Biochemical failure is a very common problem in the treatment of prostate cancer. Klotz has estimated that approximately 40% of men treated curatively by prostatectomy or radiotherapy will develop biochemical failure [
Prognostic data can provide assistance to clinical management practices if presented in the form of a nomogram or risk stratification scheme. Nomograms are of value to individual patients and their clinicians when determining the need for interventions. Risk stratification schemes, however, are more valuable in identifying patient subgroups who would benefit from participation in trials of new therapeutic approaches and for stratification purposes in these trials.
With trials of the effective new CRPC agents in mind, we sought to develop a risk stratification scheme to predict the outcome following biochemical failure using data from the TROG 96.01 randomized controlled trial which addressed the value of neoadjuvant androgen deprivation prior to and during radiotherapy for locally advanced prostate cancer. Earlier reports from the TROG 96.01 trial showed that PSA doubling time (PSADT) and time to biochemical failure (TTBF) were independent and highly prognostic variables after biochemical failure [
The trial was opened in 1996 following institutional ethical approval at 19 sites across Australia and New Zealand. All patients provided written informed consent. 802 eligible patients receiving radiotherapy (66 Gy delivered using 33 daily fractions of 2 Gy) for locally advanced prostate cancer were randomized to receive 0, 3, or 6 months maximal androgen suppression therapy (AST) prior to and during radiation. This comprised goserelin 3.6 mg (AstraZeneca Pty Ltd., Sydney, Australia) given subcutaneously every month and flutamide 250 mg (Schering-Plough Pty Ltd., Sydney, Australia) given orally three times daily. AST commenced 2 months prior to radiation in the 3-month arm and 5 months prior to radiation in the 6-month arm. Patients were stratified by age (<70 versus 70–80 versus >80 years), tumour stage (T2b and T2c versus T3 and T4), tumor differentiation (well versus moderate versus poor), and initial PSA level (<20 versus ≥20
Biochemical failure (BF) was defined according to the Phoenix definition (time from end of radiotherapy to a PSA rise of ≥2
Endpoints used in this paper were the cumulative incidences of PCSM, distant progression, and STI from BF. PCSM occurred at the time of death due to prostate cancer (attribution of cause validated in Lancet Oncology 2011 [
The analysis group consisted of 485 subjects who experienced BF prior to clinical failure or STI. Three-tier post-BF risk categorization (BFRC) schemes based on low, intermediate, and high risk groups were derived in two stages: (1) 12 “cut point range-finding” schemes were identified and evaluated using combinations of TTBF and PSADT cutpoints regularly cited in the literature as being predictive of outcome following BF and (2) new “candidate” BFRC schemes were derived based on the most prognostic ranges identified in the range-finding schemes. To ensure that they were unique and consisted of sizeable risk strata, candidate schemes had to satisfy three criteria: at least three months separation between the high and low risk PSADT; at least one year separation between the high and low risk TTBF; and a minimum of 20% of patients in each risk stratum. All BFRC schemes were evaluated in unadjusted regression models for PCSM from BF using the method of Fine and Gray [
Sensitivity analyses were undertaken to compare the performance rankings of the candidate schemes in a range of adjusted PCSM models. The first model adjusted for trial arm (0 versus 3 versus 6 months maximal AST) as duration of androgen suppression could influence outcome after BF. The second model adjusted for baseline factors as well as trial arm because these could determine the aggression of the relapse process. Additional covariates included age at BF (continuous, years), pretreatment PSA (<10 versus ≥10 and <20 versus ≥20
Competing risks methodology was used to calculate the cumulative incidences of distant progression, STI and PCSM. Competing risks were defined as STI, and death for distant progression; death for STI; and death due to other or unknown cause for PCSM. Univariable analyses were performed to determine the cumulative incidences of these endpoints in the three strata of the best BFRC and were compared using Gray’s test.
All analyses involving trial arms were conducted on an “intention to treat” basis and two-sided probability levels below 0.05 were considered significant. Analyses were performed using Stata Version 11.2.
As on 31 August 2010, 485 (60.5%) out of the 802 eligible subjects had experienced biochemical failure (BF) prior to clinical failure or STI. Of these, 343 (71%) received STI, 150 (31%) died due to prostate cancer, and 69 (14%) died due to other causes. Median follow-up time after BF was 5.6 years (IQR 3.1–8.0).
