Oncological outcomes in an Australian cohort according to the new prostate cancer grading groupings

The South Australian Prostate Cancer Clinical Outcomes Collaborative (SA-PCCOC) database is a long running prospective clinical registry which collects tumour characteristics, treatment details and oncological and functional outcome data for men with PCa across both the public and private sector in South Australia [19]. The study sample included all men in the SA-PCCOC registry with localised PCa diagnosis between 2006 and 2013 who had biopsy Gleason grade patterns recorded. During this period, registry coverage was approximately 50% of all cases in the state and included recruitment from all public hospitals, which are government run with universal access for all Australians, as well as approximately 50% of private urologists/urology services. Diagnoses before 2006 were excluded to limit cases to those graded after ISUPs revision of the grading system in 2005. Men with evidence of metastatic disease (clinical or imaging) at or within 45 days of diagnosis were also excluded, since metastatic disease may distort outcome assessment by grade.

Data on patient characteristics including age at diagnosis, public or private health care management, place of residence; clinical features including grade, prostate specific antigen [PSA] levels, stage, and symptomatic presentation (i.e. referral due to symptoms - i.e. lower urinary tract symptoms, haematuria, bone pain – versus referral for elevated PSA), primary and subsequent treatment modalities, and dates of biochemical recurrence and death were extracted from SA-PCCOC for eligible cases. An area level measure of socioeconomic status was derived from patient’s residential postcode, using the Australian Bureau of Statistics Index of Socioeconomic Advantage and Disadvantage [20]. Death data were obtained from both the South Australian Register of Births, Deaths, and Marriages and the National Death Index.

For analyses of outcomes among men receiving curative treatment, we restricted the cohort to men who received curative RP or RT within 12 months of diagnosis. RT included external beam radiotherapy (EBRT), brachytherapy, or a combination of both. Conservative management was defined as management via watchful waiting (WW), active surveillance (AS) or androgen deprivation therapy (ADT) alone.

Grade at diagnosis, grouped according to the recently endorsed five-tiered system [8], was the key variable of interest in this study. For comparative purposes only biopsy grade was considered across all treatment groups including radical prostatectomy. Key outcomes in this study were prostate cancer-specific survival (PCSS) and biochemical recurrence-free survival (BRFS). PCSS was defined as the time from diagnosis to death, where PCa was indicated on the death certificate as a primary contributing cause of death. BRFS was defined as the time from date of diagnosis to first evidence of biochemical recurrence (BCR) among men who underwent definitive treatment. BCR was defined for patients receiving RP as two consecutive PSA values of >0.2 ng/mL [21], and for those receiving primary radiation therapy, any PSA increase >2 ng/mL higher than the post-RT PSA nadir value, regardless of the serum concentration of the nadir [22]. Survival durations were calculated from the date of diagnosis until the date of BCR, death or censoring date of June 30, 2016 (i.e. most recent deaths/PSA update), which ever was earliest.

Descriptive analyses of demographic, clinical and treatment characteristics according to grade groups were undertaken, with extended Wilcoxon rank-sum tests used to assess trends across ordered groups. Survival outcomes were initially assessed using Kaplan-Meier methods with log rank tests for differences in survival by grade groups. For Kaplan-Meier curves and log rank analyses competing risks are censored.

PCSS and BRFS were also compared across biopsy grade groupings (I to V) using univariable and multivariable competing risk regression, according to Fine and Gray’s methodology [23], with death from causes other than PCa as the competing risk. We undertook analyses for the entire cohort as well as for separate treatment subgroups: conservative management, RP and curative RT. All regression models controlled for age at diagnosis (continuous), year of diagnosis (continuous), public or private healthcare management, closest preceding PSA level to diagnosis (<10, 10- < 20, 20 + ng/ml), clinical stage (<cT3 v cT3+), symptomatic presentation (yes/no), and where appropriate, specific treatment types (e.g. robot-assisted versus open surgery, brachytherapy versus EBRT, ADT) and total dose received in Grays (continuous) for RT patients. Wald’s test was used to test for significant trends across grade groups in multivariable models. The potential for pairwise interactions between grade and other baseline factors was explored using likelihood ratio tests, comparing nested models with and without interactions. Statistically significant interactions were observed for treatment approaches and grade, in relation to both PCa mortality (p = 0.03), and BCR (p < 0.001). We therefore report results of subgroup analyses for different treatment modalities.

Due to known inaccuracies in assessing grade at biopsy, a sensitivity analyses was also undertaken for the subset who underwent RP, comparing the discriminatory power of prostatectomy versus biopsy grade to predict biochemical recurrence via the Akaike Information Criteria (AIC) in separate multivariable models.

Statistical analyses were undertaken using Stata v 12.1 [24].

Data were available for a total of 4268 men, diagnosed between 2006 and 2013. Two thirds of patients were classified as grade group I or II on biopsy, according to the new grade groupings. The mean age at diagnosis increased with increasing grade, as did median PSA at diagnosis (see Table 1). The proportion of men presenting with symptoms at diagnosis also increased with increasing grade groupings. However, grade groups did not differ with respect to number of cores taken at biopsy (median = 12). In the case of grade group V, a higher proportion was managed in the public system compared with other grade groups.

