A preoperative MRI-based prediction model for biochemical failure after radical prostatectomy

doi:10.21203/rs.3.rs-1977169/v1

Download PDF

Article

A preoperative MRI-based prediction model for biochemical failure after radical prostatectomy

https://doi.org/10.21203/rs.3.rs-1977169/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 09 Jan, 2023

Read the published version in Scientific Reports →

You are reading this latest preprint version

Objective: To propose and verify whether MRI-based model can reduce the incidence of postoperative biochemical failure (BF) in patients with PCa.

Design: The center retrospectively analyzed 967 patients who accepted prostate bpMRI and radical prostatectomy (RP) from June 2018 to January 2020. Then, the included 446 patients were randomized into a research cohort (n=335) and validation cohort (n=111) at a ratio of 3:1. In addition to clinical variables, MRI models also include MRI parameters. The AUC of ROC and decision curve is analyzed.

Main outcomes: Risk of postoperative BF, defined as persistent high or re-elevated PSA levels in patients with PCa who lack evidence of clinical recurrence.

Result: Overall, in the research cohort (age, 69 [63 - 74] years) and the validation cohort (age, 69 [64 - 74] years), the incidence of postoperative BF was 22.39% and 27.02%, respectively. In the research cohort, the AUC of baseline model and MRI model were 0.780 and 0.857, respectively, with significant difference (P < 0.05). The results in the validation cohort were consistent (0.753 vs. 0.865, P < 0.05). At a 20% risk threshold, The false positive rate of MRI model was lower than that of baseline model (31% [95% CI, 9%-39%] vs 44% [95% CI, 15%-64%]), with the true positive rate has only decreased a little (83% [95% CI,63%-94%] vs 87% [95% CI, 75%-100%]). 32 of 100 RPs can been performed, with no raise in quantity of patients with missed BF.

Conclusion: We developed and verified a MRI-based prediction model that predicts the incidence of BF in patients after RP through preoperative clinical variables and MRI-related variables. This model could be helpful for clinical work.

Prostate cancer

Targeted biopsy

Magnetic resonance imaging

Biomedical failure

Radical prostatectomy

Prostate cancer (PCa) is the second most frequent cancer and the fifth leading cause of cancer death among men in 2020, with an estimated almost 1.4 million new cases and 375,000 deaths worldwide[1]. Generally, radical prostatectomy (RP) was considered to be a valid method for the treatment of localized PCa[2]. An elevated serum PSA is the most sensitive and specific early indicator of PCa recurrence after RP. Biochemical failure (BF) is defined as a persisting detectable PSA level after RP (i.e., persistent PSA) or two consecutive PSA levels increasing 0.2 ng/mL or more after a period of PSA value normalization (i.e., Biochemical Recurrence). It occurs in 30–40% of the patients within 10 years after RP, and is associated with worse cancer-specific outcomes[3].

Multi-parametric magnetic resonance imaging (mpMRI) has been a highly sensitive tool for the detection of clinically significant PCa (CSPCa)[4, 5]. Besides, mpMRI can detect the adverse pathological features of PCa patients, such as extracapsular invasion or lymph node metastasis of PCa [6–10]. Some studies described bi-parametric MRI (bpMRI), as opposed to mpMRI, without dynamic enhancement (DCE) has been showed to have a diagnostic accuracy of PCa that were similar to those of mpMRI[11]. Besides, bpMRI brought huge cost-effectiveness compared with mpMRI, and contributed to the diagnostic process and risk stratification of PCa patients[12]. In our study, we evaluated the add value of bpMRI for BF prediction. We aimed to develop and internally validate a presurgical model including bpMRI parameters along with clinical variables for predicting BF.

Study population for research cohort and validation cohort

We retrospectively analyzed consecutive 967 patients who underwent prostate bpMRI and RP from June 2018 to January 2020. The exclusion criteria were: 1) Patients who did not receive standardized MRI scans or underwent MRI scans from other centers were excluded; 2) patients with neoadjuvant therapies were excluded; 3) Patients with transurethral resection of prostate were excluded; 4) Patients who received hormone or radiotherapy after RP before BF were excluded; 5) Patients with insufficient clinical data and unable to determine the PSA results of postoperative follow-up were excluded. Finally, 446 patients met the inclusion criteria of this study (Figure.1). Then, the included 446 patients were randomized into a research cohort (n = 335) and validation cohort (n = 111) at a ratio of 3:1. All detected lesions were evaluated and classified according to PI-RADS v2.1 guidelines[13]. If a patient had multiple lesions with the same PI-RADS category, the lesion with the largest diameter was regarded as an exponential lesion.