Table
Evaluation of 12 “range finding” post-biochemical failure risk categorization (BFRC) schemes in univariable competing risk models for prostate cancer-specific mortality (PCSM) after biochemical failure based on historical prognostic cutpoints for PSA doubling time and time to biochemical failure.
High risk† | Low risk† | BFRC performance | |||||
---|---|---|---|---|---|---|---|
PSADT | TTBF | PSADT | TTBF | c-index (95% CI) | Ranking‡ |
|
|
<3 | <1 | >9 | >3 | 0.724 | (0.684–0.764) | 1 | — |
<3 | <2 | >9 | >4 | 0.705 | (0.668–0.743) | 4 | 0.095§ |
<3 | <3 | >9 | >5 | 0.685 | (0.651–0.719) | 8 | 0.001 |
<3 | <1 | >12 | >3 | 0.720 | (0.680–0.760) | 2 | 0.489§ |
<3 | <2 | >12 | >4 | 0.695 | (0.657–0.733) | 5 | 0.026 |
<3 | <3 | >12 | >5 | 0.667 | (0.634–0.700) | 9 | <0.001 |
<6 | <1 | >15 | >3 | 0.714 | (0.680–0.749) | 3 | 0.27§ |
<6 | <2 | >15 | >4 | 0.691 | (0.658–0.725) | 6 | 0.033 |
<6 | <3 | >15 | >5 | 0.659 | (0.629–0.689) | 11 | <0.001 |
<9 | <1 | >18 | >3 | 0.690 | (0.656–0.723) | 7 | 0.003 |
<9 | <2 | >18 | >4 | 0.664 | (0.633–0.695) | 10 | <0.001 |
<9 | <3 | >18 | >5 | 0.630 | (0.603–0.658) | 12 | <0.001 |
PSA: prostate-specific antigen; BFRC: biochemical failure risk categorization; PSADT: PSA doubling time (months); TTBF: time from biochemical (Phoenix) failure (years); CI: confidence interval; c-index: Harrell’s concordance index.
†Risk is defined by PSADT and/or TTBF ranges specified.
‡Performance assessed by c-index, ranked highest (best) to lowest (worst). Performance against best BFRC compared using paired Student’s
§c-index not significantly lower than best BFRC.
Performance of 24 candidate post-biochemical failure risk categorization schemes (BFRC)* in univariable competing risk models for prostate cancer-specific mortality (PCSM) after biochemical failure.
BFRC model | High risk† | Low risk† | Intermediate risk‡ | BFRC model performance§ | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
PSADT | TTBF | HR | (95% CI) |
|
PSADT | TTBF | HR |
|
HR | (95% CI) | c-Index | (95% CI) | Rank |
|
|
1 | 97 | <3 | <1 | 6.1 | (3.8–9.8) | 231 | >9 | >4 | 1 | 157 | 2.7 | (1.8–4.3) | 0.724 | (0.685–0.763) | 7 | 0.32 |
2 | 113 | <3 | <1.5 | 5.1 | (3.3–7.7) | 267 | >6 | >4 | 1 | 105 | 2.7 | (1.8–4.2) | 0.719 | (0.678–0.760) | 8 | 0.28 |
3 | 129 | <3 | <1.5 | 5.6 | (3.5–9.0) | 222 | >9 | >4 | 1 | 134 | 2.7 | (1.7–4.3) | 0.718 | (0.679–0.757) | 9 | 0.15 |
4 | 163 | <3 | <2 | 5.2 | (3.2–8.5) | 212 | >9 | >4 | 1 | 110 | 2.8 | (1.6–4.7) | 0.705 | (0.668–0.743) | 13 | 0.013 |
5 | 166 | <3 | <2 | 5.1 | (3.2–8.3) | 221 | >12 | >3 | 1 | 98 | 2.9 | (1.7–4.9) | 0.697 | (0.658–0.736) | 17 | 0.007 |
