Measurements of cancer extent in a conservatively treated prostate cancer biopsy cohort

Abstract

The optimal method for measuring cancer extent in prostate biopsy specimens is unknown. Seven hundred forty-four patients diagnosed between 1990 and 1996 with prostate cancer and managed conservatively were identified. The clinical end point was death from prostate cancer. The extent of cancer was measured in terms of number of cancer cores (NCC), percentage of cores with cancer (PCC), total length of cancer (LCC) and percentage length of cancer in the cores (PLC). These were correlated with prostate cancer mortality, in univariate and multivariate analysis including Gleason score and prostate-specific antigen (PSA). All extent of cancer variables were significant predictors of prostate cancer death on univariate analysis: NCC, hazard ration (HR)=1.15, 95% confidence interval (CI)=1.04–1.28, P=0.011; PPC, HR=1.01, 95% CI=1.01–1.02, P<0.0001; LCC, HR=1.02, 95% CI=1.01–1.03, P=0.002; PLC, HR=1.01, 95% CI=1.01–1.02, P=0.0001. In multivariate analysis including Gleason score and baseline PSA, PCC and PLC were both independently significant P=0.004 and P=0.012, respectively, and added further information to that provided by PSA and Gleason score, whereas NNC and LCC were no longer significant (P=0.5 and P=0.3 respectively). In a final model, including both extent of cancer variables, PCC was the stronger, adding more value than PLC (χ² (1df)=7.8, P=0.005, χ² (1df)=0.5, P=0.48 respectively). Measurements of disease burden in needle biopsy specimens are significant predictors of prostate-cancer-related death. The percentage of positive cores appeared the strongest predictor and was stronger than percentage length of cancer in the cores.

Introduction

The advent of prostate-specific antigen (PSA) screening has resulted in an increase in the detection of early stage tumours [1–3]. However, the benefits of radical therapy in patients with low- or intermediate-risk disease are not clear cut, and many of these patients may benefit from a more conservative approach [4]. Nomograms based on preclinical parameters have been devised to guide clinical management [5–8], but these do not routinely include measures of cancer extent.

Pathological data from biopsy specimens other than Gleason score have increasingly been used to improve the predictive accuracy of outcome models that guide clinical management, by identifying factors associated with an increased risk of disease progression in terms of PSA failure and metastatic disease post-radical prostatectomy [6, 9–14].

A cohort study by the Transatlantic Prostate Group developed a model using Gleason score and PSA to stratify patients with localised prostate cancer managed conservatively into three groups with 10-year cancer-specific mortalities of <10%, 10–30% and >30% [6]. After Gleason score and PSA, the extent of cancer in the biopsy specimen showed a strong association with prostate-cancer-related deaths on univariate and multivariate analysis. However, there are different methods of measuring cancer extent in prostate needle core biopsies. In this study, we determine which pathological assessment of the “extent of cancer” in diagnostic prostate biopsies best predicts long-term survival.

Methods

Study population, follow-up and end points

The protocol for patient selection and data collection methodology has been detailed in a previous report by the Transatlantic Prostate Group [6] and is summarised below. Patients diagnosed with biopsy-proven, localised prostate cancer between 1990 and 1996 were identified by participating UK cancer registries. Exclusion criteria included age >76 years, no baseline PSA, invasive carcinoma within 5 years prior to diagnosis, hormone treatment prior to biopsy and a follow-up period of less than 6 months. Patients were also excluded if they underwent radical prostatectomy or radiotherapy within 6 months of diagnosis or had evidence of metastasis, a baseline PSA >100 ng/ml or died within 6 months of diagnosis.

The clinical end point was defined as death from prostate cancer. Survival outcomes were ascertained from medical records and cancer registry data. Last follow-up was via the cancer registries, and the last review took place in December 2006.

Pathological assessment of biopsy specimens

Diagnostic needle core biopsy specimens were reviewed by an expert panel of urological pathologists to confirm the diagnosis and Gleason grading. Extent of cancer in biopsy specimens was determined by multiple methods (Table 1): data available included total number of cores (TNC) and number of cores with cancer (NCC) as well as the total length of all cores (TLC) and length of cancer in cores (LCC). Apart from the NCC and LCC, these data allowed calculation of the percentage of cancer-positive cores (PCC = NCC/TNC) and secondly the percentage length of cancer (PLC) in the cores (PLC = LCC/TLC). Measurements of cancer length and core lengths were all made by an expert uropathologist (DB) with microscope-mounted Vernier’s scales and rulers, respectively. Measurements of linear cancer length within individual cores did not include intervening non-tumour tissue. Gleason grading conformed to modern ISUP criteria [15].