Bpmri Protocols

BpMRI was operated on a 3T MRI system (Verio, Siemens, Germany), which involves only T2WI and DWI (b value: 2000 s/mm²), the dominant sequences in PI-RADS v2.1 for characterizing the transitional and peripheral zones, respectively[13]. Prostate volume was measured by bpMRI. All detected lesions were evaluated by PI-RADSv2.1 scores by the deputy chief or senior physicians. The prostate MRI regional model was defined by the following four-zone method. To trisect prostate along its axis, the lower third of the prostate was defined as the apex zone while the upper third as the basal zone. The middle third was further divided into peripheral and non-peripheral zones. According to four-zone method above, positive zone defined as the major part of the lesion located or a lesion involved more than half of the zone. Therefore, patients with multiple lesions may also have multiple positive zone. Besides, extracapsular extension (EPE) and seminal vesicle invasion (SVI) were recorded.

Prediction Model Design

The baseline model embodied the commonly used clinical variables consisting of age at biopsy, body mass index (BMI), PSA at diagnosis, PSA density (PSAD), suspicious digital rectal examination (DRE) (yes or no), biopsy pathology (ISUP grade) and type of surgical technique (RARP, Robot-Assisted Radical Prostatectomy or LRP, Laparoscopic Radical Prostatectomy). The MRI model included all above predictors plus PI-RADS score (1, 2, 3, 4 or 5), EPE at bpMRI (yes or no), SVI at bpMRI (yes or no), zonal location of suspected lesions (apex region, basal region, central peripheral zone and central non-peripheral zone), maximum diameter of suspected lesion and clinical stage (T1, T2 or T3). The outcome was BF. In our study, Postoperative PSA level was measured first at 1–2 months after RP ,then at 3 month intervals for the 2st year, and an interval exceed 6 months was defined as lost to follow-up.

Statistical Analysis

We developed and validated two multivariable logistic regression models to predict BF after RP. We recalibrated the risk model in the validation cohort by matching logistic regression with the logit of the predictive risk[14]. The calibration slope near 1 indicated correct fitting of the predictive model. The diagnostic correctness of the two models were surveyed and balanced by AUC of ROC. Model fitting was evaluated by calibration plots[14]. False positive rate (FPR) and true positive rate (TPR) evaluated the prediction accuracy of postoperative BF. TPR was the ratio of patients with BF above the threshold, and FPR is the proportion of patients with non-BF above the same threshold. The clinical value of the prediction model was weighed by the ratio of BFs avoided, the net benefit (NB) and the net reduction (NR) in the quantity of false positives (FPs)[15].

We analyzed the 95% CI and SE of prediction ability estimator in every predictive models, and the difference between the two models which from 2000 samples by stochastically selecting patients with substitution. It was necessary to readjust the prediction model and recalculate the prediction risk of every models in every samples in the research cohort. 95% CIs came from 2.5% and 97.5% of the re-sampling distribution. The data for the resampling process included outcome (whether there was postoperative BF) and the unregulated predicted risk analyzed according to every risk models in the validation cohort. In every sample, the simple model used for recalibration needs to be readjusted, and then the predicted risk after calibration must be recalculated. We compared the distribution of variables between the research cohort and the validation cohort. Categorical variables used for χ ² test, we used Wilcoxon test to analyze continuous variables. The above tests were bilateral tests, and P < 0.05 indicated statistical significance.

Study populations

446 consecutive patients with retrospective accumulation satisfactied the inclusion criteria. Then, through random grouping, we divided 335 patients into the research cohort and 111 patients into the validation cohort, and were separately included in the model. Table 1 summarizes the patient demographics for the two cohorts. In the research cohort (median [IQR] age, 69 [63 - 74] years) and the validation cohort (median [IQR] age, 69 [64 - 74] years), the incidence of postoperative BF was 22.39% (n = 75) and 27.02% (n = 30), respectively. Compared with the validation cohort, age at biopsy, BMI, PSA, abnormal DRE, PI-RADS v2.1 category, ISUP grade and surgical technique of the research cohort were similar.

