A validation study of new rules for the interpretation of cancer significance on prostate systematic biopsy

Transcription

1 A validation study of new rules for the interpretation of cancer significance on prostate systematic biopsy Gerald O. Ogola, Nicolas Delongchamps 2,3, and Robert Serfling 4 Center for Clinical Effectiveness, Office of the Chief Quality Officer, Baylor Scott and White Health, Dallas, Texas. 2 Department of Urology, Cochin Hospital, Paris Descartes University, Paris, France. 3 Inserm 5, INEM, Paris Descartes University, Paris, France. 4 Department of Mathematical Sciences, University of Texas at Dallas, Richardson, Texas. Corresponding author: Gerald Ogola, Ph.D. Center for Clinical Effectiveness, Office of the Chief Quality Officer, Baylor Scott and White Health, 88 N. Central Expressway Suite 5 Dallas, Texas (Voice) (FAX) GeraldO@baylorhealth.edu Conflicts of Interest: None of the authors have conflicts of interest to report Abstract word count = 25 Manuscript word count = 272 Running Head: Validation of prostate-volume-based decision rules Page of 4

2 ABSTRACT Purpose: To test with actual data a new decision rule derived by probability modeling of the number of positive cores, for deciding insignificant versus significant prostate cancer (CaP), based on prostate volume, Gleason score, tumor length on biopsy cores, and number of positive cores. Materials and Methods: A dataset of 59 cancer-involved autopsied prostate glands from patients aged 42 to 92 years with prostate volumes of 22 to 95 cc was used. An 8 coresystematic biopsy was performed on the first 47 patients. The first six cores were taken from the mid peripheral zone (MPZ), the next 6 cores from the central zone (CZ), and the last six cores from the lateral PZ (LPZ). In the last 2 patients, a saturation biopsy protocol of 36 cores was performed: the first 8 cores were taken in a similar fashion and 6 additional cores were taken in each of the 3 locations clinically insignificant CaP was defined on whole-mounts prostates as Gleason score < 7, total tumor volume <.5 cc,. For prostate volumes 2-3, 4, 5, 6-7, and 8- cc, and numbers of biopsy cores 6, 2, and 8, separate counts of Significant versus Insignificant CaP by both the model-based decision rule and the actual data were obtained. These yielded specificity (SP), sensitivity (SE), and concordance values, by prostate volume and number of cores, for evaluation of the efficacy of the decision rule. Results: The model-based rule yielded SP from 83 to, SE from 62 to, and concordance from 78 to. Being model-based, the rule and these results applied generally. Also, the findings compared favorably with currently used study-based rules and their individually fitted SP and SE, which, however, differ across studies. Conclusions: This model-based decision rule performed well with this dataset of autopsied prostates, confirming its practical feasibility and its potential to help reduce over- and undertreatment. However, additional validation studies with other datasets including higher prostate volumes are needed for further calibration and improvement of the model-based rule. KEY WORDS: Prostate cancer, biopsy, decision rule, validation, specificity, sensitivity Page 2 of 4

