Novel nomogram developed for determining suitability of metastatic castration-resistant prostate cancer patients to receive maximum benefit from radium-223 dichloride treatment—Japanese Ra-223 Therapy in Prostate Cancer using Bone Scan Index (J-RAP-BSI) Trial

To develop a novel nomogram for determining radium-223 dichloride (Ra-223) treatment suitability for metastatic castration-resistant prostate cancer (mCRPC) patients. This Japanese Ra-223 Therapy in Prostate Cancer using Bone Scan Index (J-RAP-BSI) Trial was a retrospective multicenter investigation enrolled 258 mCRPC patients in Japan with Ra-223 treatment between June 2016 and August 2020, with bone scintigraphy findings before treatment, clinical data, and survival outcome available. A nomogram was constructed using prognostic factors for overall survival (OS) based on a least absolute shrinkage and selection operator Cox regression model. A sub-analysis was also conducted for patients meeting European Medicines Agency (EMA) guidelines. Within a median of 17.4 months after initial Ra-223 treatment, 124 patients (48.1%) died from prostate cancer. Predictive factors included (1) sum of prior treatment history (score 0, never prior novel androgen receptor-targeted agents (ARTA) therapy, never prior taxane-based chemotherapy, and ever prior bisphosphonate/denosumab treatment), (2) Eastern Cooperative Oncology Group (ECOG) performance status, (3) prostate-specific antigen doubling time (PSADT), (4) hemoglobin, (5) lactate dehydrogenase (LDH), and (6) alkaline phosphatase (ALP) levels, and (7) automated bone scan index (aBSI) value based on bone scintigraphy. The nomogram using those factors showed good discrimination, with apparent and optimism-corrected Harrell’s concordance index values of 0.748 and 0.734, respectively. Time-dependent area under the curve values at 1, 2, and 3 years were 0.771, 0.818, and 0.771, respectively. In 227 patients meeting EMA recommendation, the nomogram with seven factors showed good discrimination, with apparent and optimism-corrected Harrell's concordance index values of 0.722 and 0.704, respectively. Time-dependent area under the curve values at 1, 2, and 3 years were 0.747, 0.790, and 0.759, respectively. This novel nomogram including aBSI to select mCRPC patients to receive Ra-223 with significantly prolonged OS possibility was found suitable for assisting therapeutic decision-making, regardless of EMA recommendation.


