If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Address for correspondence: Nicholas G. Zaorsky, MD, Department of Medical Oncology, University Hospitals Cleveland Medical Center, 11100 Euclid Ave, Cleveland, OH 44106, U.S.A
Affiliations
Department of Medical Oncology, University Hospitals Cleveland Medical Center, Cleveland, OH
Castration-sensitive metastatic prostate cancer is heterogeneous. Our objective is to identify metastatic prostate cancer phenotypes and their prognostic impact on survival.
Materials and Methods
The National Cancer Database was queried. The Surveillance, Epidemiology, and End Results database was used for validation. Patterns were split into: nonregional lymph node, bone only, and visceral (any brain/liver/lung). Hazard ratios (HR) with 95% confidence intervals (CI) were calculated for the univariate and multivariate Cox proportional hazards regression models, odds ratios were calculated, Kaplan-Meier curves were generated, and a nomogram of the multivariate regression model was created.
Results
The training set included 13,818 men; bone only was most common (n = 11,632, 84.2%), then nonregional lymph node (n = 1388, 10.0%), and any visceral (brain/liver/lung; n = 798, 5.8%). Risk of death was increased by metastases to a visceral organ versus nonregional lymph node (HR = 2.26; 95% CI [2.00, 2.56]), bone only metastases versus nonregional lymph node (HR = 1.57; 95% CI [1.43, 1.72]), T-stage 4 versus 1 (HR = 1.27; 95% CI [1.17, 1.36]), Grade Group 5 versus 1 (HR = 1.93; 95% CI [1.61, 2.31]), PSA > 20 ng/mL versus < 10 ng/mL (HR = 1.32; 95% CI [1.23, 1.42]), and age ≥ 80 versus < 50 (HR = 1.96; 95% CI [1.69, 2.29]). On internal validation, the model had C-indices 20.5%, 22.7%, and 14.6% higher than the current staging system for overall survival, 1-year, and 5-year survival, respectively.
Conclusion
We developed and validated prognostic metastatic prostate cancer phenotypes that can assist risk stratification to potentially personalize therapy. Our nomogram (https://tinyurl.com/prostate-met) may be used to predict survival.
Prostate cancer is the most common form of cancer, affecting approximately 1 in every 9 men, and is the second leading cause of cancer death in American men.
We hypothesize that these unique phenotypical variants of castration-sensitive metastatic prostate cancer can inform staging, treatment, and survival outcomes. Current treatment recommendations for prostate cancer are similar for all phenotypes of metastases, taking a “middle of the road” approach, recommending systemic therapy for everyone, despite disease course and survival outcomes varying widely between these sub-groups. Further investigation of these sub-types may be clinically used to predict outcomes in patients and identify patients in greatest need of treatment intensification. Biologically, the results may be used to identify the underlying pathways responsible for metastasis to particular subsites of the body; treatments may be developed to target these pathways in the clinic.
Our retrospective analysis focuses on improving prognostication, which will help clinicians identify the patients most appropriate for different therapies, including: local radiation therapy, systemic therapies (endocrine, cytotoxic, targeted, immunotherapeutic), and palliative care.
Safety and survival rates associated with ablative stereotactic radiotherapy for patients with oligometastatic cancer: a systematic review and meta-analysis.
Specifically, we sought to define distinct metastatic prostate cancer phenotypes and their associated survival times, develop and validate a nomogram predictive of survival, and compare our results to other published studies, to provide a deeper understanding of this topic.
Methods
Data Acquisition
In this study, we extracted information from 2 databases. The National Cancer Database (NCDB) was our training dataset, and then information from the Surveillance, Epidemiology, and End Results (SEER) database was used to validate our findings (Supplemental Figure 1). The NCDB is a hospital-based cancer registry that collects data from American College of Surgeons–Commission on Cancer (CoC) accredited facilities. The database includes 70% of all malignant cancers diagnosed in the United States. The NCDB has patient level data on the type of cancer, histology (eg, squamous cell, adenocarcinoma), patient demographics (eg, sex, age, race), follow up time, location of metastasis (eg, bone, liver, lung, brain, other), and vital status. Additionally, the NCDB contains information regarding therapies delivered during the first course of treatment (including surgery, radiation therapy, chemotherapy, hormone therapy, immunotherapy, and/or palliative care). Overall, the data from the NCDB have been shown to be very accurate due to quality assurance, data checks, audits as indicated by Boffa et al.
