Comparison of contemporary methods for estimating prostate tumour volume in pathological specimens

Transcription

1 Comparison of contemporary methods for estimating prostate tumour volume in pathological specimens Marlon Perera, Nathan Lawrentschuk*, Damien Bolton* and David Clouston Western Health, *Department of Surgery and Ludwig Institute for Cancer Research, University of Melbourne, Austin Hospital, and Focus Pathology, Melbourne, VIC, Australia Objective To evaluate the accuracy of various prostate tumour volume (TV) estimation methods. To determine the most appropriate estimation method for current clinical practice. Patients and Methods Radical prostatectomy (RP) specimens from multiple institutions were analysed by a single uro-pathologist between September 9 and May 11. Tumour properties including thickness, width and length were collected and TV was established using computer-assisted image analysis (CAIA). TV estimation methods including; square, cuboidal and ellipsoidal estimations were calculated using previously reported formulae. The estimation methods were compared against the gold-standard and the accuracy of identifying clinically significant tumours of TV.5 cc was determined. Results In all, 99 consecutive specimens were analysed by a single uropathologist. The median index TV on CAIA was 1. cc. Of the four estimation methods, the ellipsoid methods produced the closest correlation with the gold-standard (r.91, P =.71). This correlation lost accuracy when larger tumours (TV > cc) were excluded from the analysis (r =.73, P =.3). Sensitivity and specificity for identifying clinically significant tumours was 9% and 9% respectively, when using the ellipsoid estimation. Conclusions In current uro-pathology, the ellipsoidal estimation method appears to be the most suitable for estimating TV in prostate cancers. This method is cheap, reproducible and sensitive and can be safely used as a surrogate for CAIA volumes when such technology is not available. Keywords image processing, computer-assisted/methods, male, prostatectomy, prostatic neoplasms/pathology, prostatic neoplasms/surgery Introduction Prostate cancer tumour volume (TV) and its clinical significance in radical prostatectomy (RP) specimens is controversial [1,]. TV is of prognostic significance on univariate analysis and correlates closely with Gleason score, stage and surgical margin status of RP specimens. Given this close correlation, it is not surprising that TV loses its independence as a prognostic factor on multivariate analysis [3 7]. Due to the lack of independent prognostic significance and difficulties in accurately measuring TV, many pathologists choose not to give a quantifiable measure of TV [8]. Despite this, at the 9 International Society of Urological Pathology (ISUP) Consensus Conference in Boston, USA, there was overwhelming consensus of 97% of attending pathologists that some form of quantitative measurement of TV should be reported, although no consensus was reached regarding method(s) [9]. There are many documented methods for calculating prostate TV. Currently, computer-assisted image analysis systems (CAIA) are considered the most accurate means of determining TV [1]. However, these methods are labour intensive and time consuming and are not commonly used in routine practice. Alternative methods for the estimation of TV (or size) in clinical practice rely on calculation of basic 13 The Authors BJU International 13 BJU International doi:1.1111/bju.158 BJU Int 1; 113, Supplement, 9 3 Published by John Wiley & Sons Ltd. wileyonlinelibrary.com

2 Perera et al. tumour dimensions. We sought to evaluate the accuracy of three estimation calculations and maximum tumour diameter (MTD) compared with standardised CAIA. Further, we aimed to determine the sensitivity and specificity of these various TV estimation methods in detecting clinically significant tumours in routine practice. The Noguchi et al. [11] ellipsoid method, the Chen et al. [1] three-dimensional cube volume estimation and the Renshaw et al. [13] estimated square of the largest tumour together with MTD were selected as a follow-up of the Noguchi et al. [11] observations in. Fig. 1 Illustration of measurements obtained on pathological dissection. CST, cross-sectional thickness; L, length; W, width. Base CST Patients and Methods TV analysis was performed on RP specimens received by a large-volume specialist uropathology service with institutional ethics approval. Surgery was performed between September 9 and May 11 at multiple institutions and histology was reported by a single uro-pathologist (D.C.). Specimens were included for analysis if there was a focal lesion of prostatic malignancy. Apex L W Specimen Processing After RP, the prostate was fixed in 1% neutral buffered formalin. The seminal