with no AR expression (OR = 0.48, 95% CI = 0.36 to

6 of 11 Review | JNCI

Vol. 106, Issue 1 | djt319 | January 1, 2014

Study or subgroup ER-positive Castellano 2010 Gonzalez-Angulo 2009 Hu 2011 Loibl 2011 Luo 2010 Park 2010 Park 2012 Peters 2009 Subtotal (95% CI)

Weight, %

Odds ratio (95% CI)

6.1 3.7 5.5 4.2 4.5 6.7 6.0 0.8 37.4

1.06 (0.44 to 2.56) 0.66 (0.19 to 2.22) 0.23 (0.09 to 0.58) 0.42 (0.13 to1.29) 0.45 (0.15 to 1.33) 0.45 (0.20 to 1.05) 0.39 (0.16 to 0.96) 0.03 (0.00 to 0.55) 0.45 (0.29 to 0.69)

1.5 0.8 4.5 1.5 4.8 2.9 15.9

0.07 (0.01 to 0.55) 0.10 (0.01 to 1.82) 1.00 (0.34 to 2.95) 0.12 (0.02 to 0.95) 1.49 (0.53 to 4.18) 0.38 (0.09 to 1.56) 0.38 (0.13 to 1.10)

16.6 6.1 5.7 4.4 2.1 3.6 7.0 1.3 46.7

0.54 (0.38 to 0.75) 0.64 (0.27 to 1.56) 0.25 (0.10 to 0.64) 0.44 (0.15 to 1.30) 0.47 (0.09 to 2.55) 0.42 (0.12 to 1.43) 0.31 (0.14 to 0.70) 0.17 (0.02 to 1.55) 0.47 (0.36 to 0.61)

100.0

0.45 (0.35 to 0.58)

Total events Heterogeneity: χ² = 9.43, (P = .22); I ² = 26% Test for overall effect: Z = 3.66 (P < .001) ER-negative Gonzalez-Angulo 2009 Hu 2011 Loibl 2011 Luo 2010 Park 2010 Peters 2009 Subtotal (95% CI) Heterogeneity: χ² = 13.60, (P = .02); I ² = 63% Test for overall effect: Z = 1.78 (P = .08) Unselected for ER status Bryan 1984 Collett 1996 Gonzalez 2008 Honma 2012 Langer 1990 Peters 2012 Schippinger 2006 Yu 2011 Subtotal (95% CI) Total events Heterogeneity: χ² = 4.59, (P = .71); I ² = 0% Test for overall effect: Z = 5.76 (P < .001) Total (95% CI) Test for subgroup differences: χ² = 0.14, (P = .93)

0.01

0.1 1 10 100 Favors AR+ Favors AR-

Figure  4.  Three-year overall survival (OS) by androgen receptor (AR) expression based on coexpression of estrogen receptor (ER). Odds ratios in reference to AR expression compared with no AR expression for each trial are represented by the squares, the size of the square represents the weight of the trial in the meta-analysis, and the horizontal line crossing the square represents the 95% confidence interval (CI). The diamonds

represent the estimated pooled effect (labeled total). Estimates of odds ratios were weighted and pooled using the Mantel–Haenszel randomeffect model. Statistical heterogeneity was assessed using Cochran’s Q and I2 statistics. Test of subgroup differences relates to the test of heterogeneity between the three cohorts of ER expression as defined by Deeks et al. (21). All P values are two-sided.

0.62; OR = 0.36, 95% CI = 0.23 to 0.55, respectively). Exclusion of these studies did not reduce heterogeneity for 5-year OS (Cochran’s Q P < .001, I2 = 75%; Cochran’s Q P < .001, I2 = 77%, respectively). Among studies using IHC, use of different cutoffs for AR overexpression was not statistically different; the association between AR overexpression and outcome was similar for both the 1% and 10% thresholds, and the test for subgroup difference did not meet statistical significance for either 3-year or 5-year OS (P = .62 and P = .09, respectively). Similarly, use of difference cutoffs for ER expression was not associated with a statistically significant test for

subgroup difference for either 3-year or 5-year OS (P  =  .51 and P = .32, respectively). Corresponding authors of papers using the IHC cutoff of ≥10% for AR positivity were contacted to retrieve any available data using the ≥1% cutoff. Responses were received from three authors (4,23,24). For one study, 5% of the study cohort had AR expression between 1% and 9%, increasing the proportion of patients considered positive from 58% to 63% (Byeong-Woo Park and Seho Park, personal communication). Using the 1% rather than 10% cutoff for AR led to marginally different values of median OS

jnci.oxfordjournals.org

JNCI | Review 7 of 11

A Study Agrawal 2006 Carreño 2007 Castellano 2010 Collett 1996 Gonzalez 2008 Gonzalez-Angulo 2009 Honma 2012 Hu 2011 Loibl 2011 Luo 2010 Park 2010 Peters 2009 Peters 2012 Yu 2011 Total (95% CI)

Weight, %

Absolute risk difference (95% CI)