Table 2 summarizes the MRI characteristics for the two cohorts. The research cohort had a similar zonal location of index lesion, maximum diameter of index lesion, MRI EPE, seminal invasion and clinical stage with the validation cohort (P > 0.05).

Prediction model

In the baseline model, PSA, GG3, GG4 and GG5 were independent predictors as clinical variables, and they also had statistical significance in MRI model (Table 3). The risk for BF was positively associated with PSA and increased with GG3, GG4, GG5 and lesion of central peripheral zone. In the research cohort and validation cohort, the calibration plot showed that the MRI model had better fit than the baseline model (Figure 1).

Compared with the baseline model, AUC increased from 0.780 to 0.857 (P < 0.05) in the MRI model in the research cohort (Figure 2A and Table 4). In the validation cohort, compared with the baseline model, AUC increased from 0.753 to 0.865 (P < 0.05) (Figure 3A and Table 5).

The TPR and FPR of the models were indicated in Figure 2B fortheresearch cohort. The TPR and FPR of the calibrated risk models (Table 4) are showed in Table 5 and Figure 3B for the validation cohort. The FPR of MRI model was lower than that of baseline model, and the loss of TPR was the smallest.

Decision Curve Analysis (DCA)

Figures 2C and D showed the NBs and NRs in the quantity of FPs for the research cohort, and Figures 3C and D showed the NBs and NRs in the quantity of FPs for the validation cohort. We then applied the MRI model to the validation cohort. We found that the risk threshold exceeds 10%, a higher NB and NR in the quantity of FPs than the baseline model can be achieved, as well as BF (BF all) for each patient. For instance, at 20% risk cut-off, the NB was 14 (95%CI, 6 – 23) in the two model, 14 (95%CI, 7 – 23) in baseline model, and 18 (95%CI, 11 – 28) in MRI model, and NR in the quantity of FPs was 0 in the all model, 19 (95%CI, 6 – 37) in the baseline model, and 32 (95%CI, 0 – 56) in the MRI model. The NB of the MRI model was identical to 18 BFs per 100 men with-out negative BFs, 4 more than the baseline model. Compared with BFs in all patients with positive MRI results, the NR in the quantity of FPs based on MRI model is equivalent to 32 fewer false BFs per 100 men, while the quantity of undiagnosed BFs does not increase. Overall, 66% (95% CI, 53%-90%) of BFs could been averted, which was better than baseline model under this threshold [53% (95% CI, 33%-76%)]. In clinical work, the risk threshold of RP could be determined after physicians and patients weigh and judge the relative hazard of potentially unnecessary RP and the benefit of finding postoperative BF. So, there was no one risk threshold to decide who demands RP, but a series of risk thresholds. For example, by selecting a 20% risk threshold, a total of 66% of operations can be avoided, while 89% of postoperative BFs can still be identified.

With the emergence of many treatments for localized PCa, preoperative risk stratification of PCa patients were very important. BF is an ideal early prognostic predictor of prostate cancer after RP. Previous study showed that BF occurred when tumor tissue residue at surgery (i.e.,positive margin, Subclinical lymphatic metastasis, etc),or cancer have disseminated beyond the prostate and thus outside the surgical field at the time of operation(i.e.minimal residual disease)[16, 17].