Using the NCDB dataset (2010-2014), descriptive statistics, such as means and medians, of the patient population were determined. Next, hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated for the univariate and multivariate Cox proportional hazards regression model. The NCDB was internally validated using the Uno's C-statistic (C-index) at various time points (ie, 1-month, 6-month, 12-month, 36-month and 60-month) and for overall survival. Also, the generalized estimating equation (GEE) model was used to calculate odds ratios of site-specific metastasis. For the Cox proportional hazards model, we further removed patients from the study population used in the GEE model if they had missing values in time to death or censoring indicator. Given similar predictive accuracy in the Cox proportional hazards model with the 4 binary metastatic variables and the model with merged metastatic patterns, merged metastatic patterns were used out of clinical consideration (3 groups: nonregional lymph node, bone only, and any visceral [brain/liver/lung]). Kaplan-Meier curves were utilized to present nonparametric survival probability estimates with 95% CI for patients with different merged metastatic patterns, where log-rank tests were used for group comparison. A nomogram of the multivariate Cox proportional hazards model was created and used as survival prediction. Calibration plots were made for 3-year and 5-year survival probabilities to measure concordance of the model predictions with the actual survival outcomes (Supplemental Figure 2).
External Validation
In order to externally validate this dataset, patients with castration-sensitive metastatic prostate adenocarcinoma, diagnosed between 2010 and 2015, were extracted from the National Cancer Institute's SEER program. More specifically, to compare the distribution of each covariate in the 2 datasets, Chi-Square tests were performed for categorical covariates and t tests were performed for continuous covariates. The overview and limitations of the database and the methods are described in Supplemental File 1. Again, Kaplan Meier curves were generated (Supplemental Figure 3). SEER is a network of population-based incident tumor registries from geographically distinct regions in the US, covering 28% of the US population, including incidence, survival, and treatment (eg, radiation therapy, surgery, chemotherapy). For the current analysis, the SEER18 registry was used. The SEER registry includes data on age at diagnosis, race, marital status, and year of diagnosis. SEER*Stat 8.2.1 was used for analysis. Additionally, the Uno's C-statistic was calculated for the SEER validating dataset, therefore externally validating this dataset, as well.
Results
There were 13,818 men diagnosed with metastatic castration-sensitive prostate cancer from 2010 to 2014 in the NCDB. Supplemental Table 1 shows patient characteristics. The most common metastatic pattern was bone only (n = 11,632, 84.2%), followed by nonregional lymph node (n = 1388, 10.0%), and then any visceral involvement (brain/liver/lung; n = 798, 5.8%). The mean age of all men diagnosed with metastatic disease was 69 years old, and was similar among the different cohorts. 10,736 men were white (77.7%), 2590 were black (18.7%), and 492 were another race (3.6%). The majority of patients (n = 9963, 72.1%) had PSA > 20 ng/mL. The presence of lower PSA (< 10 ng/mL) was most common among patients with nonregional lymph node disease (n = 354, 25.5%) and any visceral disease (n = 160, 20.1%), when compared to bone only disease (n = 1588, 13.7%), P < .001. Among all patients with metastases, presence of Grade Group (GG) 4 (n = 3284, 23.8%) and GG 5 (n = 8175, 59.2%) disease was most common compared to other groups (Figure 1, Supplemental Table 1). Distribution of patients among metastatic sites are shown in Table 1. The most common pattern with any visceral involvement was of both the bone and lung (n = 298, Table 1).
Figure 1Metastatic phenotypes by site of metastasis. The left graph shows the distribution of metastatic with percent of patients on the y-axis and metastatic site(s) on the x-axis. The most common metastatic site was bone, followed by nonregional lymph node involvement. The right graph shows a heat map of metastatic site involvement (y-axis) by Grade Group (x-axis). Grade Groups 4 and 5 are most prevalent in metastatic disease, regardless of site.
