Predictors of recurrence after robotic prostatectomy for pT3b disease

Introduction

Prostate cancer is the most prevalent type of cancer among men in the European region and is globally reported as one of the most frequently occurring cancers. ¹ The diagnosis is achieved via prostate-specific antigen (PSA) levels, a digital rectal exam, multiparametric magnetic resonance imaging (mpMRI) of the prostate, and a biopsy of the prostate. ² The prognosis and treatment options are dictated by the combination of these factors.

Non-organ-confined (locally advanced) prostate cancers are treated in a multimodality fashion, usually involving the combination of radiotherapy and androgen deprivation therapy. Recent advances in treatment have involved the combination of these older therapies with abiraterone and demonstrated even greater survival improvement in these groups.^3,4 Surgical intervention is yet another primary treatment option in these patients, and several large multi-institutional studies have shown that radical prostatectomy is capable of obtaining adequate local control in the majority of patients with high-risk locally advanced disease.^5,6 The remaining gap in knowledge around the treatment of this group of patients is mainly attributed to the lack of a proper understanding of the heterogeneous nature of high-risk locally advanced prostate cancer and the pathological mechanisms of cancer recurrence.^7,8

The invasion of the seminal vesicle (T3b) suggests more advanced disease, greater risk of micrometastatic disease, and a more accelerated course of disease that is more likely to recurrence via germinal pathways. ⁹ The reported recurrence rates of organ confinement following radical prostatectomy (pT3b) are between 45% and 65%, and to date, no ideal therapy has been proposed. ¹⁰ It is important to pinpoint the pathological mechanisms of recurrence in this group of patients to determine strategies for stricter postoperative observation and the indication for adjuvant therapy or salvage therapy thereafter.

The role of mpMRI in the clinical diagnosis and cancer staging of prostate cancer has become well understood and integrated. In embracing mpMRI into the analysis of the prostate for the evaluation of prostate cancer and its variants, the Prostate Imaging Reporting and Data System (PI-RADS) utilises T2-weighted, diffusion-weighted, and dynamic contrast-enhanced (DCE) sequences to assess the likelihood of significant disease.^11,12 Studies^13,14 have shown mpMRI is likely to over-stage disease in comparison to its final histopathologic correlations, and prior meta-analyses have demonstrated the need for continued clarification for accuracy in the staging of seminal vesicle invasion (SVI) as many patients are downgraded after final pathology, and results are inconsistent across multiple centres. ¹⁵

The primary objective of this study is to determine the pre- and postoperative factors influencing the likelihood of biochemical recurrence (BCR) of prostate cancer following robotic-assisted laparoscopic prostatectomy (RALP) for the cancer stage of pT3b. This will be achieved by conducting a survival analysis using the time-to-event method. An additional objective of this study is to determine the reliability of mpMRI in the preoperative staging of cancer as an indicator of SVI.

Methods

Data were collected from the West of Scotland database of men who underwent RALP (December 2015 to August 2023) for those who qualified for pT3b status for retrospective study. The parameters analysed were preoperative patient characteristics such as age, Bosar score, PSA, size, and postoperative outcomes such as size of prostate, surgical margins, lymphadenopathy, and pathological grading.

An mpMRI was obtained from eight facilities that use both 1.5T and 3T magnets at minimum and include diffusion-weighted imaging (DWI), DCE, and T2 sequences. The acquisition protocols (including b-value and slice thickness) varied by site, and there was no standardisation of parameter settings. All mpMRI studies were reported by the uroradiologists at the referring centre and were reviewed by the senior uroradiologist before the surgical planning was done at our institution.

We performed a retrospective study on mpMRI imaging correlative to preoperative PSA and Gleason grading data on patients who had cT3b prostate cancer. Patients were stratified according to upstaging and downstaging. We also performed a matched stratification for T stage based on the outcomes. The final analysis was on the sensitivities and specificities of the various mpMRI sequences. The sequences compared were T2-weighted, DWI, apparent diffusion coefficient (ADC), and DCE. Sensitivities and specificities that pertain to the pathological outcome were analysed through the lens of the above scenarios. Exact Binomial confidence intervals (CIs) at the 0.05 level were computed.

We defined BCR as PSA positivity (PSA ⩾ 0.1 ng/mL) on two PSA tests per the norm found in the medical literature. ¹⁶ We defined the interval to BCR as the time from the date of RALP to the date of BCR. Patients who did not have BCR at the end of follow-up were censored at the time of the final PSA. We also excluded patients from the study for having incomplete or inconsistent time-to-event data.