Introduction
Radium-223 dichloride (Ra-223) is a first-in-class alpha particle-emitting radiopharmaceutical used as therapy for metastatic castration-resistant prostate cancer (mCRPC) and symptomatic bone metastasis and known to have minimal myelosuppressive effects [1]. Alpha particles have a relatively large size, and when emitted using high linear energy transfer a short path length and localized area of cell destruction is the result (< 100 μm, 2-10 cell diameters), which induces predominantly nonrepairable double-stranded DNA breaks [2]. The randomized phase 3 ALSYMPCA study reported that Ra-223 plus best standard of care prolonged median overall survival (OS) by 3.6 months as compared to a placebo plus best standard of care [median 14.9 vs 11.3 months; hazard ratio (HR) = 0.70, 95% confidence interval (CI) 0.58-0.83, p < 0.001] [3]. Additionally, that study showed that Ra-223 treatment was well tolerated and associated with a low incidence of grade 3 or 4 myelosuppression (Ra-223 vs. placebo: anemia 13% vs. 13%, neutropenia 2% vs. 1%, thrombocytopenia 7% vs. 2%). Analyses conducted thereafter also demonstrated a survival benefit when used for the treatment of chemotherapy-naïve as well as post-chemotherapy mCRPC patients [4].
Despite increasing clinical experience with Ra-223 given to mCRPC patients, clinical variables useful for predicting response to that radiopharmaceutical and also survival remain difficult to identify. Scoring systems reported include baseline Eastern Cooperative Oncology Group (ECOG) performance status, levels of hemoglobin, alkaline phosphatase (ALP), lactate dehydrogenase (LDH), and prostate-specific antigen (PSA), PSA doubling time (PSADT), and completion of six cycles of Ra-223, as well as other factors [5][6][7][8][9][10][11][12][13][14][15][16][17]. However, a validated and simple model for clinicians to assess which patients will most likely benefit from Ra-223 treatment remains to be elucidated. A validated tool that takes into consideration patient characteristics as well as biomarkers is needed to predict OS, thus allowing for a more careful selection of those eligible for the greatest benefit from such treatment.
Automated bone scan index (aBSI) technology was developed as a way to quantify the extent of skeletal tumor burden shown by bone scintigraphy as a percentage of the total skeletal weight [18] and is considered a valuable metric and potentially helpful tool for estimating total quantitative skeletal metastatic burden in patients with mCRPC. Along with rapid processing time, this automated methodology that uses artificial intelligence has been demonstrated to be accurate and reproducible [19]. In a recent prospective multi-institutional phase III study with 721 mCRPC patients, aBSI was clinically validated for use as a prognostic biomarker [20]. Furthermore, several studies have evaluated pretreatment aBSI as a prognostic imaging biomarker for Ra-223 treatment and found baseline values to be significantly associated with OS in mCRPC patients treated with Ra-223 [6,7,[9][10][11].
In 2018, a formal warning promoted by the European Medicines Agency (EMA) limited the Ra-223 prescription to mCRPC patients pre-treated with at least two systemic therapies for mCRPC or ineligible for any systemic treatment and with more than six osteoblastic lesions at bone scan [21]. This moved Ra-223 treatment to the later stages of mCRPC disease, making the patient selection process even more challenging [22,23]. To our knowledge, only one group has evaluated the prognostic factors potentially able to select patients most likely to benefit from Ra-223 in the scenario of EMA use [24] and this topic has not been discussed in detail. The aim of the present multicenter retrospective study was to develop a novel nomogram system including aBSI for OS prediction in patients with mCRPC who underwent Ra-223 therapy and moreover in patients who met EMA guidelines.  and bone metastasis who  underwent bone scintigraphy before treatment with Ra-223 between June 2016 and August 2020 were included. Bone scintigraphy was performed no more than two months before the first Ra-223 cycle and baseline laboratory tests included determinations of ALP, LDH, PSA, hemoglobin, neutrophil, and platelet levels. The established Ra-223 treatment schedule consists of six intravenous injections (55 kBq/kg of body weight) administered every 28 days in mCRPC patients without disease progression (PSA progression, clinical progression, or the appearance of visceral metastasis on CT scan), unacceptable toxicity, or declining performance status, or by patient request. Patients included in the present study were males > 18 years of age with symptomatic bone metastasis, no visceral metastasis or lymph nodes ≥ 3 cm, hemoglobin > 8.4 gm/dL, white blood cell count > 1.5 × 10 9 , and platelet count > 100 × 10 9 at the first cycle of Ra-223 therapy. mCRPC was defined based on disease progression despite androgen-deprivation therapy, with testosterone serum level < 50 ng/ml and with one or any combination of continuous rise in serum level of PSA, progression of pre-existing disease, or appearance of new metastasis [25]. Patients were excluded if they had received taxane-based chemotherapy within the previous four weeks, or showed impaired kidney function (creatinine clearance < 50 mL/min) or impaired liver function (aspartate aminotransferase (AST) > 100 IU/L, alanine aminotransferase (ALT) > 100 IU/L or γ-Glutamyl TransPeptidase (γ-GTP) > 200 IU/L), or inflammatory bowel disease. Administration of androgen deprivation therapy was continued during Ra-223 therapy and other medications, including denosumab or bisphosphonates, were prescribed at the discretion of the attending physician. The enrolled patients may have been previously given a systemic therapy line for mCRPC, such as abiraterone, enzalutamide, docetaxel, or cabazitaxel. Concomitant treatment with abiraterone or chemotherapy was not permitted during Ra-223 treatment for this cohort. The Common Terminology Criteria for Adverse Events (CTCAE), version 4.03, was used for grading adverse events [26].