vesicles were amputated and the apex and base of the prostate were removed and submitted as sagittal sections. The posterior vertical dimension of the remainder of the prostate was measured and the prostate was sliced transversely, perpendicular to the posterior surface. Routinely, 5 7 slices were obtained for each specimen (mean slice thickness.5 mm). After fixation, the outlines of all tumour foci were marked with a marking pen on the slides and the tumour foci were divided into the index tumour (largest tumour focus taking into account Gleason score) and other foci. The slides were then scanned into Adobe Photoshop Elements 7 and analysed using Scion image (version..3. freeware) to determine the total surface area (SA) of the tumour. The TV calculation = SA tumor thickness tissue shrinkage correction factor. The tissue shrinkage correction factor was determined by the change in surface area of slices before and after fixation for five specimens. In these cases, the prostate was sliced initially after < h of formalin fixation. The transverse slices were scanned and a surface area for each slice was obtained using CAIA. The tissue slices were then subdivided and fixed with overnight processing. Sections were then cut at μm and stained with haematoxylin and eosin. The slices were subsequently reconstructed digitally and the surface areas of each corresponding slice was determined by CAIA. The average reduction in surface area after fixation was calculated. Calculation and Evaluation of Estimation Methods The calculated values were based on the linear dimensions of the index tumour only. The number of slices containing index tumour were recorded. Tumour length (MTD) was defined as the largest tumour dimension on any cross-section and tumour width was considered as the maximal width perpendicular to the MTD. Index tumour thickness was calculated as the number of slices containing index tumour multiplied by the average slice thickness of the respective specimen. Several formulae required multiplication by a constant (k), used in accordance with published literature. Figure 1 depicts the tumour dimensions used for TV calculation. Once the tumour biometric variables were recorded, the data was entered into Excel 8 program (Microsoft Corp, Redmond, WA, USA). The calculations were then performed for each estimation method in question, as stated below: 1. MTD = length. Estimated square method described by Renshaw et al. [13] = length width 3. Three-dimensional cuboid described by Chen et al. [1] = k(.) length width thickness. Ellipsoid estimation described by Noguchi et al. [11] = k(π/) length width thickness 5. Ideal estimation to fit the present study = k (variable) length width thickness Statistical Analysis The accuracy of each estimation method was assessed against the gold standard of CAIA by comparing TV results for each tumour specimen. TV was considered a non-parametric variable and Wilcoxon signed-rank test was used to assess the significance of each mathematical transformation against the respective CAIA value. The product of each estimation 13 The Authors 3 BJU International 13 BJU International

3 Methods of estimating prostate tumour volume in pathological specimens method was plotted against the CAIA volume using SPSS v17 statistics package (SPSS Inc., Chicago, IL, USA). Linear regressions were then fitted for each estimation method and compared with the line of equality using the Student s t-test. Clinically insignificant tumours are defined as having a volume of <.5 cc, Gleason score, stage T with negative margins as previously reported [1,15]. Sensitivity and specificity were established for each estimation method for predicting tumours of >.5 cc. Results During the study period, 99 consecutive specimens were assessed prospectively by a single uro-pathologist. The median (range) age of the cohort was (1 83) years and malignancy confirmed in all specimens, with a predominance of the acinar adenocarcinoma subtype in 81 (93.%). TV Calculation A correction value of 1.1 was applied to all TV calculations to allow for tumour shrinkage. The mean (range) change in surface areas of transverse slices before and after fixation, processing and staining, was 9. (8 9)%, with an estimated vertical shrinkage of %. This produced a fixed correction factor of 1.1 and was used in all TV measurements. TV was determined by CAIA for all 99 RP specimens, with only TV of the index tumour used for statistical analysis. The median (range) index TV was 1. (.3 8.) cc and was considered clinically insignificant in 5 (1.7%) cases with a TV of.5 cc. Evaluation of Estimation Method Using MTD, estimated square and the cuboid estimation methods, the median TV was calculated and compared with a median TV of 1. cc on CAIA. Linear regressions for each method were plotted