Summary statistics for patients were presented as median and interquartile range (IQR) for the continuous variables (age, PSA level, and prostate volumetrics) and as a percentage for the categorical variables (lymph node status, Grade Group, and status of surgical margin) and IQR. We compared the groups by using the Mann–Whitney U test for the continuous variables and by the chi-square test (Fisher’s exact test, as appropriate) for the categorical variables.

BCR-free survival was estimated using the Kaplan–Meier method, and groups were compared using the log-rank method. To reveal the predictors of BCR, both univariate and multivariate Cox proportional hazards regression were applied. The results were displayed as hazard ratios (HRs) and 95% confidence intervals (CIs). All the candidate variables were considered for the multivariate model, regardless of the univariate analysis results. Multivariate analysis excluded patients whose lymph nodes had not been evaluated.

Multivariate analysis was conducted on the data, as it was available; missing data were excluded list-wise. Regarding lymph nodes, patients whose status was not evaluated were categorised separately. All statistical tests were two-tailed. A value of p < 0.05 was defined as statistically significant. The statistical analysis was conducted using Python (lifelines, pandas, scipy).

Results

A total of 1587 patients underwent RALP between December 2015 and August 2023, of whom 149 had pT3b prostate cancer on final pathology. Three patients were excluded because of duplicate or inconsistent records, leaving 146 patients for descriptive clinicopathological analysis. Of these, 76 patients had documented BCR and 70 had no documented BCR. A further seven patients were excluded from time-to-event analysis because of missing or inconsistent follow-up or recurrence timing data, leaving 139 patients for Kaplan–Meier and Cox regression analysis. In this survival-analysis cohort, 73 patients developed BCR and 66 were censored at last follow-up.

Patient demographics and tumour characteristics

The median age for the overall cohort was 64 years (IQR = 59–70). Median PSA at diagnosis was 13.0 ng/mL (IQR = 8.0–20.4), and median prostate size was 44 mL (IQR = 34–58). Positive surgical margins were present in 67 patients (45.9%), while positive lymph nodes were detected in 68 patients (46.6%); in 29 patients (19.9%), lymph nodes were not assessed. The distribution of International Society of Urological Pathology (ISUP) Grade Groups was: Grade Group 2, 41 patients (28.1%); Grade Group 3, 58 (39.7%); Grade Group 4, 12 (8.2%); and Grade Group 5, 35 (24.0%) (Table 1).

OPEN IN VIEWER

Table 1.

Baseline demographics and tumour characteristics (n = 146).

Variable	Overall (n = 146)
Age, years	64 (59-70)
PSA, ng/mL	13.0 (8.0–20.4)
Prostate size, mL	44 (34-58)
Grade Group, n (%)
2	41 (28.1)
3	58 (39.7)
4	12 (8.2)
5	35 (24.0)
Positive surgical margins, n (%)	67 (45.9)
Lymph node status, n (%)
Positive	68 (46.6)
Negative	49 (33.6)
Not assessed	29 (19.9)

Values are median (IQR) unless stated otherwise.

Comparison between BCR and no BCR groups

There were no statistically significant differences in median age (No BCR: 64.0, IQR = 59.0–70.0 vs BCR: 65.0, IQR = 60.8–69.0; p = 0.730), PSA (11.8, IQR = 8.0–20.4 vs 13.4, IQR = 8.1–19.7; p = 0.613), or prostate size (43.0, IQR = 34.0–55.0 vs 46.5, IQR = 34.1–62.5; p = 0.247) between the groups. Grade Group distribution differed significantly between groups (p = 0.012), with BCR patients having higher proportions of Grade Group 4 (14.5% vs 1.4%) and Grade Group 5 disease (27.6% vs 20.0%). Positive surgical margins were significantly more frequent in the BCR group (59.2% vs 31.4%, p < 0.001). Among patients in whom lymph nodes were assessed, positive nodal disease was more prevalent in the BCR group (70.5% vs 44.6%, p = 0.008) (Table 2).

OPEN IN VIEWER

Table 2.

Comparison of clinicopathological variables by BCR status in the descriptive cohort (n = 146).