6 | 171 | <3 | <2 | 7.3 | (3.9–13.7) | 183 | >12 | >4 | 1 | 131 | 3.9 | (2.1–7.5) | 0.695 | (0.657–0.733) | 19 | 0.004 |
7 | 119 | <4 | <1 | 5.7 | (3.7–8.8) | 246 | >9 | >3 | 1 | 120 | 2.6 | (1.7–4.2) | 0.732 | (0.695–0.769) | 1# | — |
8 | 119 | <4 | <1 | 7.0 | (4.3–11.3) | 225 | >12 | >3 | 1 | 141 | 3.1 | (1.9–5.1) | 0.730 | (0.693–0.767) | 2 | 0.70 |
9 | 120 | <4 | <1 | 8.2 | (4.8–14.1) | 208 | >24 | >3 | 1 | 157 | 3.5 | (2.0–6.0) | 0.729 | (0.692–0.767) | 3 | 0.68 |
10 | 188 | <4 | <2 | 8.1 | (4.2–15.5) | 180 | >12 | >4 | 1 | 117 | 3.7 | (1.9–7.3) | 0.705 | (0.669–0.741) | 14 | 0.019 |
11 | 196 | <4 | <2 | 7.4 | (3.8–14.7) | 163 | >18 | >4 | 1 | 126 | 2.9 | (1.4–5.9) | 0.697 | (0.661–0.733) | 16 | 0.005 |
12 | 196 | <4 | <2 | 8.0 | (3.9–16.4) | 159 | >24 | >4 | 1 | 130 | 3.1 | (1.4–6.5) | 0.697 | (0.660–0.733) | 18 | 0.005 |
13 | 150 | <5 | <1 | 7.6 | (4.4–13.0) | 201 | >24 | >3 | 1 | 134 | 3.4 | (1.9–6.0) | 0.727 | (0.691–0.762) | 4 | 0.55 |
14 | 149 | <5 | <1 | 5.2 | (3.4–7.9) | 239 | >9 | >3 | 1 | 97 | 2.5 | (1.5–4.1) | 0.726 | (0.691–0.762) | 5 | 0.41 |
15 | 149 | <5 | <1 | 6.3 | (3.9–10.3) | 218 | >12 | >3 | 1 | 118 | 3.1 | (1.8–5.1) | 0.726 | (0.690–0.762) | 6 | 0.45 |
16 | 201 | <5 | <2 | 8.4 | (4.3–16.6) | 173 | >12 | >4 | 1 | 111 | 3.9 | (1.9–7.9) | 0.700 | (0.665–0.735) | 15 | 0.007 |
17 | 209 | <5 | <2 | 7.8 | (3.8–15.9) | 156 | >18 | >4 | 1 | 120 | 3.0 | (1.4–6.4) | 0.693 | (0.658–0.728) | 20 | 0.002 |
18 | 209 | <5 | <2 | 8.5 | (4.0–18.2) | 152 | >24 | >4 | 1 | 124 | 3.2 | (1.4–7.2) | 0.693 | (0.658–0.728) | 21 | 0.002 |
19 | 141 | <6 | <1 | 5.0 | (3.3–7.7) | 235 | >18 | >2 | 1 | 109 | 2.4 | (1.5–3.9) | 0.714 | (0.676–0.753) | 10 | 0.18 |
20 | 184 | <6 | <1 | 6.8 | (3.9–11.8) | 190 | >24 | >3 | 1 | 111 | 2.7 | (1.5–5.0) | 0.713 | (0.679–0.748) | 11 | 0.07 |
21 | 140 | <6 | <1 | 4.7 | (3.1–7.1) | 242 | >12 | >2 | 1 | 103 | 2.4 | (1.5–3.8) | 0.712 | (0.674–0.751) | 12 | 0.12 |
22 | 220 | <6 | <2 | 7.7 | (3.9–15.0) | 167 | >12 | >4 | 1 | 98 | 3.3 | (1.6–6.9) | 0.691 | (0.657–0.725) | 22 | 0.001 |
23 | 228 | <6 | <2 | 7.0 | (3.5–14.3) | 150 | >18 | >4 | 1 | 107 | 2.4 | (1.1–5.3) | 0.685 | (0.651–0.719) | 23 | <0.001 |
24 | 228 | <6 | <2 | 7.7 | (3.6–16.4) | 146 | >24 | >4 | 1 | 111 | 2.6 | (1.1–6.0) | 0.685 | (0.651–0.719) | 24 | <0.001 |
PSA: prostate-specific antigen; BFRC: biochemical failure risk categorization;
*BFRC schemes presented include the best and worst three schemes for each high risk PSADT cutpoint from the 72 evaluable schemes.
†Risk is defined by PSADT and/or TTBF ranges specified.
‡PSADT and TTBF ranges are intermediate between the high and low risk ranges.