Mean(SD) was reported for continuous variable, n(%) was reported for categorical variable. TX is the T stage cannot be or has not been assessed. NX is the cancer in the lymph nodes cannot be or has not been assessed. This is often because T and N stage become clinically irrelevant, and therefore not worked-up, if the patient is presenting with metastatic disease. P value: To test the association between merged metastatic pattern and covariates, Chi-square test was performed for categorical covariates, and Kruskal-Wallis H test (nonparametric one-way ANOVA) was performed for continuous covariates. Age and PSA were regarded both as continuous and categorical variable, and smaller P value was reported for each variable.
Table 2 shows hazard ratios between metastatic pattern, T stage, N stage, race, GG, PSA, and age group. The following significantly increase the risk of death: metastases to any visceral organs versus nonregional lymph nodes (HR = 2.26 [95% CI 2.00, 2.56]), bone only versus nonregional lymph node metastases (HR = 1.57 [95% CI 1.43, 1.72]), T stage 4 versus T stage 1 (HR = 1.27 [95% CI 1.17, 1.36]), GG 5 versus GG 1 (HR = 1.93 [95% CI 1.61, 2.31]), PSA > 20 ng/mL versus < 10 ng/mL (HR = 1.32 [95% CI 1.23, 1.42]), and age ≥ 80 versus < 50 (HR = 1.96 [95% CI 1.69, 2.29]).
Table 2Hazard Ratios (95% CI) From Univariate/Multivariate Cox Proportional Hazards Model
Description
Univariate (Unadjusted)
Multivariate (Adjusted)
HR (95% CI)
P value
HR (95% CI)
P value
Subgroup of Metastatic Pattern
Visceral (any brain/liver/lung) versus Nonregional lymph node
TX is the T stage cannot be or has not been assessed. NX is the cancer in the lymph nodes cannot be or has not been assessed. This is often because T and N stage become clinically irrelevant, and therefore not worked-up, if the patient is presenting with metastatic disease.
Figure 2 shows the odds ratio within each metastatic site for 10 ng/mL increase in PSA, 10-year increase in age, race: black versus white, and race: other versus white (Panels A and B) and for different GG (Panels C and D). Higher GG is correlated with an increase in odds of bone metastases: GG 3 versus 1 (OR = 2.016 [95% CI 1.507, 2.696]), GG 4 versus 1 (OR = 2.290 [95% CI 1.770, 2.963]), and GG 5 versus 1 (OR = 2.510 [95% CI 1.966, 3.204]). However, the same was not true for higher GG and odds of metastasis to any visceral organ. Also, Figure 2 Panel E shows there is no association between age at diagnosis and site(s) of metastases.
Figure 2Metastatic phenotypes by Grade Groups, race, PSA, age. Panels A to D show odds ratios within each metastatic site (bone, liver, lung, brain) for 10 ng/mL increase in PSA, 10-year increase in age, race: black versus white, and race: other versus white (Panels A and B) and for different Grade Groups (Panels C and D). In Panel A and C, the x-axis depicts the metastatic site, while the y-axis shows the odds ratio. The dotted horizontal line is a ratio of 1, or no association. Significant findings include that the likelihood of bone metastases increases with Grade Group 3 versus 1, 4 versus 1, and 5 versus 1 (Panel C). Panels B and D show heat maps for the same information, respectively. Metastatic site is on the y-axis and the specific variable is on the x-axis. A darker blue or red, respectively, indicates a higher odds ratio, while a lighter white or green color, respectively, indicates a lower odds ratio. Panel E shows the distribution of age at diagnosis for specific metastatic patterns. The y-axis is age and the x-axis depicts the various metastatic phenotypes, showing no correlation between age at diagnosis and site(s) of metastases. Panel F depicts the distribution of race for specific metastatic patterns. Most patterns are dispersed amongst races.
The nonparametric Kaplan-Meier curves depicted a lowest survival probability for patients with any visceral [brain/liver/lung] metastatic pattern, followed by bone only, and nonregional lymph node with a log-rank test P value < .0001 (Figure 3). A novel nomogram was created to predict 3- and 5- year survival probability for men diagnosed with metastatic disease. Parameters include metastatic pattern, T stage, N stage, race, PSA, GG, and age. Any visceral site of metastasis, GG 5 disease, and age ≥ 80 portended the worst prognosis (https://tinyurl.com/prostate-met). Compared with the current staging system, which only include the merged metastatic pattern as a predictor, the proposed Cox proportional hazards model had a 20.5%, 22.7%, and 14.6% higher C-index for overall survival, 1-year, and 5-year survival, respectively. The C-index of 1-year survival is 0.68 (95% CI [0.67, 0.69]) for internal validation and 0.66 (95% CI [0.64, 0.67]) for external validation, and 0.63 (95% CI [0.59, 0.66]) for 5-year survival internal validation and 0.63 (95% CI [0.59, 0.67]) for external validation. All suggest good reliability of the nomogram (Figure 3).