Variable	No BCR (n = 70)	BCR (n = 76)	p-value
Age, years	64.0 (59.0–70.0)	65.0 (60.8–69.0)	0.730
PSA, ng/mL	11.8 (8.0–20.4)	13.4 (8.1–19.7)	0.613
Prostate size, mL	43.0 (34.0–55.0)	46.5 (34.1–62.5)	0.247
Grade Group, n (%)			0.012
2	24 (34.3)	17 (22.4)
3	31 (44.3)	27 (35.5)
4	1 (1.4)	11 (14.5)
5	14 (20.0)	21 (27.6)
Positive margins, n (%)	22 (31.4)	45 (59.2)	<0.001
Lymph node status, a n (%)			0.008
Positive	25 (44.6)	43 (70.5)
Negative	31 (55.4)	18 (29.5)

Continuous variables are median (IQR), tested with Mann–Whitney U; categorical variables are n (%), tested with chi-square.

a

Lymph node comparison excludes 29 patients in whom nodes were not assessed.

Predictors of BCR

The Kaplan–Meier analysis demonstrated an overall median BCR-free survival of 31.4 months, with estimated 12-month and 24-month BCR-free survival rates of 65.8% and 55.4%, respectively (Table 3 and Figure 1).

OPEN IN VIEWER

Table 3.

Kaplan–Meier BCR-free survival (BFS) estimates by subgroup.

Subgroup	N	12-month BFS	24-month BFS	Median BFS (months)
Overall	139	65.8%	55.4%	31.4
Negative margins	75	80.2%	71.0%	41.7
Positive margins	64	49.3%	37.6%	10.5
Negative nodes	47	72.1%	69.4%	Not reached
Positive nodes	64	62.2%	46.8%	19.0
Grade Group 2–3	94	69.8%	62.6%	41.7
Grade Group 4–5	45	57.5%	40.7%	16.4

BFS: BCR-free survival. Node-stratified analysis includes only patients in whom lymph nodes were assessed (n = 111). Log-rank p-values: margins p = 0.0006; nodes p = 0.030; Grade Group p = 0.009.

OPEN IN VIEWER

Figure 1.

Kaplan-Meier analysis of predictors of BCR.

BCR-free survival differed significantly by surgical margin status: patients with positive margins had a median BCR-free survival of 10.5 months compared with 41.7 months for those with negative margins (log-rank p = 0.0006). Similarly, patients with positive lymph nodes had significantly shorter BCR-free survival than those with negative nodes (median 19.0 months vs not reached; log-rank p = 0.030). Patients with Grade Group 4–5 disease had a median BCR-free survival of 16.4 months compared with 41.7 months for Grade Group 2–3 (log-rank p = 0.009) (Figure 1).

On univariate Cox regression analysis within the 139-patient survival-analysis cohort, higher Grade Group (HR = 1.27, 95% CI = 1.03–1.55; p = 0.022), positive surgical margins (HR = 2.24, 95% CI = 1.40–3.59; p < 0.001), and positive lymph nodes (HR = 1.84, 95% CI = 1.05–3.21; p = 0.033) were each significantly associated with an increased hazard of BCR. Age, PSA, and prostate size were not significant predictors on univariate analysis.

On multivariate Cox regression analysis including 106 patients with complete data for all covariates, positive surgical margins remained the only independent predictor of BCR (HR = 2.20, 95% CI = 1.18–4.09; p = 0.013). The multivariable model included 106 patients; 33 were excluded because 28 had no nodal assessment and 5 had missing preoperative data. Positive lymph node status approached but did not reach statistical significance (HR = 1.68, 95% CI = 0.94–3.00; p = 0.081), and Grade Group was not independently predictive after adjustment (HR = 1.16, 95% CI = 0.90–1.50; p = 0.248). Age, PSA, and prostate size were not independently associated with recurrence. The overall model was statistically significant (likelihood ratio test p = 0.005; concordance index 0.68) (Table 4).

OPEN IN VIEWER

Table 4.

Cox proportional hazards regression analysis of predictors of biochemical recurrence.