against CAIA plots and tested against the line of equality (Figs A, 3A and A). This was then repeated after larger tumours (TV of cc) were excluded from the analysis (Figs B, 3B and B). All plotted regressions from MTD, estimated square and cuboid estimations differed significantly from CAIA plots, regardless of TV (P <.1). The ellipsoid estimation method produced a median TV of 1.5 cc, which did not represent a significant difference when compared with the median of the CAIA volumes (P =.7). As above, a linear regression for the ellipsoid estimation was fitted and compared with the CAIA values. The difference in regressions was not statistically significant (P =.71; Fig. 5A). This finding lost significance when larger tumours (TV of cc) were excluded from the analysis (P =.3; Fig. 5B). The ideal estimation method was obtained by calculating the coefficient of the previously mentioned formula that provided the closest correlation with the CAIA TVs. In the present study, when TV > cc was excluded, the ideal coefficient was Fig. Comparison of MTD estimation with tumour volumes < cc. Red line, line of equality; dotted line, MTD vs CAIA regression. A. MTD estimation vs CAIA B. MTD estimation vs CAIA (volumes < cc) MTD estimation, cc 3 1 MTD estimation, cc Fig. 3 Comparison of square estimation with tumour volumes < cc. Red line, line of equality; dotted line, square estimation vs CAIA regression. A. Square estimation vs CAIA B. Square vs CAIA (volumes < cc) Square estimation, cc 3 1 Square estimation, cc The Authors BJU International 13 BJU International 31

4 Perera et al. Cuboid estimation, cc 3 1 A. Cuboid estimation vs CAIA Cuboid estimation, cc B. Cuboid vs CAIA (volumes < cc) Fig. Comparison of cuboid estimation with tumour volumes < cc. Red line, line of equality; dotted line, cuboid vs CAIA regression Ellipsoid estimation, cc 3 1 A. Ellipsoid estimation vs CAIA Ellipsoid estimation, cc B. Ellipsoid estimation vs CAIA (volumes < cc) Fig. 5 Comparison of ellipsoid estimation with tumour volumes < cc. Red line, line of equality; dotted line, ellipsoid vs CAIA regression Table 1 Summary of results of estimation methods. Method CAIA MTD Estimated square Cuboid estimation Ellipsoid estimation Volume estimation formula* L L W. M W CST π/ L W CST Median tumour volume, cc Does this differ significantly from the mean of the Yes,P<.1 Yes,P<.1 Yes,P<.1 No,P=.7 CAIA volume? Gradient of line of best fit NA.11 (r =.3, P <.1.3 (r =.8, P <.1.75 (r =.91, P <.1.99 (r =.91, P <.1 Does this differ significantly from the line of Yes,P<.1 Yes,P<.1 Yes,P<.1 No,P=.5 equality (Students t-test)? Gradient of line of best fit for tumour volumes NA.3 (r =.53).8 (r =.71).89 (r =.73) 1.1 (r =.73) <cc Does this differ significantly from the line of Yes,P<.1 Yes,P<.1 Yes,P<.1 Yes,P=.3 equality (Students t-test)? Number of patients with tumours of >.5 cc 9/99 8/99 /99 3/99 37/99 Number of patients correctly identified with NA 8/9 /9 /9 33/9 tumours of >.5 cc % sensitivity (sens.) and specificity (spec.) of sens: 99. sens: 9 sens: 89 sens: 9 detecting tumours of >.5 cc spec: 3 spec: 9 spec: 98 spec: 9 *L, maximum length; W, perpendicular width; CST, cross-sectional thickness..5, compared with. of the cuboid estimation and.5 of the ellipsoid estimation. These results are summarised in Table 1. Identification of Clinically Significant Tumours according to Size The estimation methods were compared with CAIA for their ability to correctly classify tumours using the currently accepted threshold values for clinical significance, of.5 cc for TV and 1 mm for MTD. The sensitivity and specificity of MTD, estimated square, cuboid and ellipsoid estimation methods in identifying tumours of >.5 cc are summarised in Table 1. The mean (median, range) MTD was 1.8 (15.,.1 7) mm. In all, 3 patients had an index MTD of >1 mm. When compared with the index TV by CAIA, 8 of the 3 cases 13 The Authors 3 BJU International 13 BJU International

5 Methods of estimating prostate tumour volume in pathological specimens with a MTD of >1mmhadaTVof>.5 cc. While, 1 of the 5 cases with an index MTD of <1 mm had an index TV of >.5 cc. Excluding two outliers, the index TV was always <.5 cc when the MTD was 7 mm, and always >.5 cc when the index MTD was mm. Discussion Prostate TV provides important information in determining the likely patient outcome after RP for biochemical recurrence, tumour progression and patient survival. Although it is not an independent prognostic factor on multivariate analysis, it is an important part of the pathology report providing an indication of tumour size. Given that most tumours with a TV of > cc are of higher stage ( T3, data not shown), estimated TV may play a more important role in T disease. Substaging T disease is problematic because ptc can include such a wide variation