Table 1.

Measurements of cancer extent made including abbreviations and calculations

Different measurements of cancer extent on prostate core biopsies	Abbreviations and calculations
Total number of cores	TNC
Number of cores with cancer	NCC
Total length of cores	TLC
Length of cancer in all cores	LCC
Percentage of cancerous cores	PCC = NCC/TNC×100
Percentage length of cancer	PLC = LCC/TLC×100

Statistics

The primary end point for this study was death from prostate cancer. Univariate and multivariate analyses were performed by proportional hazard models. All P values were two-sided and 95% confidence intervals were based on normal distribution.

Results

Patient selection identified a cohort of 2,333 eligible men. Centrally reviewed, Gleason grading results were available for 1,656 (71%) cases, 474 cases were excluded on histological grounds (missing histology 284, incorrect pathological specimen 43, histologically negative 135, no tissue left in the block 6 and six cases were histologically ungradable for Gleason score) and for 203 cases the histology specimen was not requested for review. Cases diagnosed from prostate chips were not included in this study, and analysis was confined to the 744 cases diagnosed on the basis of needle core biopsies. The mean age at diagnosis was 69.2 years (range 44–76 years). Median follow-up was 106 months (range 6–197 months).

The total number of cores per patient ranged from one to nine, and the proportion positive is shown (Fig. 1). In univariate analysis, NCC was a significant predictor of prostate cancer death (hazard ratio (HR)=1.15, 95% confidence interval (CI)=1.04–1.28, χ² (1df)=6.4, P=0.011) but was less informative than the PCC (HR=1.01, 95% CI=1.01–1.02, χ² (1df)=27.1, P<0.0001). The KM survivor curve for PPC analysed in three groups (<50%, 50 to <100% or 100%) is shown in Fig. 2a.

Figure 1.

Distribution of cancer-positive cores according to the total number of cores reviewed, in cases diagnosed by needle biopsy (n=727)

Figure 2.

Kaplan–Meir plot estimating prostate cancer cause-specific survival according to the extent of cancer in the biopsy: a percentage cancer-positive cores n=727; percentage length of cancer in the cores, b quintiles, c 30% cutoff point (n=737) for men diagnosed by needle biopsy

In multivariate analysis, in the standard model including Gleason score (≤5, 6, 7, 8, 9 or 10) and PSA (ng/ml) (0–4, >4–10, >10–25, >25–50, >50–100), PCC was significant independently of Gleason score and PSA and provided additional prognostic value (Δχ² (1df)=8.1, P=0.04). In contrast, NCC provided almost no additional prognostic value (Δχ² (1df)=0.45, P=0.5), Table 2.

Table 2.

The added value of pathological assessments of the extent of cancer (PCC, NCC, PLC, LCC) in a multivariate model including Gleason score (G) and baseline PSA level for predicting prostate cancer mortality in men with conservatively treated prostate cancer diagnosed by needle biopsy

Variable	Number	χ2 (1df)h	Δχ2	P value
Positive cores (%)
Ga	727	59.8	–
G + PSAb	727	71.3	11.6	<0.001
G + PSA + PCCc	727	79.5	8.1	0.004
G + PSA + NCCd	727	71.4	0.45	0.50
Length of cancer (%)
G	737	63	–
G + PSA	737	73.1	10.1	0.002
G + PSA + PLC quintilese	737	78.7	5.6	0.018
G + PSA + PLC binaryf	737	79.5	6.4	0.012
G + PSA + LCCg	737	74.1	0.96	0.3
Combined model
G	723	59.7	–
G + PSA	723	70.5	10.8	0.001
G + PSA + PCC	723	78.3	7.8	0.005
G + PSA + PLC binary	723	75.1	4.5	0.03
G + PSA + PLC quintiles	723	74.7	4.2	0.04
G + PSA + PCC + PLC binary	723	78.8	0.5	0.5