As expected, treatment patterns varied considerably across grade groups. RP was the primary treatment in 40% of men with grade group I and 51% with grade group II, but only 14% for men with grade group V disease. RT as the primary treatment was less variable. The proportion receiving RT with curative intent ranged from 27% for grade group I to 38% for grade group III and 39% for grade group V. Twenty seven percent of men in grade group I and 10% in grade group II underwent observation without immediate treatment.

Five-year PCSS decreased from 98% (95% CI 97–99%) for grade group I to 64% (95% CI 59–69%) for grade group V. Five-year BRFS decreased from 91% (95% CI 89–93) among men in grade group I to 67% (95% CI 57–75) for grade group V. The median follow-up time for the whole cohort was 72 months (inter-quartile range 52–96 months).

Figure 1 presents Kaplan-Meier PCa-specific survival curves for each of the new five-tier grade groupings I – V for the whole cohort. Survival decreased incrementally with higher grade grouping, as expected. Table 2 presents results of unadjusted and covariate adjusted competing risk regressions for PCa mortality across grade groups. Adjustment for covariates attenuated the effect of grade on risk of PCa death, in some instances quite considerably. This is likely to be explained by associations with other prognostic factors (age and PSA levels) across grade groups. Using backwards elimination modelling we confirmed that differences in age and pre-treatment PSA levels were the main factors contributing to the attenuation effect. Even so, grade remained the strongest independent predictor of death from PCa.

Figure 2 presents survival curves for BCR by grade groups for patients treated curatively, indicating poorer outcomes with increasing grade group. Results from competing risk regression analyses for BCR by grade groups are shown in Table 3. These analyses show a clear gradient in risk of BCR across grades following RP (p for trend < 0.001). For patients who received curative RT, we did not observe a constant increase in sHRs across grade groups, though the trend overall was statistically significant (<0.001). Including type of RT (EBRT versus brachytherapy), total dose and concurrent or adjuvant ADT in the model did not alter this pattern. Likewise, findings did not change when men receiving neo-adjuvant or adjuvant ADT were excluded.

To specifically compare outcomes between incremental grade groups we repeated multivariable competing risk regression models with each grade group referenced to the previous grade grouping (Table 4). With respect to PCa death we observed non-significant trends toward higher risk for Grade group III compared with II among men undergoing radical prostatectomy (sHR = 1.4, CI 0.3-6.5) and men receiving curative radiotherapy (sHR = 1.6, CI 0.9–3.1). The lack of statistical significance is likely to be due to the low number of PCa deaths among men in these grade groupings. With respect to BCR, comparisons showed statistically significant higher risk of progression for grade group III compared with grade group II for both treatment groups (RP: sHR = 2.1, CI 1.5–2.8; RT: sHR = 2.0, CI 1.3–3.1). Comparisons of grade groups IV and grade group III indicated statistically significant differences for risk of PCa death for men undergoing RP (sHR = 4.2, CI 1.0–17.1) and men managed conservatively (sHR = 1.9, CI 1.0–3.4) but not for men undergoing curative RT, and no difference in relation to risk of BCR for either curative approach. Risk of PCa death and BCR were both elevated for men in grade group V compared with IV for RP and RT patients, but the difference only reached statistical significance in relation to BCR for men undergoing RP. Among men who were managed conservatively we observed a significant difference in risk of PCa death for grade group V compared grade group IV (sHR = 2.1, CI 1.3–3.4), but saw no difference in risk death between grade groups II and III).

Comparison of biopsy and prostatectomy grade groups among men who underwent radical prostatectomy indicated that prostatectomy grade was marginally superior to biopsy grade in predicting BCR following RP, as indicated by the difference in AIC for the two models (Table 5).

In undertaking this study, we did not commission a review of the original grade assignment but rather reclassified groupings based on recorded primary and secondary Gleason patterns. Nor did we include tertiary pattern 5 in grade classification, as it was not always recorded.

Also, we were unable to account for potential confounding by factors such as comorbidity and frailty, due to a lack of information on these measures. Comorbidity/frailty may be contributing to poorer clinical outcomes for men in the lower grade groups, particularly those undergoing radiotherapy or conservative management due to being unfit for surgery. Not being able to account for these factors may have obscured the influence of grade, leading to less clear incremental effect across grade groups in these treatment subgroups. This is supported by the smaller effect sizes for grade among men receiving RT compared to those observe for the RP subgroup.

Since state-wide coverage was only 50% and private patients were likely underrepresented in SA-PCCOC during the study period, our results may be affected by selection bias. Given public patients are likely to be older and have more comorbidities and, or more advanced disease, the likely impact of such bias would be toward reduced strength of association between grade groups and clinical outcomes. This may be another factor explaining the less clear distinction in men receiving RT or managed conservatively.

The strengths of this study in relation to assessing applicability of the new grading classification in a community setting are a relatively long follow-up time and multi-institutional nature of our cohort.