Figure 3Nomogram to predict survival, Kaplan-Meier survival curves, and C-index of Cox proportional hazards models over time for prediction of all-cause survival Figure 3. includes a nomogram generated to predict 3- and 5- year survival probability. Parameters include metastatic pattern, T stage, N stage, Race, PSA, Grade Group, and age. The right panel shows Kaplan-Meier curves with survival probability on the y-axis and the time to death in months on the x-axis. Nonregional lymph node involvement has a higher survival probability than bone only metastases, which has a higher survival probability than any visceral metastases. The bar graph shows Uno's C-statistic (C-index) of Cox proportional hazards models over Time for Prediction of Overall Survival in the Internal Validation Cohort NCDB, External Validation Cohort SEER versus Current Staging System with NCDB. A higher C-index indicates a better measure of discrimination ability. Error bars denote bootstrapped 95% CIs. The first column “Overall” of x-axis shows the summary measure of time-dependent area under the curve. The current staging system only includes the merged metastatic patterns (nonregional lymph node, bone only, and visceral (any brain/liver/lung)) in the Cox proportional hazards model.
Additionally, Table 3 contains summarized information from a literature analysis assessing the significance of metastatic phenotypes and different prognostic factors.
Recurrent metastatic disease, after prostate bed radiotherapy without castration
Institutional
Longer PSA doubling time, involvement of nodes or axial skeleton, and a lower number of metastases are associated with improved prostate cancer specific survival
Prostate cancer specific survival No validation No nomogram Retrospective Institutional population
Metastatic spread and pain patterns confer different prognosis in patients with metastases. Bone may serve as a crucial microenvironment in the development of CRPC and disease progression
Cancer-specific mortality No nomogram No validation Retrospective Institutional population
In men with castration-resistant prostate cancer, visceral metastases predict shorter overall survival: what predicts visceral metastases? Results from the SEARCH database.
Visceral metastases predicted worse survival. Age, year, treatment center, PSA, and time from CRPC to metastases were significant in predicting survival.
Overall survival No validation No nomogram Retrospective
Prognostic significance of young age and non-bone metastasis at diagnosis in patients with metastatic prostate cancer: a SEER population-based data analysis.
Sites of synchronous distant metastases and prognosis in prostate cancer patients with bone metastases at initial diagnosis: a population-based study of 16,643 patients.
Prognostic factors for survival: lung metastases- older age, advanced T stage, and higher Gleason score; liver metastases- older age and higher Gleason score; brain metastases- advanced T stage
Overall survival No nomogram No validation Retrospective
Current analysis
13,818 (NCDB), 9318 (SEER)
De novo metastatic disease
NCDB (training) SEER (validating)
Nomogram Nonregional lymph node, bone only, and any visceral metastases (brain/liver/lung) have distinct survival patterns
Permutations of metastases (Table 1) All ages Nomogram Validation PSA data Retrospective
Our study evaluates the impact of various clinicopathologic factors on outcomes of metastatic prostate cancer patients. High GG is associated with the presence of metastases to any site. Also, a lower PSA (< 10 ng/mL) was identified in 15% patients. It is most common in patients with nonregional lymph node disease and any visceral disease when compared to bone only disease. Although this lower PSA group is uncommon, this is the proportion that are clinically seen. This group of patients has a high rate of prostate cancer specific mortality, as explored by Mahal et al.