in extent of disease, from small foci of carcinoma in both lobes to widespread tumour throughout both lobes. Given that most prostate cancer is organ confined in contemporary RP specimens, and that there is wide variation in tumour size in the Tc category, it would seem that TV could play an important role in stratifying tumours within the T stage. Unfortunately, there is no uniform approach to estimation of prostate TV and currently remains variable depending on pathology institution. Of the previously reported estimation methods, the ellipsoid method corresponds most closely to the CAIA results when all TV were considered. Both the estimated square and the cuboid estimation method significantly underestimated the TV when using the gold standard. Interestingly, the ellipsoid method tended to overestimate the TV for smaller tumours with TV of < cc. Noguchi et al. [11] reported similar findings stating that the ellipsoid method overestimated the tumour volume by 9%. They argued that the ellipsoid method overestimates TV because the calculation assumes that the tumours are spherical whereas in the prostate, the tumours tend to be crescent shaped. The present results support this argument, as the smaller volume tumours grow in a linear or crescent fashion, and become more spherical as the tumour reaches a larger size and extends more centrally from the peripheral zone. This would account for the closer correlation between the ellipsoid method and CAIA when the larger tumours are included in the assessment. Conversely, all other methods of TV estimation became less accurate when larger tumours (TV of > cc) where included in the analysis. This was similar to the Chen et al. [1] findings, with the greatest accuracy being seen for tumours of cc. The ellipsoid method and the cuboid method are closely related, with the main difference being confined to the coefficient applied to correct the raw cubic volume of the three linear dimensions used to determine the estimated TV. Chen et al. [1] noted that the volume using the uncorrected product of the three dimensions tended to overestimate the TV and applied a coefficient of. based on the slope of the regression lines in his study. Noguchi et al. [11] based their coefficient of π/ (.5) on a similar estimation method used to measure the volume of lung cancer [1]. The ideal coefficient for the present study was.5. In their studies, both Chen et al. [1] and Noguchi et al. [11] suggested that the ellipsoid method overestimated the TV by up to 3%. This discrepancy from the present study may be a reflection of differences in slice thickness, as the present study allowed for variation in slice thickness between specimens. Noguchi et al. showed that a slice thickness of mm gave an overestimation of 9.5% compared with a thickness of 3 mm. The results presented here more closely replicate routine pathology practice, where the prostate gland is sliced freehand, compared with the Noguchi et al. laboratory where the prostate was sliced mechanically at 3-mm intervals. Clinically insignificant tumours are defined as a TV of <.5 cc, Gleason score <7 or stage T with negative margins [1,15]. Therefore, it is important that the estimation methods should accurately sort tumours < and >.5 cc. In the present study, 83.3% of tumours were clinically significant with a TV of >.5 cc as determined by CAIA. All of the methods underestimated the number of clinically significant index tumours. The ellipsoid and cuboid methods incorrectly estimated TV of.5 cc in 1 (5.3%) and 7 (9.%) cases, respectively. Chen et al. [1] had similar findings with the ellipsoid estimation, with under-reporting of.9% of cases as clinically insignificant that had CAIA volumes of >.5 cc. An MTD of 1 mm has been proposed as a substitute size criterion to replace the value of.5 cc for clinically insignificant tumours. This is a reasonable analogy, as a sphere with a diameter of 1 mm has a volume of.5 cc. Renshaw et al. [17] showed tumours with a MTD of <1 mm had a 15% risk of biochemical recurrence, compared with 73% for tumours of > mm, findings supported by Stamey et al. [5]. In the present study, 3 cases had a MTD of >1 mm, and of these, 8 (93.8%) had a TV of >.5 cc. Importantly, 1 of the 5 cases with a MTD of <1mmhadaTVof>.5 cc when measured by CAIA, indicating that up to 5% of cases with a clinically insignificant MTD may have a clinically significant TV by CAIA. Both the ellipsoid and cuboid methods failed to identify most of these borderline tumours. In the present study, all tumours with MTD of 7mmhadaTVof.5 cc, while all tumours with a MTD of >mmhadatvof >.5 cc. The same thresholds in the Chen et al. [1] study were 9 and 5 mm respectively, and these differences are most probably due to the addition of a shrinkage factor of 1.1 to the TV calculations in the present study, and also differences in slice thickness between the studies. The limitations of the present study are that it is from a single institution and this may influence outcomes. However, as the 13 The Authors BJU International 13 BJU International 33