Several commonly used multivariate risk tools based on the categories determined by pre-diagnosis PSA, T stage by DRE and biopsy grading group have been used to predict postoperative PSA results[18, 19]. Several studies support that the inclusion of MRI-derived parameters in a risk model could Increase the accuracy of BCR prediction. For example, a multivariable model including MRI PIRADS, along with clinical and pathological variables outperformed the European Association of Urology classification and CAPRA score for predicting BCR (C-index: 77% vs 62% vs 60%, respectively)[20]. Moreover, in another study[8], a presurgical model incorporating PI-RADS, fusion-targeted biopsy grade and extraprostatic extension on MRI had better accuracy for predicting BCR (AUC: 0.68–0.71) than the D’Amico classification (AUC: 0.66–0.71). However, these findings took biochemical recurrence as the endpoint, and persistent PSA (༞0.2ng/ml) after RP also required preoperative intervention. In Norihito Soga. et al study, Three sub-groups were defined in terms of D’Amico classification risk (low, intermediate, and high) and The GP score(Gleason score multi-plied by prostate-specific antigen). There was no significant difference in the non-BF rate between the low risk and low GP score subgroups or the intermediate risk and intermediate GP score subgroups. But the non-BCFrate of the high GP score subgroup was significantly lower than that of the high-risk subgroup (42.1% VS 66.1%, p = 0.008). Based on multivariate analysis, a high GP score (P = 0.001; HR 3.78; 95%CI 1.95–7.35) was a significant independent risk factor for BCF after prostatectomy. However they are prediction models that are only limited to clinical parameters[21]. In other previous studies, Teloken.P E. et al found that the location of transition zone indicates a better biochemical recurrence (BR) free survival after adjusting for poor clinicopathological features[22]. Nari Shin. et al demonstration the zonal location of lesions on MRI, in addition to PI-RADS category, may be helpful for estimating risks of postoperative BF[9]. The above studies confirmed the role of MRI in predicting BF, but did not further develop a new prediction model. With MRI parameters were included in our prediction model, it showed better model fitting and higher diagnostic accuracy, avoids more BFs, and maintains a similar level of sensitivity to BFs contrast to the baseline model.

We applied a DCA to the 2 risk prediction models to compare net benefits of “treat none” with “treat all”. Treat none refers to RP for localized prostate cancer, while treat all refers to neoadjuvant androgen deprivation, extended radical operation and Lymph node dissection. Due to the higher negative effects of treat-all strategy, such aggressive therapy were performed only when there was exact risk of locally advanced PCa. We chose high risk thresholds for DCA. Compared with “treat all” and “treat none” methods, the net benefit of risk thresholds greater than or equal to 15% was always higher for all models. Furthermore, the novel MRI model demonstrated better calibration characteristics and higher net benefit than baseline model. The above DCA indicated that when the location of index lesions on bpMRI were included in the prediction model, it showed better model fitting and higher predictive accuracy, decreasing unnecessary treatment while increasing sensitivity for BF compared with the baseline model.

Limitation

The results of our proposed model are similar to those previously reported. However, there are several limitations. This study combed from retrospective, single center data and was internally validated. These factors may lead to verification bias and are not universal[23].

We developed and verified a MRI-based prediction model that predicts the incidence of BF in patients after RP through preoperative clinical variables and MRI-related variables. This model could be helpful for clinical work.

DRE, Digital Rectal Examination; IQR, Interquartile Range; PI-RADSv2, Prostate Imaging-Reporting and Data System version 2; PSA, prostate-specific antigen; PSAD, PSA density; ISUP, International Society of Urological Pathology; GG, Grading Group; RARP, Robot-Assisted Radical Prostatectomy ; LRP, Laparoscopic Radical Prostatectomy；EPE, Extraprostatic Extension；AUC, area under the receiver operating characteristic curve; FPR, false-positive rate; MRI, magnetic resonance imaging; NA, not applicable; NB, net benefit; NRFP, net reduction in false-positives; PABF, percentage of avoided biomedical failures; percentage of avoided biopsies; TPR, true-positive rate;

Acknowledgements:

Ethics approval and consent to participate:

All methods were performed in accordance with the relevant guidelines and regulations. This was a retrospective study that received approval by the Hospital Ethics Committee of the First Affiliated Hospital of Nanjing Medical University. A written informed consent was obtained from all subjects.

Consent for publication

Not applicable.

Availability of data and materials

All data generated or analysed during this study are included in this published article [and its supplementary information files]

Competing interest

The authors have no conflicts of interest to declare.

Funding

No.

Authors' contributions

Minjie Pan, Shouchun Li, Fade Liu and Linghui Liang wrote the main manuscript text.

Linghui Liang, Jinwei Shang and Wei Xia prepared figures 1-3.

Minjie Pan, Shouchun Li, Fade Liu，Linghui Liang, Jinwei Shang, Wei Xia, Gong Cheng and Lixin Hua reviewed the manuscript.

Acknowledgements

Thanks to all departments that contributed to this study.

Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F: Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021, 71(3):209–249.
Sanda MG, Cadeddu JA, Kirkby E, Chen RC, Crispino T, Fontanarosa J, Freedland SJ, Greene K, Klotz LH, Makarov DV et al: Clinically Localized Prostate Cancer: AUA/ASTRO/SUO Guideline. Part II: Recommended Approaches and Details of Specific Care Options. The Journal of urology 2018, 199(4):990–997.
Mottet N, Bellmunt J, Bolla M, Briers E, Cumberbatch MG, De Santis M, Fossati N, Gross T, Henry AM, Joniau S et al: EAU-ESTRO-SIOG Guidelines on Prostate Cancer. Part 1: Screening, Diagnosis, and Local Treatment with Curative Intent. European urology 2017, 71(4):618–629.
Meyerson BL, Streicher J, Sidana A: A magnetic resonance imaging-based prediction model for prostate biopsy risk stratification. Ther Adv Urol 2018, 10(11):357–358.
Boesen L: Multiparametric MRI in detection and staging of prostate cancer. Danish medical journal 2017, 64(2).
Mehralivand S, Shih JH, Harmon S, Smith C, Bloom J, Czarniecki M, Gold S, Hale G, Rayn K, Merino MJ et al: A Grading System for the Assessment of Risk of Extraprostatic Extension of Prostate Cancer at Multiparametric MRI. Radiology 2019, 290(3):709–719.
Rajwa P, Mori K, Huebner NA, Martin DT, Sprenkle PC, Weinreb JC, Ploussard G, Pradere B, Shariat SF, Leapman MS: The Prognostic Association of Prostate MRI PI-RADS v2 Assessment Category and Risk of Biochemical Recurrence after Definitive Local Therapy for Prostate Cancer: A Systematic Review and Meta-Analysis. The Journal of urology 2021, 206(3):507–516.
Manceau C, Beauval JB, Lesourd M, Almeras C, Aziza R, Gautier JR, Loison G, Salin A, Tollon C, Soulie M et al: MRI Characteristics Accurately Predict Biochemical Recurrence after Radical Prostatectomy. J Clin Med 2020, 9(12).
Shin N, Park SY: Postoperative Biochemical Failure in Patients With PI-RADS Category 4 or 5 Prostate Cancers: Risk Stratification According to Zonal Location of an Index Lesion. AJR American journal of roentgenology 2020, 215(4):913–919.
Dell'Oglio P, Stabile A, Dias BH, Gandaglia G, Mazzone E, Fossati N, Cucchiara V, Zaffuto E, Mirone V, Suardi N et al: Impact of multiparametric MRI and MRI-targeted biopsy on pre-therapeutic risk assessment in prostate cancer patients candidate for radical prostatectomy. World journal of urology 2019, 37(2):221–234.
Kang Z, Min X, Weinreb J, Li Q, Feng Z, Wang L: Abbreviated Biparametric Versus Standard Multiparametric MRI for Diagnosis of Prostate Cancer: A Systematic Review and Meta-Analysis. AJR American journal of roentgenology 2019, 212(2):357–365.
Porter KK, King A, Galgano SJ, Sherrer RL, Gordetsky JB, Rais-Bahrami S: Financial implications of biparametric prostate MRI. Prostate cancer and prostatic diseases 2020, 23(1):88–93.
Turkbey B, Rosenkrantz AB, Haider MA, Padhani AR, Villeirs G, Macura KJ, Tempany CM, Choyke PL, Cornud F, Margolis DJ et al: Prostate Imaging Reporting and Data System Version 2.1: 2019 Update of Prostate Imaging Reporting and Data System Version 2. European urology 2019, 76(3):340–351.
Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, Pencina MJ, Kattan MW: Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology 2010, 21(1):128–138.
Vickers AJ, Elkin EB: Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making 2006, 26(6):565–574.
Murray NP, Aedo S, Fuentealba C, Reyes E, Salazar A, Lopez MA, Minzer S, Orrego S, Guzman E: Minimal Residual Disease Defines the Risk and Time to Biochemical Failure in Patients with Pt2 and Pt3a Prostate Cancer Treated With Radical Prostatectomy: An Observational Prospective Study. Urology journal 2020, 17(3):262–270.
Murray NP, Aedo S, Fuentealba C, Reyes E, Salazar A: Minimum Residual Disease in Patients Post Radical Prostatectomy for Prostate Cancer: Theoretical Considerations, Clinical Implications and Treatment Outcome. Asian Pacific journal of cancer prevention: APJCP 2018, 19(1):229–236.
D'Amico AV, Whittington R, Malkowicz SB, Schultz D, Blank K, Broderick GA, Tomaszewski JE, Renshaw AA, Kaplan I, Beard CJ et al: Biochemical outcome after radical prostatectomy, external beam radiation therapy, or interstitial radiation therapy for clinically localized prostate cancer. JAMA 1998, 280(11):969–974.
Cooperberg MR, Pasta DJ, Elkin EP, Litwin MS, Latini DM, Du Chane J, Carroll PR: The University of California, San Francisco Cancer of the Prostate Risk Assessment score: a straightforward and reliable preoperative predictor of disease recurrence after radical prostatectomy. The Journal of urology 2005, 173(6):1938–1942.
Morka N, Simpson BS, Emberton M, Norris JM: Re: Giorgio Gandaglia, Guillaume Ploussard, Massimo Valerio, et al. Prognostic Implications of Multiparametric Magnetic Resonance Imaging and Concomitant Systematic Biopsy in Predicting Biochemical Recurrence After Radical Prostatectomy in Prostate Cancer Patients Diagnosed with Magnetic Resonance Imaging-targeted Biopsy. Eur Urol Oncol 2020;7:739 – 47. European urology oncology 2021, 4(1):127–128.
Soga N, Ogura Y, Wakita T, Kageyama T, Furusawa J: The GP Score, a Simplified Formula (Bioptic Gleason Score Times Prostate Specific Antigen) as a Predictor for Biochemical Failure after Prostatectomy in Prostate Cancer. Curr Urol 2019, 13(1):25–30.
Teloken PE, Li J, Woods CG, Cohen RJ: The Impact of Prostate Cancer Zonal Origin on Pathological Parameters at Radical Prostatectomy and Subsequent Biochemical Failure. The Journal of urology 2017, 198(6):1316–1323.
Tangen CM, Goodman PJ, Till C, Schenk JM, Lucia MS, Thompson IM, Jr.: Biases in Recommendations for and Acceptance of Prostate Biopsy Significantly Affect Assessment of Prostate Cancer Risk Factors: Results From Two Large Randomized Clinical Trials. J Clin Oncol 2016, 34(36):4338–4344.