8.7 3.0 8.5 8.1 5.8 9.1 7.3 9.9 9.3 4.4 9.2 7.8 6.0 3.0

-0.41 (-0.48 to -0.34) -0.13 (-0.53 to 0.26) -0.06 (-0.10 to -0.02) -0.04 (-0.13 to 0.05) -0.21 (-0.41 to -0.02) -0.14 (-0.24 to -0.04) -0.07 (-0.13 to -0.02) -0.05 (-0.09 to -0.01) -0.06 (-0.12 to -0.01) -0.18 (-0.31 to -0.06) -0.04 (-0.07 to -0.00) -0.13 (-0.24 to -0.02) -0.18 (-0.41 to 0.05) -0.44 (-0.69 to -0.18)

Odds ratio (95% CI) 0.09 (0.05 to 0.16) 0.57 (0.11 to 3.07) 0.37 (0.21 to 0.66) 0.74 (0.39 to 1.39) 0.30 (0.11 to 0.82) 0.49 (0.30 to 0.81) 0.38 (0.18 to 0.80) 0.61 (0.42 to 0.88) 0.60 (0.38 to 0.95) 0.24 (0.07 to 0.85) 0.60 (0.37 to 0.97) 0.47 (0.24 to 0.91) 0.48 (0.19 to 1.24) 0.11 (0.02 to 0.60)

100.0

-0.14 (-0.20 to -0.07)

0.40 (0.29 to 0.56)

Heterogeneity: χ² = 46.47, (P < 0.001); I² = 72% Test for overall effect: Z = 5.27 (P < .001)

0.01

0.1 1 10 100 Favors AR+ Favors AR-

B Study Agoff 2003 Agrawal 2006 Castellano 2010 Gonzalez 2008 Gonzalez-Angulo 2009 Honma 2012 Hu 2011 Loibl 2011 Luo 2010 Micello 2010 Park 2010 Yu 2011 Total (95% CI)

Weight, %

Absolute risk difference (95% CI)

Odds ratio (95% CI)

6.0 9.0 8.9 7.1 8.8 8.4 9.0 9.0 8.5 7.8 9.0 8.5

-0.39 (-0.63 to -0.16) -0.30 (-0.38 to -0.21) -0.06 (-0.11 to -0.01) -0.30 (-0.50 to -0.09) -0.23 (-0.33 to -0.13) -0.71 (-0.78 to -0.64) -0.07 (-0.11 to -0.03) -0.06 (-0.13 to 0.00) -0.13 (-0.24 to -0.03) 0.01 (-0.08 to 0.11) -0.05 (-0.09 to -0.00) -0.25 (-0.36 to -0.14)

0.17 (0.05 to 0.57) 0.27 (0.18 to 0.40) 0.58 (0.38 to 0.90) 0.23 (0.09 to 0.60) 0.35 (0.22 to 0.56) 0.03 (0.01 to 0.05) 0.50 (0.34 to 0.72) 0.71 (0.49 to 1.02) 0.50 (0.28 to 0.88) 1.10 (0.52 to 2.30) 0.66 (0.45 to 0.96) 0.27 (0.15 to 0.47)

100.0

-0.21 (-0.33 to -0.09)

0.34 (0.21 to0.56)

Heterogeneity: χ² = 124.89, (P < .001); I² = 91% Test for overall effect: Z = 4.29 (P < .001)

0.01

0.1 1 10 100 Favors AR+ Favors AR-

Figure 5.  Five-year patient outcome by androgen receptor (AR) expression. A) Five-year overall survival. B) Five-year disease-free survival. Odds ratios in reference to AR expression compared with no AR expression for each trial are represented by the squares, the size of the square represents the weight of the trial in the meta-analysis, and the

horizontal line crossing the square represents the 95% confidence interval. The diamonds represent the estimated pooled effect (labeled total). Estimates of odds ratios were weighted and pooled using the Mantel– Haenszel random-effect model. Statistical heterogeneity was assessed using Cochran’s Q and I2 statistics. All P values are two-sided.

and DFS but did not change the statistical significance. Of the two other authors who responded, one did not have any further information available (Rong Hu, personal communication), and another described that the association between AR and outcome was smaller, but was unable to provide updated data (Naoko Honma, personal communication). Among studies using IHC, there did not appear to be any difference between those where assessment was made using nuclear

compared with nonnuclear staining. The odds ratios for 3-year OS for nuclear vs nonnuclear assessment were 0.42 (95% CI  =  0.30 to 0.59) and 0.17 (95% CI  =  0.02 to 1.55), and the test for subgroup difference was 0.43. Similar data were seen for 5-year OS (OR = 0.55, 95% CI = 0.44 to 0.70 vs OR = 0.35, 95% CI = 0.19 to 0.62; subgroup difference P = .14). Although the definition of early vs metastatic breast cancer has been relatively standard over time, between 1987 and 2002,

8 of 11 Review | JNCI

Vol. 106, Issue 1 | djt319 | January 1, 2014

patients with supraclavicular nodal metastases at diagnosis of breast cancer were classified as having metastatic disease (25). Before and after this time, these patients were considered to have early breast cancer. Exclusion of studies published during this time did not affect the association between AR-expression with either 3-year or 5-year OS (OR  =  0.46, 95% CI  =  0.36 to 0.58; OR  =  0.38, 95% CI  =  0.27 to 0.54, respectively). Furthermore, there was no apparent improvement in heterogeneity for 5-year OS (Cochran’s Q P