§Performance assessed by C-index, ranked highest (best) to lowest (worst). Performance against best BFRC compared using paired Student’s
#The best BFRC scheme.
¶A
Sensitivity analyses using multivariable models confirmed that our findings were not influenced by potentially important confounding covariables (data not shown). The rankings remained stable across all models, with the best BFRC in the unadjusted model also being the most predictive scheme in the models adjusting for trial arm (c-index 0.747), as well as for prognostic factors known at time of BF (c-index 0.751). In addition, the best BFRC in the unadjusted model was also the most predictive (c-index 0.744) in schemes derived using 12 months of PSAs post-BF to calculate PSADT instead of 6 months of PSAs.
Cumulative incidences of distant progression and STI for the best BFRC scheme are presented in Figures
Outcomes after biochemical failure for the most predictive post-treatment failure category (BFRC) stratification scheme. (a) Cumulative incidence of distant progression. (b) Cumulative incidence of secondary therapeutic intervention.
Mortality after biochemical failure for the most predictive post-treatment failure category (BFRC) stratification scheme. (a) Cumulative incidence of prostate cancer-specific mortality. (b) Cumulative incidence of other cause mortality.
Table
Pre- and post-treatment characteristics of the 485 subjects who developed biochemical (Phoenix) failure before clinical failure according to the most predictive post-biochemical failure risk category (BFRC) stratification scheme.
Post-biochemical failure risk category | ||||||
---|---|---|---|---|---|---|
Low | Intermediate | High | ||||
( |
( |
( |
||||
Gleason score | ||||||
2–6 | 117 (47%) | 36 (30%) | 23 (19%) | |||
7 | 105 (43%) | 53 (44%) | 54 (45%) | |||
8–10 | 24 (10%) | 31 (26%) | 42 (35%) | |||
| ||||||
T stage | ||||||
T2b | 73 (30%) | 23 (19%) | 16 (13%) | |||
T2c | 87 (35%) | 33 (28%) | 36 (30%) | |||
T3,4 | 86 (35%) | 64 (53%) | 67 (56%) | |||
| ||||||
PSA ( |
||||||
<10 | 45 (18%) | 24 (20%) | 20 (17%) | |||
≥10 and <20 | 102 (41%) | 38 (32%) | 24 (20%) | |||
≥20 | 99 (40%) | 58 (48%) | 75 (63%) | |||
| ||||||
Risk group* | ||||||
Intermediate | 45 (18%) | 10 (8%) | 4 (3%) | |||
High | 201 (82%) | 110 (92%) | 115 (97%) | |||
| ||||||
Primary treatment | ||||||
0 months AST | 107 (44%) | 45 (38%) | 40 (34%) | |||
3 months AST | 75 (30%) | 35 (29%) | 45 (38%) | |||
6 months AST | 64 (26%) | 40 (33%) | 34 (28%) | |||
| ||||||
Age at BF (years) | ||||||
Median (range) | 74 (54–89) | 69 (54–88) | 69 (44–81) | |||
| ||||||
PSADT (months) | ||||||
Median (IQR) | 13.2 (9.4–19.8) | 5.3 (4.3–6.4) | 3.0 (1.9–3.8) | |||
| ||||||
TTBF (years) | ||||||
Median (IQR) | 4.6 (3.4–7.3) | 2.2 (1.5–2.7) | 0.8 (0.5–1.4) | |||
| ||||||
STI | ||||||
No STI | 121 (49%) | 13 (11%) | 8 (7%) | |||
STI without distant progression | 89 (36%) | 61 (51%) | 47 (39%) | |||
STI with distant |
36 (15%) | 46 (38%) | 64 (54%) |
*D’Amico et al. risk classification.
This study has confirmed that TTBF and PSADT can be used to identify sizeable risk categories of men with poor, intermediate, and highly favourable outcomes after BF. The main strength of our study is that its findings are based on prospectively collected 10-year follow-up data from a randomised, clinical controlled trial. A further strength is the internal validation of the prognostic importance of the combination of the two variables. Our sensitivity analyses confirmed that the prognostic value of the combination was not influenced in multivariable models adjusting for treatment arm and other factors which could affect outcome, or in models using either 6 or 12 months of PSAs to estimate PSADT.