Table 1. Patient Demographics of Research and Validation Cohort

Variable	Research Cohort (n=335)	Validation Cohort (n=111)	P Value^a
Age at biopsy, Median (IQR)	69 (63 - 74)	69 (64 -74)	0.495
BMI(Kg/m²), Median (IQR)	24.22 (22.49 – 26.03)	23.94 (22.15 – 26.67)	0.965
Preoperative PSA(ng/dL), Median (IQR)	10.53 (7.10 – 16.51)	10.13 (7.22 – 14.38)	0.579
PSAD(ng/ml² ), Median (IQR)	0.30 (0.18 – 0.53)	0.27 (0.18 – 0.43)	0.275
Abnormal DRE, n (%)	65 (19.40)	16 (14.41)	0.237
PI-RADS v2.1 category, n (%)
1	14 (4.18)	6 (5.41)	0.222
2	14 (4.18)	9 (8.11)
3	89 (26.57)	20 (18.02)
4	132 (39.40)	49 (44.14)
5	86 (25.67)	27 (24.32)
ISUP grade, n (%)
GG1	101 (30.15)	31 (27.93)	0.685
GG2	69 (20.60)	20 (19.96)
GG3	73 (21.79)	25 (21.97)
GG4	77 (22.99)	32 (24.44)
GG5	15 (4.48)	3 (4.04)
Surgical technique, n (%)
RARP	263 (78.51)	87 (78.38)	0.977
LRP	72 (21.49)	24 (21.62)
Biochemical failure n (%)	75 (22.39)	30 (27.02)	0.318

Table 2.MRI Characteristics of Research and Validation Cohort

MRI Variable	Research Cohort	Validation Cohort	P value
Zonal location of index lesions
Negative MRI	28 (8.36)	15 (13.51)	0.111
Prostatic apex region	165 (49.25)	46 (41.44)	0.153
Basal region	42 (12.54)	15 (13.51)	0.789
Peripheral zone	171 (51.04)	54 (48.65)	0.662
Central non-peripheral zone	109 (32.54)	29 (26.13)	0.205
Maximum diameter of index lesion at bpMRI (cm), median (IQR)	1.10 (0.70 – 1.60)	1.10 (0.80 – 1.50)	0.275
MRI EPE, n (%)	43 (12.84)	14 (12.61)	0.951
Seminal invasion, n (%)	10 (2.99)	2 (1.98)	0.741
Clinical stage, n (%)
T1	27 (8.06)	14 (12.61)	0.296
T2	262 (78.21)	80 (72.07)
T3	46 (13.73)	17 (15.3)