G Gleason score, PSA prostate-specific antigen, PCC percentage of cancer-positive cores, NCC number of cancer-positive cores, PLC percentage length of cancer, PLC percentage length of cancer, LCC length of cancer

^aG score groups: ≤5, 6, 7, 8, ≥9

^bPSA (ng/ml) groups: 0–4, >4–10, >10–25, >25–50, >50–100

^cPCC (%) groups: < 50, 50 to <100, 100

^dNCC (1, 2, 3, ≥4)

^ePLC (%) quintiles: 1.8–14.1,14.2–29.5, 30–49.5, 50–76, 76.1–100

^fPLC binary: ≤30, >30, cut off

^gLCC (mm) continuous

^hAll chi-squared values (χ2) are for trend and are all on one degree of freedom (df)

The extent of cancer measured as the absolute (LCC) and percentage length of cancer in the needle cores (PLC) was recorded for 737 cases. In univariate analysis, PLC was a significant predictor of prostate cancer death, when assessed as a continuous variable (HR=1.01, 95% CI=1.01–1.02 χ² (1df)=21.26, P=0.0001) or as a grouped variable. The most informative grouping was based on quintiles (χ² (1df)=27, P<0.0001) (Fig. 2b), but this was only slightly more informative than a simpler variable based on two groups (≤30%, >30%) (HR=2.21, 95% CI=1.6–3.1, χ² (1df)=25, P<0.0001) (Fig. 2c). Although LCC was also a significant predictor of prostate cancer death, it was less informative (HR=1.02, 95% CI=1.01–1.03, χ² (1df)=9.8, P=0.002).

In multivariate analysis, PLC was significant independently of Gleason score and PSA and provided some additional prognostic value to the standard model. The added value of the PLC variable based on two groups (≤30%, >30%) was very similar to that based on quintiles (Δχ² (1df), 5.6, P=0.018, and Δχ² (1df), 6.3, P=0.012 respectively), but LCC provided almost no additional prognostic value (Δχ² (1df), 0.96, P=0.33) (Table 2).

An analysis of the 723 cases with both extent of cancer variables (PCC and PLC) available allowed a direct comparison of their predictive power. The predictive power of the variable PCC (<50%, 50% to <100% or 100%) was greater than that of PLC measured either as a two- or five-group variable (Δχ² (1df)=7.8, P=0.005, Δχ² (1df)=4.5, P=0.03, Δχ2 (1df), 4.2, P=0.04), respectively (Table 2).

Detailed multivariate analysis for the final model including PCC, Gleason score (<7, =7, >7) and baseline PSA (ng/ml) (0–4, >4–10, >10–25, >25–50, >50–100) is shown in Table 3.

Table 3.

Multivariate analysis of pathological assessments of the extent of cancer measured as PCC, in a model including G and baseline PSA level for predicting prostate cancer mortality in men with conservatively treated prostate cancer diagnosed by needle biopsy (n=727)

Variable	N (%)	Hazard ratio (95% CI)	χ2 (1df)a contribution	P value
Gleason score
<7	262 (36)	1 (ref)	38	<0.0001
7	271 (37)	1.78 (1.17–2.73)
>7	194 (27)	3.47 (2.26–5.31)
PSA (ng/ml)
0–4	52 (7)	1 (ref)	11	0.001
4–10	138 (19)	0.68 (0.35–1.36)
10–25	261 (36)	0.68 (0.37–1.26)
25–50	160 (22)	1.5 (0.82–2.75)
50–100	116 (16)	1.35 (0.72–2.51)
PCC (%)
<50	138 (19)	1 (ref)	7	0.008
50 to <100	255 (35)	1.38 (0.82–2.33)
100	334 (46)	1.74 (1.06–2.85)

χ² chi-squared, df degrees of freedom, CI confidence interval, ref reference category, PSA prostate-specific antigen, PCC percentage of cancer-positive cores, PLC percentage length of cancer in the cores

^aAll chi-square values (χ²) are for trend and are all on one degree of freedom (df)

Discussion

The present dilemma arising out of the advent of PSA screening is how best to manage the increasing proportion of patients with intermediate- and low-risk prostate cancer, as defined by standard models using age, PSA and Gleason score. This is recognition that many of these patients have indolent disease in whom the benefits of radical prostatectomy do not outweigh the surgical morbidity. The parameters that define these two groups are becoming more stringent as clinicians adopt more conservative management protocols [16, 17]. However, even with standard models of prognostication, there is heterogeneity within the low–intermediate-risk (Gleason 6 and 7 and low PSA) patient groups and a recognition that a proportion of patients will develop progressive disease in whom radical treatment is then warranted [18].