Variable	n	Ev	UV HR	UV 95% CI	MV HR	MV 95% CI	UV p	MV p
Age (per year)	139	73	1.02	0.99–1.06	1.04	1.00–1.09	0.253	0.078
PSA (per ng/mL)	137	73	1.01	0.99–1.02	1.00	0.99–1.02	0.280	0.610
Prostate size (per mL)	132	71	1.01	1.00–1.02	1.00	0.99–1.01	0.065	0.820
Grade Group (per unit)	139	73	1.27	1.03–1.55	1.16	0.90–1.50	0.022	0.248
Positive margins	139	73	2.24	1.40–3.59	2.20	1.18–4.09	<0.001	0.013
Positive nodes a	111	59	1.84	1.05–3.21	1.68	0.94–3.00	0.033	0.081

UV: univariate; MV: multivariate; Ev: events; HR: hazard ratio; CI: confidence interval. The multivariate model includes 106 patients with complete data for all covariates. Bold values with shading indicate statistical significance (p < 0.05). The multivariate model included 106 patients with complete data for all six covariates (33 excluded: 28 with nodes not assessed, 5 with missing preoperative data). In the 146-patient descriptive cohort, lymph nodes were not assessed in 29 patients. Within the 139-patient survival-analysis cohort, 28 patients had no lymph node assessment and were excluded from node-specific Cox and Kaplan–Meier analyses.

a

Univariate analysis for nodes excludes 28 patients in whom lymph nodes were not assessed.

mpMRI accuracy in cT3b prostate cancer

Fifty-three patients were staged as cT3b on mpMRI. On final pathology, 30 patients (56.6%) were confirmed as pT3b, 20 (37.7%) were downstaged to pT3a, two (3.8%) were upstaged to pT4, and one (1.9%) was downstaged to pT2. Neither preoperative PSA (logistic regression, p = 0.972) nor Gleason grading (chi-square, p = 0.355) was significantly associated with concordance between clinical and pathological staging.

Among individual MRI sequences, T2-weighted imaging demonstrated the highest sensitivity for SVI (96.6%, 95% CI = 82.8–99.4%) but the lowest specificity (26.1%, 95% CI = 12.5–46.5%). DWI offered the highest specificity (65.2%, 95% CI = 44.9–81.2%) but the lowest sensitivity (48.3%, 95% CI = 31.4–65.6%). Wide CIs for all estimates reflect the limited sample size of this exploratory sub-analysis (Table 5).

OPEN IN VIEWER

Table 5.

Sensitivity and specificity of mpMRI sequences for detecting seminal vesicle invasion (cT3b).

Sequence	Sensitivity	95% CI	Specificity	95% CI
T2-weighted	96.6%	82.8–99.4%	26.1%	12.5–46.5%
DWI	48.3%	31.4–65.6%	65.2%	44.9–81.2%
ADC	75.9%	57.9–87.8%	60.0%	38.7–78.1%
DCE	50.0%	33.2–66.8%	33.3%	15.2–58.3%

CI: confidence interval; DWI: diffusion-weighted imaging; ADC: apparent diffusion coefficient; DCE: dynamic contrast enhancement. 95% CIs calculated using the Wilson score method. n = 53 patients staged cT3b on mpMRI.

Discussion

In this study, 52.5% of patients with pT3b prostate cancer experienced BCR following RALP, with a median time to recurrence of only 3.6 months. This high early recurrence rate is consistent with the established prognostic significance of SVI as an indicator of poor oncological outcomes. ⁹ The Kaplan–Meier analysis demonstrated a median BCR-free survival of 31.4 months, with estimated 12-month and 24-month BCR-free survival rates of 65.8% and 55.4%, respectively. These outcomes are broadly consistent with published series reporting BCR rates of 45–65% for pT3b disease, including the large multi-institutional analysis by Briganti et al.,^5,6 which evaluated over 2,400 patients with high-risk and locally advanced prostate cancer across several European centres and demonstrated that radical prostatectomy provides effective local control even in adverse pathology subgroups. Our findings add to this evidence base from a contemporary UK robotic series with prospectively maintained data.

When patients who did and did not develop BCR postoperatively were compared, age, PSA, and prostate size were either not or less significant in distinguishing BCR in the patients studied. In comparison, high Grade Group, positive surgical margins, and positive lymph node involvement were shown to be associated with BCR in the study population via univariate analysis, followed by prognostic separation in the Kaplan–Meir analysis in the same population on single Cox regression. Using multivariate analysis within the study, positive surgical margins were the first and most significant predictor of BCR, with an HR of 2.2 (p = 0.013). Positive lymph node involvement was associated with an HR of 1.68 with a p-value of 0.081, a marginally insignificant trend, and high Grade Group; in contrast, it did not hold significance with an HR of 1.16 and a p-value of 0.248. The effect of the Grade Group and lymph node involvement seen in the multivariate analysis was likely attributed to the marginal prognostic value on their own in combination with positive surgical margins and the low statistical power of the complete case analysis (n = 106). The significance of the Kaplan–Meir with log-rank p = 0.009 for Grade Group and p = 0.030 for nodule status respectively upholds the value of their individual clinical contributions to the study cohort.