6 Perera et al. study focus is on the methods used, this may have less impact than if it was of clinical consequence. Ideally, a large prospective study with multiple pathologists would create tighter datasets and potentially allow examination of inter-observer variability within methods should they exist. In conclusion, we have shown that in routine pathology laboratories the ellipsoid method provides a close estimate of TV when compared with the gold standard of CAIA. The ellipsoid estimation provides acceptable sensitivity and specificity in determining clinically significant tumours. This method also performs reasonably when TVs ranges between.5 and cc and thus may play a role in stratifying pt disease. This measurement is simple, reproducible and is considered appropriate for routine surgical pathology practice. Understanding the limitations and accepting the need for TV will allow pathologists and clinicians involved with prostate cancer to ask more informed questions and this may lead to more accurate use of TV in future practice and studies. Conflict of Interest None declared. References 1 Wolters T, Roobol MJ, van Leeuwen PJ et al. Should pathologists routinely report prostate tumour volume? The prognostic value of tumour volume in prostate cancer. Eur Urol 1; 57: 81 9 Srigley JR. Key issues in handling and reporting radical prostatectomy specimens. Arch Pathol Lab Med ; 13: Fukuhara H, Kume H, Suzuki M et al. Maximum tumor diameter: a simple independent predictor for biochemical recurrence after radical prostatectomy. Prostate Cancer Prostatic Dis 1; 13: 7 Nelson BA, Shappell SB, Chang SS et al. Tumour volume is an independent predictor of prostate-specific antigen recurrence in patients undergoing radical prostatectomy for clinically localized prostate cancer. BJU Int ; 97: Stamey TA, McNeal JE, Yemoto CM, Sigal BM, Johnstone IM. Biological determinants of cancer progression in men with prostate cancer. JAMA 1999; 81: 1395 Wheeler TM, Dillioglugil O, Kattan MW et al. Clinical and pathological significance of the level and extent of capsular invasion in clinical stage T1- prostate cancer. Hum Pathol 1998; 9: 85 7 PartinAW,YooJ,CarterHBetal.The use of prostate specific antigen, clinical stage and Gleason score to predict pathological stage in men with localized prostate cancer. J Urol 1993; 15: 11 8 Kench J, Clouston D, Delprados W, Eade T, Ellis D, Horvath L. Prognostic factors in prostate cancer. Key elements in structured histopathology reporting of radical prostatectomy specimens. Pathology 11; 3: van der Kwast TH, Amin MB, Billis A et al. International Society of Urological Pathology (ISUP) Consensus Conference on Handling and Staging of Radical Prostatectomy Specimens. Working group : T substaging and prostate cancer volume. Mod Pathol 11; : Sherwin JC, Mirmilstein G, Pedersen J, Lawrentschuk N, Bolton D, Mills J. Tumor volume in radical prostatectomy specimens assessed by digital image analysis software correlates with other prognostic factors. J Urol 1; 183: Noguchi M, Stamey TA, McNeal JE, Yemoto CE. Assessment of morphometric measurements of prostate carcinoma volume. Cancer ; 89: 15 1 Chen ME, Johnston D, Reyes AO, Soto CP, Babaian RJ, Troncoso P. A streamlined three-dimensional volume estimation method accurately classifies prostate tumors by volume. Am J Surg Pathol 3; 7: Renshaw AA, Chang H, D Amico AV. Estimation of tumor volume in radical prostatectomy specimens in routine clinical practice. AmJClin Pathol 1997; 17: Stamey TA, Freiha FS, McNeal JE, Redwine EA, Whittemore AS, Schmid HP. Localized prostate cancer. Relationship of tumor volume to clinical significance for treatment of prostate cancer. Cancer 1993; 71 (Suppl. 3): Epstein JI, Walsh PC, Carmichael M, Brendler CB. Pathologic and clinical findings to predict tumor extent of nonpalpable (stage T1c) prostate cancer. JAMA 199; 71: Binks S, Clelland C, Layton C. A comparison of pathological methods of measuring lung cancer volume. J Clin Pathol 199; 9: 5 17 Renshaw AA, Richie JP, Loughlin KR, Jiroutek M, Chung A, D Amico AV. The greatest dimension of prostate carcinoma is a simple, inexpensive predictor of prostate specific antigen failure in radical prostatectomy specimens. Cancer 1998; 83: 78 5 Correspondence: Marlon Perera, Department of Surgery, Western Health, Footscray, VIC 31, Australia. marlonlperera@gmail.com Abbreviations: CAIA, computer-assisted image analysis; MTD, maximum tumour diameter; TV, tumour volume; RP, radical prostatectomy. 13 The Authors 3 BJU International 13 BJU International