This has driven the exploration of pathological factors in preoperative specimens to predict poor outcome. In the absence of validated molecular markers of prognostic valve, which are practical in large screening populations, many studies have focused on measurements of disease burden in needle core biopsies.

The number of positive cores in sextant biopsies have been shown to correlate with tumour stage [19–22] and surgical margin status in radical prostatectomy specimens [23, 24] and the presence pelvic lymph node metastasis [25]. Similarly, the percentage of positive cores has been shown to be predictive of stage [26], extra-capsular extension [27, 28], distant metastasis, biochemical progression and death following radical prostatectomy [29].

In addition, the percentage cancer involvement of biopsies is predictive of capsular perforation and seminal vesicle involvement [30]. The total length of cancer has also been shown to correlate with stage, positive surgical margins and biochemical recurrence following radical prostatectomy, respectively [24]. Brimo et al. showed that the greatest percentage of cancer, the total percentage of cancer and the greatest millimetric length of cancer in the needle cores were closely associated with clinical stage [31].

A systematic review by Harnden et al. determined the significance of measurements of disease burden in biopsy specimens for predicting risk of progression after radial prostatectomy [32]. The authors were unable to make firm conclusions on absolute measurements made of tumour burden. Tumour burden represented as a percentage was found to be more consistent, and most studies they reviewed showed that the percentage of cancer involvement was an independent predictor of PSA recurrence following radical prostatectomy.

In the current study, the percentage of positive cores (FC) was strongly predictive of prostate cancer death over the long follow-up period (median follow-up 117 months) despite the heterogeneity of the study population with regard to standard prognostic indicators (age, PSA and Gleason score at presentation) and treatment-related factors (initial hormone treatment). These findings are consistent with previous studies in which the percentage of positive cores was predictive of stage, surgical margin status [33] and biochemical recurrence [10, 34–37].

The utility of the percentage of positive cores is dependant on the number of cores taken to sample disease extent, which acts as a corrective factor for the size of the gland. Thus, the accuracy of this measure increases with the number of cores taken. In this sense, the measure is likely to overestimate disease extent when the core number is low.

Conversely, when the core number is high (approaching 21 biopsies) and the fraction of positive cores is low, the measurement will more accurately predict low disease burden which is especially important during selection of patients for active surveillance programmes [38]. As only 10% of the cohort was composed of sextant biopsies, there were too few cases to analyse adequately. There was no heterogeneity within this subset of cases. It is hoped that a more contemporary cohort conforming more to modern practice will be published in the future.

Similar principles apply with measurements of the total fractional length of cancer in biopsy specimens. In this study, the power of the total fractional length of cancer in needle biopsies was almost uninformative. This may be explained by the measurement underestimating the size of large tumours when the biopsy number is low and thus may not reflect its true clinical usefulness in predicting prognosis. In fact, several studies have shown that the total percentage length of cancer was predictive of pathological stage and PSA recurrence post-radical prostatectomy [11, 24, 39]. However, other studies also failed to demonstrate a correlation between the percentage of prostate cancer in the biopsy and clinical stage or biochemical recurrence following radical prostatectomy [40, 41].

Limiting factors in this study include its retrospective nature, low biopsy numbers compared to contemporary series and the heterogeneity of the study population when analysing the significance of disease burden in the biopsies. However, despite the aforementioned confounding factors, no contemporary series, let alone one with in which patients are managed by watchful waiting, has enough follow-up to make long-term measurements of this nature. Furthermore, direct comparisons of predicted PSA failure after radical prostatectomy and prostate-cancer-related death in patients managed by watchful waiting or active surveillance need to be made with caution, as PSA failure does not necessarily imply clinical progression without correction for PSA doubling time [42–44].

Conclusions

This study is unique in correlating measurements of disease burden in diagnostic specimens with long-term survival in patients managed by watchful waiting. The percentage of positive cores was strongly predictive of prostate-cancer-related death on multivariate analysis and added additional information to models utilising PSA and Gleason score to refine management decision in patients with low-risk disease who would qualify for conservative management.