A significant portion of the cancers were T stage 1 or 2 and N stage 0, matching the picture that is clinically seen where disease appears confined to the prostate and then work-up reveals otherwise, or a patient presents with metastases and then work-up reveals the primary tumor is in the prostate. A patient's risk of death is increased by metastases to any visceral organs versus nonregional lymph node, bone only versus nonregional lymph node metastases, T stage 4 versus 1, GG 5 versus 1, PSA > 20 ng/mL versus < 10 ng/mL, and age ≥ 80 versus < 50. Additionally, we created a novel nomogram that clinicians can use to generate survival estimates that takes into account metastatic pattern, T stage, N stage, Race, PSA, Grade Groups, and age. Lastly, we analyze current, relevant literature on this topic (Table 3); we suggest future staging systems consider number of metastases, performance status, PSA kinetics, hemoglobin, pain, alkaline phosphatase, albumin, and lactate dehydrogenase.
Prognostic significance of young age and non-bone metastasis at diagnosis in patients with metastatic prostate cancer: a SEER population-based data analysis.
In men with castration-resistant prostate cancer, visceral metastases predict shorter overall survival: what predicts visceral metastases? Results from the SEARCH database.
Sites of synchronous distant metastases and prognosis in prostate cancer patients with bone metastases at initial diagnosis: a population-based study of 16,643 patients.
Supplemental Table 2 shows that all studies were institutional or used large retrospective databases. We identify other factors that may be considered in prognostication. In general, the presence of visceral metastases, advanced age, higher Gleason score were all associated with worse survival. Several studies created and validated nomograms, which included the following additional covariates: performance status, alkaline phosphatase, albumin, lactate dehydrogenase, hemoglobin, PSA doubling time, pain on exam, number of metastases.
Our study is unique because we use NCDB as our training dataset and SEER as external validation, we have a large number of patients without age limitations, and we created a novel nomogram that clinicians can use as a prognostic calculator. Future work that applies this nomogram to previously completed or ongoing clinical trials to assess metastases in the context of specific systemic therapies is ongoing. Additionally, by being the only study to use NCDB, our work further supports and compliments these prior, similar studies that were completed. NCDB pulls information from 34 million patients in the system. It contains 70% of newly diagnosed cancers from the 1500 CoC accredited facilities with diagnoses since 2004 for patients of all ages. Overall, the data from the NCDB have been shown to be very accurate due to quality assurance, data checks, audits as indicated by Boffa et al.
The SEER-Medicare database is more limited, only having information on 12% of newly diagnosed cancers and 1.8 million patients in the database dating back to 1991. Also, the population is older (≥ 65 years old) because of the affiliation with Medicare. Other studies in Table 3 used institutional databases, SEER, SEARCH, or clinical trials that also have inherent limitations.
studies, in addition to their retrospective nature. The majority used only a single dataset and therefore only a minority of them have a validation data set. Only 3 studies
have nomograms and therefore can be easily implemented clinically. A majority of the studies analyzed a relatively small number of patients (about < 1000).
In men with castration-resistant prostate cancer, visceral metastases predict shorter overall survival: what predicts visceral metastases? Results from the SEARCH database.
Additionally, others have limited age ranges or are from a targeted data set (eg, clinical trials, veterans affairs database [SEARCH]), making their results less widely generalizable.
In men with castration-resistant prostate cancer, visceral metastases predict shorter overall survival: what predicts visceral metastases? Results from the SEARCH database.
It is important to acknowledge that all of these studies are novel works in this field that enable our study to further build on the subject of metastatic disease. A future systematic review of the studies assessing the impact of metastatic locations is underway, and it is beyond the scope of the current work.
There are several limitations to our study. We are limited by the data provided by NCDB. We only have data from initial diagnosis, and we have no information on genomics and transcriptomics (eg, Decipher score), or circulating tumor cell markers, which often help redefine and restratify patients early in their disease. If we had the aforementioned data, then we would have better predictive accuracy. We do not have information on how the patient was diagnosed, imaging, the number of metastases, comorbidities, the volume of disease burden, or recurrence. Future predictive systems should integrate more specific characteristics known to be prognostic, including: number of bone metastases, location, type of bone metastases (blastic vs. lytic), size, associated pain.
Damron TA, Morgan H, Prakash D, Grant W, Aronowitz J, Heiner J. Critical evaluation of Mirels' rating system for impending pathologic fractures. 2003; Heidelberg.
Newer scans (eg, PSMA PET, fluciclovine PET, etc.) are now able to detect metastatic disease more easily, which may lead to stage migration- when disease that would have been staged as localized is now being more correctly classified as metastatic.