These findings carry direct clinical implications for postoperative management. First, patients with pT3b disease and positive surgical margins should be considered for early salvage or adjuvant therapy rather than surveillance alone, given the markedly reduced BCR-free survival in this group (median 10.5 vs 41.7 months). Second, the lack of association between preoperative PSA, age, or prostate volume and BCR suggests that these conventional parameters should not be used to withhold surgery from otherwise suitable candidates with suspected locally advanced disease. Third, the identification of pathological rather than preoperative variables as the key drivers of recurrence reinforces the importance of definitive histopathological assessment in guiding postoperative decision-making.

The treatment landscape for locally advanced prostate cancer continues to evolve. Landmark trials including STAMPEDE and PEACE-1 have demonstrated significant survival benefits with the addition of abiraterone to androgen deprivation therapy and radiotherapy in high-risk and locally advanced disease.^3,4 These systemic intensification strategies represent an important alternative to surgery, particularly for patients with the highest-risk profiles. Our data may help inform shared decision-making between surgical and non-surgical approaches: patients in whom adverse pathological features are anticipated preoperatively—particularly those at high risk of positive margins—may derive greater benefit from combined systemic and radiation-based strategies, while surgery remains a viable option for appropriately selected patients even in locally advanced disease.^7,8

We also analysed the diagnostic capability of mpMRI in an exploratory sub-analysis of 53 patients staged as cT3b preoperatively. In the final pathology, 37.7% of patients were staged to pT3a and 3.8% of patients were staged to pT4. Poststaging demonstrates an example of mpMRI’s overstaging complexity due to limitations. As determined by previous meta-analyses, mpMRI demonstrates significant variance in staging accuracy metrics across the known centres.^13–15 Furthermore, the variance suggested could be in the domain of pT3a and pT3b in the disease, which could be of fundamental importance to new treatment paradigms, as the more aggressive focused treatment could be justified.

Evaluation of individual mpMRI sequences demonstrated that T2-weighted imaging offered the highest sensitivity (96.6%, 95% CI = 82.8–99.4%) but the lowest specificity (26.1%, 95% CI = 12.5–46.5%) for SVI, whereas DWI had the highest specificity (65.2%, 95% CI = 44.9–81.2%) but the lowest sensitivity (48.3%, 95% CI = 31.4–65.6%). The wide CIs reflect the limited sample size and underscore the exploratory nature of this sub-analysis. Nonetheless, these findings support the rationale for multiparametric protocols that combine sequences to optimise diagnostic accuracy and highlight the need for larger, multicentre studies to establish the true diagnostic performance of individual sequences in detecting SVI.

There are some limitations in this study. Outcomes can be promising in retrospective studies but may lack structure as allocation to the treatment is not orchestrated. Furthermore, since the systems of management were changing in the time period of the study, outcomes may be impacted by those systems. For instance, the tier of salvage over adjuvant radiotherapy was changed because of the role shift by the RADICALS-RT study outcomes. Finally, the new technology of both imaging and surgical systems within a time period of study may provide additional complications. Of the 29 participants (19.9%) in the study, the status of the lymph nodes was not taken, prohibiting them from the multivariated analysis and thus creating a loss of power. Finally, in the time period of the study, the reporting of the parameters that impacted the outcomes, such as the involvement of the neck of the bladder and extraprostatic extension, was reported inconsistently and thus was excluded from the analysis. On review of the parameters of mpMRI, (n = 53) these yielded inconsistent high margins in the four primary locations and varying results of technology may have been responsible for this. It was also noted that the expanse of time of 20.5 months from the study to the time of closure may also be impacted because of a documented lack of time of recurrence and/or increased interval between imaged sites, thus affecting the importance of the outcome.

In summary, this study demonstrates that in patients with pT3b prostate cancer treated with RALP, positive surgical margins are the strongest independent predictor of BCR, with Grade Group and nodal involvement providing additional prognostic information on Kaplan–Meier and univariate analysis. Preoperative factors such as age, PSA, and prostate volume were not associated with recurrence and should not be overemphasised in surgical decision-making. These findings support intensified surveillance and early consideration of salvage or adjuvant therapy in patients with adverse pathology, particularly those with positive margins. While mpMRI remains an essential staging tool, clinicians must be aware of its potential to overstage SVI, and treatment decisions should incorporate integrated radiological and pathological assessment.