The optimal cutpoint ranges found to identify men at very high risk of early PCSM were PSADTs in the range <4 to <5 months and TTBF <1 year. These cutpoints were substantially lower than those found to be successful candidate surrogate endpoints for PCSM [
The best risk scheme identified in this study had a modestly predictive optimism-corrected c-index of 0.730 for PCSM after BF. The high risk category for this scheme comprised men with PSADT <4 months and/or TTBF <1 year. Within a year of biochemical failure cumulative incidences of distant progression and STI for this category were 49% and 77%, respectively. In spite of the early introduction of STI, PCSM at 5 years after BF was 45%. These data suggest that many of these high risk men had microscopic metastases at the time of BF. Had modern imaging advances been available at the time, it is quite possible some of these men could have had imaging evidence of macroscopic metastastic disease. In either event, men in the high risk fail category could have benefited from immediate inclusion in trials of the new agents effective against CRPC, had they been available.
Although not the specific intention of this study, the low risk stratum in our optimal risk scheme identified a sizeable subgroup of men who could be safely reassured that their prognosis is good enough to avoid STI for many years and possibly indefinitely. The cutpoints identified in this subgroup were PSADT >9 months and/or TTBF >3 years. In these men cumulative incidences of distant progression and PCSM at 5 years were only 17% and 4%, respectively. At this time point cumulative incidence of death due to causes other than prostate cancer was 10%. This is a potentially important finding because two randomized trials designed to determine the value of early STI after biochemical failure [
For comparative purposes we have presented prognostically significant variables at biochemical failure identified in the most recently updated studies published since 2000 [
Eleven studies of prognostic factors after biochemical failure since 2000*.
Authors | Type of Series | Number in series | Endpoint | Pretreatment variables | Variables at biochemical failure |
PSA at STI ( |
Other prognostic variables | ||
---|---|---|---|---|---|---|---|---|---|
Gleason score | Clinical stage | TTBF years | PSADT months | ||||||
Sandler et al. (2000) [ |
Single hospital, radiotherapy, retrospective | 154 | Survival |
NS | NS | — | Continuous | — | — |
D’Amico et al. (2002) [ |
Single hospital, radiotherapy, retrospective | 381 | Survival |
— | — | NS | ≤12 | >10 | — |
Moul (2003) [ |
Community database, prostatectomy | 1352 | Survival |
>7 | — | ≤1 | ≤12 | ≤5, ≤10 | — |
Okotie et al. (2004) [ |
Single hospital, prostatectomy, retrospective | 126 | Metastases | — | — | — | <6 | — | — |
Stephenson et al. (2007) [ |
Multihospital, prostatectomy and radiotherapy, retrospective | 1540 | PSA progression | >7 | Positive margins | — | ≤10 | Continuous | — |
Freedland et al. (2005) [ |
Single hospital, prostatectomy, retrospective | 379 | Survival |
>7 | — | ≤3 | <3 |
— | — |
Dotan et al. (2005) [ |
Single hospital, prostatectomy, retrospective | 239 | Bone metastases | — | — | — | Continuous | Continuous |
— |
Slovin et al. (2005) [ |
Single hospital, prostatectomy, retrospective | 148 | Metastases | >7 | ≥T3 | — | Continuous | Continuous | — |
Zhou et al. (2005) [ |
Community databases, prostatectomy and radiotherapy, retrospective | 1159 | Survival |
>7 |
— | NS | <3 | — | — |
Tollefson et al. (2007) [ |
Single hospital, prostatectomy, retrospective | 1064 | Survival (overall) metastases | — | — | — | <12 |
— | — |
Buyyounouski et al. (2008) [ |
Single hospital, radiotherapy, |
248 | Survival |
> 6 (metastases only) | — | <1.5 | NS | — | PSA nadir <2 |
NS: nonsignificant; TTBF: time to biochemical failure; PSADT: PSA doubling time; STI: secondary therapeutic intervention; PC: prostate cancer.
*Only the most recent update of each series is included in the table.
†Review article.