As a result, both groups appear to have improved survival times. Additionally, we did not include treatment information, which would impact outcomes, depending on which therapy was used and when, particularly if the sub-type is associated with response or nonresponse to specific systemic therapies. However, this was deliberate. Our retrospective study is designed to improve prognostication, not for comparative effectiveness research of treatments. We also do not want to automatically introduce bias. There is often a bias in who receives treatments and if they do, which treatments they receive, making it difficult to draw clear conclusions from that information.
Furthermore, we may have unintentionally introduced a selection bias by excluding important data points if they were missing values integral to our analysis. NCDB only houses data from CoC accredited hospitals; therefore, these institutions likely have more resources than nonaccredited facilities. Lastly, there can be limitations to applying nomograms clinically as well.
We characterized patterns of metastasis for prostate cancer. We found that the following subgroups have distinct survival patterns: nonregional lymph node, bone only, and any visceral (brain/liver/lung). Our nomogram (https://tinyurl.com/prostate-met) may be used to predict survival.
Clinical Practice Points
•
Prostate cancer is the most common form of cancer, affecting approximately 1 in every 9 men. Prostate cancer cells tend to preferentially metastasize to the lymph nodes and bones (vs. visceral organs like the lungs, brain or liver). Current treatment recommendations for prostate cancer are similar for all phenotypes of metastases, taking a “middle of the road” approach, recommending systemic therapy for everyone, despite disease course and survival outcomes varying widely between these sub-groups.
•
Our retrospective analysis focuses on improving prognostication, which will help clinicians identify the patients most appropriate for different therapies, including: hospice, palliative treatment to metastases, and aggressive treatment with stereotactic body radiation therapy.
•
Our study shows that a patient's risk of death is increased by metastases to any visceral organs versus nonregional lymph node, bone only versus nonregional lymph node metastases, T stage 4 versus 1, GG 5 versus 1, PSA > 20 ng/mL versus < 10 ng/mL, and age ≥ 80 versus < 50. We also created a novel nomogram that clinicians can use to generate survival estimates that takes into account metastatic pattern, T stage, N stage, Race, PSA, Grade Groups, and age (https://tinyurl.com/prostate-met). Lastly, we analyzed current, relevant literature on this topic; we suggest future staging systems consider number of metastases, performance status, PSA kinetics, hemoglobin, pain, alkaline phosphatase, albumin, and lactate dehydrogenase.
Acknowledgments
We thank Carson Wills for help with Graphpad Prism to make supplementary figures.
Disclosure
The authors have stated that they have no conflicts of interest.
CRediT authorship contribution statement
Shelby A. Labe: Validation, Investigation, Resources, Writing – original draft, Writing – review & editing, Visualization, Project administration. Xi Wang: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – review & editing. Eric J. Lehrer: Validation, Writing – review & editing. Amar U. Kishan: Writing – review & editing. Daniel E. Spratt: Writing – review & editing. Christine Lin: Software, Writing – review & editing. Alicia K. Morgans: Writing – review & editing. Lee Ponsky: Writing – review & editing. Jorge A. Garcia: Writing – review & editing. Sara Garrett: Writing – review & editing. Ming Wang: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – review & editing. Nicholas G. Zaorsky: Conceptualization, Validation, Investigation, Resources, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.
Safety and survival rates associated with ablative stereotactic radiotherapy for patients with oligometastatic cancer: a systematic review and meta-analysis.
Prognostic significance of young age and non-bone metastasis at diagnosis in patients with metastatic prostate cancer: a SEER population-based data analysis.
In men with castration-resistant prostate cancer, visceral metastases predict shorter overall survival: what predicts visceral metastases? Results from the SEARCH database.
Sites of synchronous distant metastases and prognosis in prostate cancer patients with bone metastases at initial diagnosis: a population-based study of 16,643 patients.
Damron TA, Morgan H, Prakash D, Grant W, Aronowitz J, Heiner J. Critical evaluation of Mirels' rating system for impending pathologic fractures. 2003; Heidelberg.
00045-3&id=gr1.jpg){kind=link}
00045-3&id=gr2.jpg){kind=link}
00045-3&id=gr3.jpg){kind=link}