EPE, Extraprostatic Extension

Table 3. Logistic Regression Prediction Models of Biochemical Failure for Research Cohort

	Baseline Model			MRI Model
Characteristic	Coefficient	OR (95%CI)	P value	Coefficient	OR (95%CI)	P value
Intercept	1.106	3.023 (0.047 – 189.847)	0.599	1.905	6.722e+00 (6.394e-02 - 6.726e+02)	0.417
Age at biopsy	-0.036	0.965 (0.923 – 1.008)	0.111	-0.035	9.568e-01 (9.205e-01 - 1.013e+00)	0.150
BMI	-0.093	0.911 (0.816 – 1.015)	0.095	-0.111	8.952e-01 (7.931e-01 - 1.007e+00)	0.068
PSA	0.048	1.050 (1.015 – 1.089)	0.007	0.047	1.048e+00 (1.009e+00 - 1.091e+00)	0.017
PSAD	0.295	1.343 (0.413 – 4.159)	0.606	-0.335	7.155e-01 (1.953e-01 - 2.519e+00)	0.599
Abnormal DRE	0.482	1.619 (0.748 – 3.410)	0.211	0.132	1.141e+00 (4.725e-01 - 2.635e+00)	0.762
Surgical technique
RARP	NA	NA	NA	NA	NA	NA
LRP	0.261	1.298 (0.615 – 2.652)	0.482	0.201	1.223e+01 (5.402e-01 - 2.680e+00)	0.620
ISUP grade
GG1	NA	NA	NA	NA	NA	NA
GG2	0.521	1.683 (0.539 – 5.503)	0.371	0.302	1.352e+00 (3.946e-01 - 4.817e+00)	0.631
GG3	1.182	3.260 (1.181 –9.999)	0.028	0.778	2.177e+00 (6.707e-01 - 7.681e+00)	0.205
GG4	1.966	7.145 (2.791 - 20.969)	<0.001	1.755	5.784e+00 (1.990e+00 - 1.937e+01)	0.002
GG5	4.097	60.133 (11.908 - 476.667)	<0.001	3.682	3.974e+01 (6.481e+00 - 3.557e+02)	<0.001
Prostatic apex region				0.568	1.764e+00 (8.422e-01 - 3.760e+00)	0.135
Basal region				0.300	1.350e+00 (4.589e-01 - 3.745e+00)	0.573
Central peripheral zone				0.912	2.490e+00 (1.112e+00 - 5.803e+00)	0.030
Central non-peripheral zone				0.050	1.050e+00 (4.116e-01 - 2.615e+00)	0.917
Maximum diameter of index lesion				0.474	1.607e+00 (7.568e-01 - 3.582e+00)	0.229
MRI EPE				-1.740	1.755e-01 (3.243e-03 - 6.841e+00)	0.364
Seminal invasion				0.315	1.370e+00 (1.225e-01 - 2.730e+01)	0.808
PI-RADS v2.1 category
1				NA	NA	NA
2
3				4.404	8.175e+01 (1.475e-72 - NA)	0.875
4				4.878	1.313e+02 (2.486e-72 - NA)	0.861
5				4.846	1.273e+02 (4.078e-72 - NA)	0.862
Clinical stage, n (%)
T1				NA	NA	NA
T2				-6.311	1.816e-03 (NA - 9.966e+70)	0.821
T3				-4.011	1.812e-02 (NA - 1.827e+72)	0.886