3 INTRODUCTION Planning for effective prostate cancer (CaP) biopsy sessions and interpreting marginally positive biopsy results is challenging. With millions of CaP biopsies performed annually, avoiding over-treatment poses a world-wide dilemma of large scale -3. To avoid over-treatment, one needs to distinguish between insignificant and significant CaP with sufficiently high specificity (SP), and this requires assigning some marginally positive cases to active surveillance 4-6. How to select such candidates? To avoid under-treatment, decision-making with sufficiently high sensitivity (SE) is required, and this entails increasing the number of biopsy cores as prostate volume increases, keeping in mind the burden on the patient. In what manner should this increase proceed? To address these questions, a probability model 7 for the number of positive cores in a biopsy session, as a function of prostate gland volume, tumor foci volume, and number of cores, was recently developed. The model was applied to construct favorable prostate-volume-specific decision rules for insignificant versus significant CaP on the basis of high specificity, high sensitivity and number of positive cores. By favorable, we mean the best trade-off between specificity, sensitivity and number of positive cores, while prioritizing high specificity over high sensitivity among available options. This is consistent with prostate cancer screening, where most efforts are directed towards increasing specificity, because it is highly unlikely that men will be tested only once in their lifetime, but will undergo serial tests, perhaps annually, and hence a false negative test is less likely to be serious 8. Gleason score and tumor length in the biopsy cores were incorporated to develop a fully structured decision rule for distinguishing significant versus insignificant CaP in a prostate biopsy session. In the present study, we aim to validate the fully structured decision rule introduced above using data collected from cancer-involved autopsied prostate glands dataset. MATERIAL AND METHODS Probability Model, Specificity and Sensitivity We developed a probability model 7 for the number of positive cores in a biopsy session, as a function of prostate gland volume, number of tumor nodules and their volumes, and number of cores. For prostate volumes through 2 cc and number of cores 6, 2, 8, and 24, we evaluated model-based specificity (SP) and sensitivity (SE) for decision rules of the form Decide insignificant CaP if D < x, for specified x and with D the number of positive cores. Page 3 of 4

4 Initially, we set aside Gleason score and tumor lengths in cores and defined insignificant CaP" simply as total tumor volume T <.5 cc. For this purpose, we defined model-based SP and SE, respectively, as the conditional probability of deciding insignificant CaP when CaP is truly insignificant, and the conditional probability of deciding significant CaP when CaP is truly significant. For SP we chose a total tumor volume T below the commonly adopted clinically significant threshold of.5 cc. In particular we used T =.25 cc, and defined SP(x n, V) = P(D x n, V, T =.25 cc) = P(D = x n, V, T =.25 cc), and for SE we chose T above.5 cc. In particular we used T = 2. cc and defined SE(x n, V) = P(D > x n, V, T = 2. cc) = P(D = x n, V, T = 2. cc). x x= min {k,n} x=x + Based on the derivations of (SP, SE) from 7, we obtained favorable choices of n and x. Since biopsy results also yield tumor lengths (or percent of cancer) in the cores and Gleason score, we combined this information with our model-based rules to develop a fully structured decision rule, as follows. For Gleason sum 7 or higher, immediately conclude Significant CaP. Otherwise, consider tumor lengths and if percent of CaP is sufficiently large in sufficiently many cores, then conclude Significant CaP. Otherwise, finally, apply the appropriate model-based rule to decide insignificant versus significant CaP. Validation Data The validation dataset was collected from consecutive prostate glands from deceased men, and was provided by the University Hospital and the Onondaga County Medical Examiner, Syracuse, NY and by the National Disease Research Interchange, Philadelphia, PA, USA. Informed consent was obtained by tissue suppliers from the next of kin. All samples were deidentified to protect the identity of the individual. Age, race and cause of death were recorded. The men had no known history of prostate cancer. 9 The validation dataset consisted of 59 patients aged 42 to 92 years with prostate volumes ranging from 22 to 95 cc. Biopsy data on 47 patients was from an 8-core biopsy scheme and on 2 patients was from a saturated 36-core scheme. The following data elements were provided in the validation dataset: prostate volume, number of biopsy cores, number of positive cores, positive core locations, core lengths, tumor Page 4 of 4