Predictive value of the prognostic stratification schemes in Table
Author | Variable | Cutpoints, mortality, number of patients | Cutpoints, Mortality, Number of patients | c-index* | ||||||
---|---|---|---|---|---|---|---|---|---|---|
PSADT | ≥6 months | <6 months | 0.675 | |||||||
Okotie et al. [ |
|
|
||||||||
|
|
|||||||||
| ||||||||||
D’Amico et al. [ |
PSADT | ≥12 months | <12 months | 0.598 | ||||||
Tollefson et al. [ |
|
|
||||||||
|
|
|||||||||
| ||||||||||
GS | ≤7 | >7 | ||||||||
TTBF | >3 years | ≤3 years | >3 years | ≤3 years | 0.694 | |||||
Freedland et al. [ |
PSADT | ≥9 months | <9 months | ≥9 months | <9 months | ≥9 months | <9 months | ≥9 months | <9 months | |
|
|
|
|
0% | 0% |
|
|
|||
|
|
|
|
|
|
|
|
|||
| ||||||||||
GS | ≤7 | >7 | 0.654 | |||||||
Zhou et al. [ |
PSADT | ≥3 months | <3 months | ≥3 months | <3 months | |||||
|
|
|
|
|||||||
|
|
|
|
|||||||
| ||||||||||
Stephenson et al. [ |
GS | ≤7 | >7 | 0.647 | ||||||
PSADT | >10 months | ≤10 months | >10 months | ≤10 months | ||||||
|
|
|
|
|||||||
|
|
|
|
|||||||
| ||||||||||
Buyyounouski et al. [ |
GS | ≤6 | >6 | 0.682 | ||||||
TTBF | ≥1.5 years | <1.5 years | ≥1.5 years | <1.5 years | ||||||
|
|
|
|
|||||||
|
|
|
|
*Harrell’s concordance index (higher c-index indicates that the model has a better predictive power).
†Due to subject numbers, subsets with PSADT <3 months and >15 months are not presented in the Freedland study.
A few limitations of our study need to be acknowledged. Firstly, it is a secondary retrospective study not prespecified in the trial protocol. However, the disciplined prospective collection of data in the context of a randomized trial, as in this study, does avoid many of the unseen selection biases that exist in most retrospective clinical studies. Secondly, the radiation dose used in the trial (66 Gy) was low by modern standards [
Although we performed internal validation of our findings in this paper, external validation in a wider range of clinical scenarios is important. These datasets should comprise men with different initial risk profiles, and who have undergone a wider range of curative treatments than used in this study, for example, prostatectomy alone in earlier stage disease, and long-term AST and radiation in later stage disease. In recommending external validation, however, we need to caution that the proportion of men with low risk disease prior to primary treatment who develop “high risk” biochemical failures according to our definition is likely to be very small. In our dataset only 4 (6.8%) of 59 men with intermediate risk cancer who developed BFs were classified as high risk. In those with low risk disease treated by prostatectomy, other prognostic factors, such as Gleason score, margin status, seminal vesicle, or nodal involvement at prostatectomy, may assume greater prognostic importance and might need inclusion for a risk categorization scheme to be effective. We suspect therefore that men experiencing BF after radiation and long-term AST for LAPC will be most likely to derive benefits from a risk stratification based on PSADT and TTBF.
Finally, if risk stratification schemes based on TTBF and PSADT derived shortly after biochemical failure are to be reproducible, there is a need for international consensus on the most appropriate means of calculating PSADT within months of biochemical failure and ensuring that TTBF is measured accurately [
This study has shown that time to biochemical failure and PSA doubling time can be combined to define risk stratification schemes after biochemical failure in men with locally advanced prostate cancer treated with short-term androgen suppression therapy and radiotherapy. External validation of these stratification schemes is necessary, particularly in datasets evaluating long-term androgen suppression therapy and radiotherapy.
A. Steigler received financial support from AstraZeneca to attend a meeting. D. Lamb received financial support from AstraZeneca to attend meetings. D. Joseph received honoraria associated with membership of AstraZeneca’s Breast Cancer Medical Advisory Board. N. Spry received honoraria associated with AstraZeneca and Schering Plough. K.-H. Tai received FROGG-AstraZeneca education grant in 2003 and was supported by AstraZeneca to attend two Casodex Investigators Meetings in 2001 and 2004.
The authors declare that they have any conflict of interests.
The authors acknowledge trial funding by the Australian Government National Health and Medical Research Council (Project Grant Applications 9936572, 209801, and 455520); Hunter Medical Research Institute (Newcastle, Australia); AstraZeneca Pty Ltd. and (Sydney, Australia); and Schering-Plough Pty Ltd., Sydney, Australia. Ms. Rosemary Bradford is thanked for her skillful preparation of the paper.