Table 4. Performance of the two Risk Prediction Models in the Research Cohort

		Model		Comparison
Performance Parameter	Risk Threshold, %	Baseline	MRI	MRI vs Baseline
AUC (95% Cl)	NA	78 (62 – 85)	86 (75 – 94)	8 (1 to 32)^a
TPR, % (95% Cl)	10	99 (85 – 98)	85 (78 – 96)	-14 (-20 to 11)
	15	84 (78 – 98)	80 (72 – 96)	-4 (-9 to 1)
	20	86 (74 – 98)	81 (61 – 90)	-5 (-10 to 2)
FPR, % (95% Cl)	10	74 (28 – 94)	52 (14 – 60)	-22 (-39 to -8)
	15	58 (23 – 83)	41 (11 – 44)	-17 (-25 to 4)
	20	44 (14 – 62)	32 (10 – 40)	-12 (-29 to -1)
NB, % (95% Cl)	10	15 (11 – 24)	24 (13 – 34)	9 (-2 to 13)
	15	20 (13 – 32)	25 (15 – 36)	5 (-8 to 20)
	20	18 (9 – 30)	20 (12 – 32)	2 (0 to 5)
NRFP, % (95% Cl)	10	7 (0 – 12)	-1 (-4 – 8)	-8 (-14 to 3)
	15	13 (0 – 23)	8 (0 – 15)	-5 (-9 to 19)
	20	14 (2 – 30)	30 (0 – 50)	16 (8 to 26)
PABF, % (95% Cl)	10	60 (52 – 80)	80 (70 – 90)	20 (14 to 27)
	15	50 (41 – 82)	67 (61 – 93)	17 (9 to 27)
	20	57 (38 – 80)	69 (55 – 92)	12 (0 to 25)

Abbreviations: AUC, area under the receiver operating characteristic curve; FPR, false-positive rate; MRI, magnetic resonance imaging;

NA, not applicable; NB, net benefit; NRFP, net reduction in false-positives; PABF, percentage of avoided biomedical failures; percentage of avoided biopsies; TPR, true-positive rate. ^aP < 0.05 for the comparison of AUCs.

Table 5. Performance of the two Risk Prediction Models in the Validation Cohort

		Model		Comparison
Performance Parameter	Risk Threshold, %	Baseline	MRI	MRI vs Baseline
AUC (95% Cl)	NA	75 (63 – 82)	86 (76 – 93)	11 (5 to 23)^a
TPR, % (95% Cl)	10	100 (86 – 100)	87 (82 – 100)	-13 (0 to 19)
	15	87 (80 – 100)	83 (72 – 97)	-3 (0 to 28)
	20	87 (75 – 100)	83 (63 – 94)	-3 (0 to 37)
FPR, % (95% Cl)	10	75 (29 – 96)	53 (16 – 63)	-22 (-67 to -12)
	15	59 (22 – 82)	43 (13 – 47)	-16 (-69 to 0)
	20	44 (15 – 64)	31 (9 – 39)	-13 (-55 to 0)
NB, % (95% Cl)	10	17 (12 – 28)	23 (15 – 31)	6 (-2 to 12)
	15	15 (10 – 25)	18 (12 – 29)	3 (0 to 10)
	20	14 (7 – 23)	18 (11 – 28)	3 (0 to 9)
NRFP, % (95% Cl)	10	13 (0 – 18)	0 (0 – 14)	-13 (-18 to 1)
	15	17 (0 – 28)	13 (0 – 20)	-3 (-10 to 2)
	20	19 (6 – 37)	32 (0 – 56)	13 (-6 to 20)
PABF, % (95% Cl)	10	62 (56 – 85)	84 (71 – 94)	22 (8 to 38)
	15	55 (42 – 80)	66 (60 – 91)	11 (6 to 30)
	20	53 (33 – 76)	66 (53 – 90)	13 (0 to 31)

Abbreviations: AUC, area under the receiver operating characteristic curve; FPR, false-positive rate; MRI, magnetic resonance imaging;

No competing interests reported.

Download PDF

Journal Publication

published 09 Jan, 2023

Read the published version in Scientific Reports →

Editorial decision: Major revision
27 Oct, 2022
Reviews received at journal
14 Oct, 2022
Reviewers agreed at journal
11 Oct, 2022
Reviewers invited by journal
28 Sep, 2022
Editor assigned by journal
27 Sep, 2022
Editor invited by journal
20 Aug, 2022
Submission checks completed at journal
20 Aug, 2022
First submitted to journal
19 Aug, 2022

You are reading this latest preprint version

A preoperative MRI-based prediction model for biochemical failure after radical prostatectomy

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Methods

Study population for research cohort and validation cohort

Bpmri Protocols

Prediction Model Design

Statistical Analysis

Results

Discussion

Limitation

Conclusions

Abbreviations

Declarations

References

Tables

Additional Declarations

Status:

Journal Publication

Version 1