5 lengths, percent of tumor involvement in cores, tumor volumes (index and total), and Gleason scores. A biopsy protocol diagram of the prostate labelled with locations of biopsy core was provided with the validation dataset. The biopsy locations were labelled -36, with 6-core systematic biopsy represented by locations labelled -6; 2-core systematic biopsy represented by locations labelled -6 and 3-8; and 8-core systematic biopsy represented by locations - 8. (Figure ) Although our model-based rules also included the case of 24-core biopsies, and we could have extracted 24-core biopsies from the 36-core biopsies, the rule under study recommends 24 cores only as a possibility for prostate volumes above cc, which were lacking in the present dataset. Due to small numbers of cases for some volumes, we combined into groups as needed, resulting in the prostate volume groups and number of cores subgroups 2-3 cc (with 6 cores), 4 cc (with 6, 2, and 8 cores), 5 cc (with 6, 2, and 8 cores), 6-7 cc (with 2 and 8 cores), and 8- cc (with 2 and 8 cores). Using the dataset, we validated the fully structured rule for prostate volumes through 9 cc and numbers of biopsy cores 6, 2, and 8, by extracting 6-core and 2-core biopsies from the actual 8- and 36-core biopsies, and also using the 8-core biopsies as is. We obtained SP, SE, and concordance for our fully-structured rule, by prostate volume and number of biopsy cores. Also, as a matter of interest, we obtained SP, SE, and concordance for just the model-based rule. Validation Approach: Separately by available prostate volumes and numbers of cores, we determined the empirical SP, SE, and concordance of both the fully structured rules and the model-based components. In particular, for each combination of prostate volume and number of cores, we classified the corresponding validation data into a table of form Prostatectomy Result Rule Decision Significant Insignificant Significant A [a] B [b] Insignificant C [c] D [d] where A, B, C, D are the counts using the fully structured rule and a, b, c, d are those using just the model-based component based only on the number of positive cores. For each such table, we reported SP, SE, and concordance values using the formulas SP = D/(B+D), SE = A/(A+C), Page 5 of 4

6 and concordance = (A+D)/(A+B+C+D), and similar formulas with the a, b, c, d counts. High values of SP, SE, and concordance reflected strong performance. RESULTS Figure displays the biopsy protocol diagram labeling the biopsy needle locations 9. Table summarizes the model based decision rule to be validated. The fully structured rule, incorporating tumor length and Gleason score is exhibited in Table 2. Table 3 provides the SP, SE, and concordance values for both the fully structured rule and the component using number of positive cores only, for each combination of prostate volume (or prostate volume range) and number of biopsy cores. For example, for 4 cc prostate volume, this table gives for the fully structured rule: (SP, SE, concordance) = (%, 86%, 9%), (%, %, %), and (%, %, %) for number of cores = 6, 2, and 8, respectively. The corresponding values for the component using number of positive cores only have comparable SP but lower SE and lower concordance. It is seen that the SP values range from 83% (a singular low) to %, the SE values from % (a singular low) to 5%, and the concordance values from 78% to %. Also, as an overall summary, the table provides (SP, SE, concordance) for the group of 6-core biopsies combining the counts for the separate prostate volumes, and likewise for the 2- and 8-core biopsies. These (SP, SE, concordance) values are (%, 84%, 92%), (9%, 88%, 89%), and (%, 88%, 94%), respectively. These results show that the SP and SE of our fully structured rules with this particular data set are highly competitive with those of rules in current practice. Also, the results are consistent with those of the model-based component provided in Table. These findings confirm with this data set that our fully structured rule based on probability modeling should, if applied in practice, be effective in lessening both over- and under-treatment. DISCUSSION The findings here show that our fully structured decision rule with a model-based component performs very well with the present prostatectomy dataset. For the 6-core biopsies, which were used only for prostate volumes 2-5 cc, the SP is % in all cases and the SE ranged from 75% to 87%. For the 2-core biopsies, used for volumes 4- cc, the SP ranged from 83% to % and the SE was % in all cases except for one (6-7 cc with SE = 62%). Page 6 of 4