aBSI
All patients underwent bone scintigraphy examinations with technetium-99m methylene diphosphonate ( 99m Tc-MDP; PDRadiopharma Inc., Tokyo, Japan) at no more than two months before the first Ra-223 cycle, with the skeletal disease burden defined expressed based on the aBSI with use of BONENAVI®, version 2 (PDRadiopharma, Inc, Tokyo, Japan), a specially developed software package [27]. The methodology for this automated platform has been previously described in detail [18]. Briefly, different anatomic regions of the skeleton were segmented, followed by the detection of abnormal hot spots and classification as metastatic lesions. The weight fraction of the skeleton for each metastatic hot spot was calculated, then the aBSI value calculated as the sum of all such fractions. A representative case is presented in Fig. 1.

Statistical analysis
For patient characteristics, continuous variables were summarized using median and range, while categorical variables were described using frequency and percentage. The least absolute shrinkage and selection operator (LASSO) Cox regression model [28] was applied to examine potential prognostic factors related to OS, which was determined based on the first Ra-223 administration until death from prostate cancer. Potential prognostic factors related to OS included initial stage (II, III/IV), prior treatment history sum score, ECOG performance status (0/1, 2, 3), PSADT (< 90/ ≥ 90 days), hemoglobin, platelet count, LDH, ALP, PSA, and aBSI. In univariate Cox regression models, hemoglobin, platelet count, and aBSI showed nonlinearity with the outcome; thus, changepoint regression models were assumed for hemoglobin and platelet count, with the risk increased in a linear manner when those values were below 13 g/dL and 15.8 × 10 9 /L, respectively, the lower limit of normal for each. For aBSI, a change-point regression model was assumed, with the risk reaching a plateau at 5. These change-point regression models were considered to fit as well as the more flexible modeling with a restricted cubic spline function. The LASSO Cox regression model with optimal shrinkage parameter used to minimize the partial likelihood deviance, estimated by tenfold cross-validation, was applied for these potential prognostic factors. As a prognostic factor for OS, a non-zero regression coefficient was employed.
A nomogram for the prediction of 1-, 2-, and 3-year survival rates was drawn according to the obtained LASSO Cox regression model. To assess the overall discriminative capability and internal validity of the predictive model, apparent concordance index (C-index) and optimismcorrected C-index values, respectively, were calculated based on the bootstrap method. Time-dependent area under the curve (AUC) values were determined based on 500 bootstrap samples and used as a measure of the predictive ability of 1-, 2-, and 3-year survival. Agreement between predicted and observed 2-year survival rates was evaluated using a calibration plot based on 500 bootstrap samples. Additionally, to evaluate the clinical usefulness of the prediction model, a decision curve analysis of 2-year survival was performed.
Statistical analyses were performed using the R software, version 4.0.3 (www. Rproj ect. org) and SPSS, version 21.0, packages. The "glmnet" package was used with the LASSO Cox regression model. Apparent and optimism-corrected C-indexes were calculated based on Sect. 23.7.1 of the book by Steyerberg regarding clinical prediction models to evaluate the overall predictive ability and internal validity of the prediction model [29]. The nomogram was drawn using the "rms" package, while the "hdnom" package was used to calculate time-dependent AUC and draw the calibration plot. With the "stdca" package, decision curve analysis was performed to evaluate the clinical usefulness of the nomogram, with net benefits calculated for different probability thresholds. Net benefit for survival analysis was defined according to Vickers, et al. [30].