7 For the 8-core biopsies, used for volumes 4- cc, the SP is % in all cases and the SE is % in all cases except for one (6-7 cc with SE = 62%). Looking at all cases together for the 6-core, 2-core, and 8-core biopsies, the corresponding (SP, SE) values are (%, 84%), (9%, 88%), and (%, 88%), respectively. This favorable finding supports the fundamental design of our fully structured rule, which prioritizes on SP in order to avoid over-treatment, uses more cores with larger prostates to obtain reasonable SE in order to avoid under-treatment, and includes a component based on the number of positive cores as data. It is also seen from Table 3 that by itself the latter component is similar in SP but lower in SE, as expected. Being derived using a probability model, rather than from a particular dataset, our rule possesses generality of potential application. It is reassuring to see that it performs well with this particular data set, the only one investigated to date, and this suggests going forward with validation studies using other available data sets, including prostate volumes -2 cc. For comparison with other methods in current practice, a review of leading methods indicates (SP, SE) such as (%, 4%), (99%, 7%), (99%, 34%), (98%, 53%), (98%, 52%), (98%, 23%), (97%, 67%), (96%, 5%), (96%, 27%), (95%, 56%), (89%, 33%), (78%, 7%), and (75%, 77%), with the two in bold including prostate volume as an input in the statistical model. These were derived, typically, by fitting logistic regression models to particular data sets, the different optimal fitted models differ across the different data sets of patients, thus making it problematic to decide which to use with a patient. Also, these concern a range of 6 to 2 biopsy cores. In contrast, the fully structured rule not only compares well in (SP, SE) but also, very importantly, properly adjusts for prostate volume. The following guideline follows from Table. Prostate Volume (cc) Number of cores 6 6, 2, or 8 2 or 8 8 or 24 We recognize that biopsy protocols used for primary diagnosis of CaP have changed over the last few years due to the development of magnetic resonance imaging (MRI), which enables detection of most high grade and large tumors Image fusion software currently available allows biopsy cores to be targeted within specific suspicious areas detected solely by MRI, whatever the gland volume. Studies have reported that MRI targeted biopsies significantly outperform conventional systematic biopsies in the detection of significant CaP 8,22. However, some studies have also recommended that systematic biopsies should not be omitted for optimal staging of disease, as long as negative predictive value for MRI is still imperfect 8. Page 7 of 4

8 Despite MRI having high diagnostic accuracy (about 95%) for diagnosing clinically significant CaP, the MRI fusion targeted biopsy cannot replace systematic biopsy as confirmatory biopsy for men enrolled in active surveillance protocol 7. We have reviewed the checklists for Standard of Reporting for MRI-targeted Biopsy Studies (START) 2 criteria, and it will be interesting to compare results from such studies with our findings. However, from some published studies, it is clear that fewer total number of cores are used in MRI-targeted biopsy protocols. 3 Despite the evolution of biopsy protocols through the increased use of MRI targeted approach, the findings in this work are still applicable to populations of patients that are potential candidates for active surveillance. These patients often have normal prostate MRI and have low risk cancer, and usually undergo systematic biopsies at 6 months. Also, many urologists still perform solely systematic biopsies without any pre-biopsy MRI. CONCLUSIONS The model-based fully structured decision rule performs very well when validated with actual prostatectomy data and is consistent with model-based expectations. This suggests the value of this decision rule in judicious practical application. Also, the findings indicate that additional modeling and additional validation studies, for possible further calibration and improvement of the decision rule, would be worthwhile. ACKNOWLEDGMENT The authors gratefully thank Dr. Gabriel Haas, M.D., and Dr. G. L. Thompson, Ph.D., for helpful remarks and encouragement. Page 8 of 4