Nomogram model
Based on the LASSO Cox regression model with seven predictors, a nomogram was constructed to predict 1-, 2-, and 3-year survival rates (Fig. 2). For assessing the overall discriminative capability of the nomogram, the apparent C-index was 0.748, while the optimism-adjusted C-index for evaluating internal validity was 0.734, indicating acceptable discrimination. The mean time-dependent AUC values at 1, 2, and 3 years obtained from 500 bootstrap samples were 0.771, 0.818, and 0.771, respectively (Fig. 3a).
A calibration plot used for assessing the consistency of the observed survival rate after 2 years and predictions obtained with the use of the nomogram is illustrated in Fig. 3b. Based on the nomogram, the lower survival rate prediction for 2 years was slightly overestimated at approximately 30%, as the actual rate ranged from approximately 10% to 30%. Similarly, the higher 2-year survival rate was slightly underestimated at approximately 75%, with the actual rate ranging from approximately 75 to 90%. On the other hand, the predicted 2-year survival rate was approximately 60%, which had good agreement with the observed values.
Results of decision curve analysis for 2-year survival are presented in Fig. 3c. Satisfactory positive net benefits were obtained as compared with death and survival predictions for all of the patients based on most of the threshold probabilities, indicating the potential clinical usefulness of this novel nomogram.   Table 3. Seven factors, including sum score of prior treatment history, ECOG performance status, levels of PSADT, hemoglobin, LDH, and ALP, and aBSI value, were selected as prognostic factors for OS. Based on the LASSO Cox regression model with six predictors, a nomogram was constructed to predict 1-, 2-, and 3-year survival rates (Fig. 4).
For assessing the overall discriminative capability of the nomogram, the apparent C-index was 0.722, while the optimism-adjusted C-index for evaluating internal validity was 0.704, indicating acceptable discrimination. The mean time-dependent AUC values at 1, 2, and 3 years obtained from 500 bootstrap samples were 0.747, 0.790, and 0.759, respectively (Fig. 5a). A calibration plot used for assessing the consistency of the observed survival rate after 2 years and predictions obtained with use the nomogram is illustrated in Fig. 5b. Results of decision curve analysis for 2-year survival are presented in Fig. 5c.