9 REFERENCES. Bostwick DG, Qian J. High-grade prostatic intraepithelial neoplasia. Modern pathology : an official journal of the United States and Canadian Academy of Pathology, Inc 24;7: Klotz L. Prostate cancer overdiagnosis and overtreatment. Current opinion in endocrinology, diabetes, and obesity 23;2: Loeb S, Bjurlin MA, Nicholson J, et al. Overdiagnosis and overtreatment of prostate cancer. European urology 24;65: Chodak GW, Warren KS. Watchful waiting for prostate cancer: a review article. Prostate cancer and prostatic diseases 26;9: Dahabreh IJ, Chung M, Balk EM, et al. Active surveillance in men with localized prostate cancer: a systematic review. Annals of internal medicine 22;56: Bill-Axelson A, Holmberg L, Garmo H, et al. Radical prostatectomy or watchful waiting in early prostate cancer. The New England journal of medicine 24;37: Serfling R, Ogola G. Probability modeling of the number of positive cores in a prostate cancer biopsy session, with applications. Statistics in medicine Mould RF. Introductory Medical Statistics. 3 ed. Bristol and Philadelphia: Institute of Physics Publishing; Delongchamps NB, de la Roza G, Jones R, Jumbelic M, Haas GP. Saturation biopsies on autopsied prostates for detecting and characterizing prostate cancer. BJU international 29;3: Serfling R, Shulman M, Thompson GL, et al. Quantifying the impact of prostate volumes, number of biopsy cores and 5alpha-reductase inhibitor therapy on the probability of prostate cancer detection using mathematical modeling. The Journal of urology 27;77: Anast JW, Andriole GL, Bismar TA, Yan Y, Humphrey PA. Relating biopsy and clinical variables to radical prostatectomy findings: can insignificant and advanced prostate cancer be predicted in a screening population? Urology 24;64: Moore CM, Kasivisvanathan V, Eggener S, et al. Standards of reporting for MRI-targeted biopsy studies (START) of the prostate: recommendations from an International Working Group. European urology 23;64: Robertson NL, Emberton M, Moore CM. MRI-targeted prostate biopsy: a review of technique and results. Nature reviews Urology 23;: Hu JC, Chang E, Natarajan S, et al. Targeted prostate biopsy in select men for active surveillance: do the Epstein criteria still apply? The Journal of urology 24;92: Grenabo Bergdahl A, Wilderang U, Aus G, et al. Role of Magnetic Resonance Imaging in Prostate Cancer Screening: A Pilot Study Within the Goteborg Randomised Screening Trial. European urology Peltier A, Aoun F, Lemort M, Kwizera F, Paesmans M, Van Velthoven R. MRI-targeted biopsies versus systematic transrectal ultrasound guided biopsies for the diagnosis of localized prostate cancer in biopsy naive men. BioMed research international 25;25: Pepe P, Garufi A, Priolo G, Pennisi M. Can MRI/TRUS fusion targeted biopsy replace saturation prostate biopsy in the re-evaluation of men in active surveillance? World journal of urology Radtke JP, Teber D, Hohenfellner M, Hadaschik BA. The current and future role of magnetic resonance imaging in prostate cancer detection and management. Translational andrology and urology 25;4: Page 9 of 4

10 9. Toner L, Weerakoon M, Bolton DM, Ryan A, Katelaris N, Lawrentschuk N. Magnetic resonance imaging for prostate cancer: Comparative studies including radical prostatectomy specimens and template transperineal biopsy. Prostate international 25;3: Turkbey B, Brown AM, Sankineni S, Wood BJ, Pinto PA, Choyke PL. Multiparametric prostate magnetic resonance imaging in the evaluation of prostate cancer. CA: a cancer journal for clinicians Valerio M, McCartan N, Freeman A, Punwani S, Emberton M, Ahmed HU. Visually directed vs. software-based targeted biopsy compared to transperineal template mapping biopsy in the detection of clinically significant prostate cancer. Urologic oncology 25;33:424.e Radtke JP, Schwab C, Wolf MB, et al. Multiparametric Magnetic Resonance Imaging (MRI) and MRI-Transrectal Ultrasound Fusion Biopsy for Index Tumor Detection: Correlation with Radical Prostatectomy Specimen. European urology 26. Page of 4