Discussion
The present multicenter cohort of mCRPC patients who underwent Ra-223 therapy not associated with a trial and with aBSI results derived from pretreatment bone scintigraphy available is the largest group known to be investigated. The findings demonstrated the potential real-world impact of this therapy. Moreover, the novel nomogram established for the selection of these patients, regardless of EMA recommendation, and considered suitable to obtain maximum OS from treatment with Ra-223, is the first such calculating device reported.
There are various reports that used multivariate analysis to examine clinical factors associated with good OS for mCRPC patients treated with Ra-223, such as the study by Saad showing ECOG performance status of 0, no or mild pain, normal ALP, higher hemoglobin (≥ 10 g/dL) with concomitant use of enzalutamide or denosumab [5]. In the report presented by Fosbøl, lower aBSI (< 5.0%) and normal LDH were noted [7], while ECOG performance status of 0 and lower LDH (< 250 U/L) were indicated in the paper by van der Doelen [8]. In other studies, Frantellizzi reported ECOG performance status of 0 and lower aBSI (< 3.0%)'s [9], and Nakashima lower aBSI (< 2.0%) [10], while Anand noted normal ALP and lower aBSI [11], Frantellizzi good ECOG performance status (0-1), normal hemoglobin To obtain the probability of progression predicted by the nomogram, values for the patients were located at each axis and then a vertical line was drawn to the "Points" axis to determine how many points to attribute to each variable. The total points for all variables were calculated, then the sum on the "Total points" line was used to assess individual probability using values predicting risk for OS  performance status of 0-1, no or mild pain, lower PSA, normal ALP, no prior abiraterone/enzalutamide therapy, and low aBSI (< 5.0%). These various reports have indicated several prognostic indicators, though the details presented differ and no single method has yet to be established. Thus, it is considered that the novel nomogram developed in the present study to predict prognosis following Ra-223 treatment is important for the selection of eligible patients who would receive the greatest benefit.
In 2018, the EMA limited the Ra-223 prescription to patients pre-treated with at least two systemic therapies for mCRPC or ineligible for any systemic treatment and more than six bone lesions. This moved Ra-223 treatment to the later stages of the disease, making the patient selection process even more difficult. Jarvis et al. [23] evaluated the survival benefit of EMA guidance using 121 patients who received one prior systemic therapy before Ra-223 (group 1) and 70 patients who received two prior therapies (group 2) and found median OS in group 1 was significantly improved, compared to group 2 (448 days vs. 341 days (p = 0.003)). Under the current EMA guidance, patients are offered Ra-223 at a later stage in their treatment pathway, and suitable patients who might benefit from Ra-223 would be disadvantaged if it is used later in the mCRPC treatment pathway. Moreover, only one group has evaluated the prognostic factors potentially able to select patients most likely to benefit from Ra-223 in the scenario of EMA use. Bauckneht et al. [24] evaluated the several prognostic values in 494 mCRPC patients treated with Ra-223 before and after EMA restricted use and found 1) 89 patients treated with Ra-223 after the EMA restriction showed lower median OS compared with 405 patients treated before the restricted EMA use (12.4 vs. 23.7 months, p < 0.001) and 2) their prognostic scoring system (Four parameters neutrophil-to-lymphocyte ratio (< vs. ≥ 3.1), ECOG performance status (0-1 vs. 2-3), number of bone metastases (< vs. ≥ 220) and ALP (< vs. ≥ 220) identified three distinctive prognostic groups: the low-risk group (score 0-2), the intermediate-risk group (score 3-4), and the high-risk group (score 5-10)) was a reliable prognostic tool not only before the EMA use but also after the EMA restriction. Although we have first established a nomogram for determining the suitability of mCRPC patients who met EMA guidance to receive maximum benefit from Ra-223 in the present study, in the future, a validated and simple model for clinicians to assess which patients will most likely benefit from Ra-223 treatment needs to be discussed in detail.
As compared to other investigations, the present analysis was conducted with a larger multicentric population and simultaneously modeled seven clinical and imaging variables using regression models with a LASSO penalty for variable selection. Unlike other statistical modeling methods, the LASSO Cox regression model uses shrinkage as a property, which results in a more stable selection of variables. Furthermore, the identification of predictive factors is not dependent on statistical significance. This study has some limitations, including its retrospective nature and performance at multiple centers, making generalization of findings limited and statistical errors possible. A prospective multicenter trial with a larger cohort would help to clarify the usefulness of the present nomogram in clinical settings for selection of mCRPC patients to receive Ra-223 treatment. Additionally, in patients with Ra-223 treatment failure, the following therapy was not uniform and various treatments were performed by the treating physicians. Finally, no external validation of the nomogram was performed.

Conclusion
The novel nomogram developed in this retrospective multicenter study was found to be a valid and reliable tool for the selection of mCRPC patients at the baseline who can receive maximum beneficial effects of Ra-223 therapy, regardless of EMA recommendation. Its use for the multidimensional assessment of patients at the time of enrollment for treatment with Ra-223 provides the treating physician with therapeutic decision-making parameters that can be used in combination, such as the sum score of prior treatment history based on ARTA therapy, taxane-based chemotherapy, and bisphosphonate/denosumab treatment, ECOG performance status, PSADT, hemoglobin, LDH, ALP, and aBSI.
Data availability Data of this study are available upon a reasonable request from the corresponding author, Kazuhiro Kitajima. They are not publicly available due to privacy restrictions.
Code availability Not applicable.

Declarations
Ethics approval We clearly state that human participants have the approval of an appropriately named ethics committee including the Helsinki declaration.

Consent to participate
The Ethics Committee of Hyogo College of Medicine (number 3662) and other institutions approved the present study and written informed consent was obtained from all patients.

Consent for publication
The authors confirm that this manuscript has not been published or presented elsewhere, in part or in entirety, and is not under consideration by any other journal. All authors have made substantial contributions to this work and have read and approved the final submitted version.

Conflict of interest
Author Kitajima has received research grants from PDRadiopharma, Inc. (Tokyo, Japan). The sponsor of the study had no role in the design or conduct of the study, data collection or interpretation, or preparation of the report. There are no other potential conflicts of interest relevant to this study to be reported.