11 TABLES AND FIGURES Table. Model-based specificity (SP) and sensitivity (SE) for selected insignificant CaP thresholds (x ), by prostate volume (V) and number of biopsy cores (n). Favorable cases in bold prioritize on high SP and offer n versus SE tradeoffs. n V x (SP, SE) x (SP, SE) x (SP, SE) x (SP, SE) 2 (67, 93) (95, 53) 2 (47, ) (89, 73) 3 (6, 9) (95, 45) 4 (69, 8) (97, 29) (47,) (88, 8) 2 (75, 97) (96, 68) 5 (75, 7) (98, 2) (55, 96) (92, 65) 2 (82, 9) (98, 5) 6 (6, 9) (94, 5) 2 (87, 83) (99,39) 7 (66, 85) (96, 4) (52, 98) (9, 72) 8 (69, 8) (97, 34) (57, 95) (92, 62) 9 (72, 75) (97, 28) (6, 9) (94, 53) (75, 7) (98, 24) (64, 87) (95, 46) (67, 84) (96, 4) (58, 94) (92, 6) 2 (69, 8) (96, 35) (6, 9) (94, 54) 3 (72, 77) (97, 3) (64, 88) (94, 49) 4 (73, 74) (97, 28) (66, 85) (95, 44) 5 (75, 7) (98, 25) (68, 83) (96, 4) 6 (76, 68) (98, 22) (69, 8) (96, 36) 7 (78, 65) (98, 2) (7, 78) (97, 33) 8 (79, 63) (98, 8) (72, 75) (97, 3) 9 (8, 6) (74, 73) 2 Page of 4 (98, 7) (8, 58) (99, 5) (97, 28) (75, 7) (98, 26)

12 Table 2. Fully Structured Decision Rule for Insignificant versus Significant CaP.. Check Gleason sum. If 7 or higher, conclude Significant CaP. Otherwise proceed to next step. 2. Check tumor lengths in the positive biopsy cores. If At least core contains >. cm tumor (or 67%), Or at least 2 cores each contain >.8 cm tumor (or 53%), Or at least 3 cores each contain >.7 cm tumor (or 47%), Or at least 4 cores each contain >.65 cm tumor (or 43%), conclude Significant CaP. Otherwise proceed to next step. 3. Apply model-based decision rules based on number of positive cores, for given prostate volume and number of biopsy cores. Conclude either Significant CaP or Insignificant CaP. Page 2 of 4

13 Table 3. Validation Results for Fully Structured Decision Rule [for Model-Based Component Only] Prostate Number Rule Prostatectomy Result Volume (cc) of Cores Decision Significant Insignificant (SP,SE) Concordance Significant 3 [] [] (, 75) Insignificant [3] 8 [8] [(, 25)] 92 [75] 6 Significant 6 [] [] (, 86) Insignificant [7] 3 [3] [(, )] 9 [3] 4 2 Significant 7 [2] [] (, ) Insignificant [5] 3 [3] [(, 27)] [5] 8 Significant 7 [] [] (, ) Insignificant [6] 3 [3] [(, 4)] [4] 6 Significant 7 [3] [] (, 87) Insignificant [5] 6 [6] [(, 37)] 93 [64] 5 2 Significant 8 [4] [] (83, ) Insignificant [4] 5 [5] [(83, 5)] 93 [64] 8 Significant 8 [4] [] (, ) Insignificant [4] 6 [6] [(, 5)] [7] Significant 5 [4] [] (9, 62) 78 [72] Insignificant 3 [4] 9 [9] [(9, 5)] 8 Significant 5 [4] [] (, 62) 83 [78] Insignificant 3 [4] [] [(, 5)] 8-2 Significant 2 [] [] (,) Insignificant [] 3 [3] [(, 5)] [8] 8 Significant 2 [] [] (,) Insignificant [] 3 [3] [(, 5)] [8] 6 Significant 6 [4] [] (, 84) Insignificant 3 [5] 7 [7] [(, 2)] 92 [58] All 2 Significant 22 [] 2 [2] (9, 88) Insignificant 3 [4] 2 [2] [(9, 44)] 89 [66] 8 Significant 22 [] [] (, 88) Insignificant 3 [5] 22 [22] [(, 4)] 94 [68] Page 3 of 4

14 Figures Figure. The saturation biopsy scheme: cores 8, extended biopsy protocol; cores 6, MPZ; cores 7 2, LPZ; cores 3 8, CZ; cores 9 36 (in red) are additional cores; 9 24, MPZ; 25 3, LPZ; 3 36, CZ. Reprinted with permission from Dr. Nicolas Delongchamps 9. Page 4 of 4