Validation of nine prognostic models in patients with bone metastases undergoing surgery.

Background

The number of patients living with metastatic cancer, as well as survival in the metastatic setting, is increasing.[ 1 , 2 ] Following lungs and liver, bone is the third most frequent location of secondary metastases. The 10-year incidence of bone metastases in cancer patients is estimated at 8%, with prostate and lung cancer patients presenting with substantially higher 10-year incidences of 29% and 13%, respectively.[ 3 , 4 ] Thus, adequate prognostication to choose the most suitable treatment strategy in case of symptomatic bone metastases is becoming more and more important. Patients with painful metastatic bone disease can often be managed by analgesics, bisphosphonates, denosumab and radiotherapy (RTX).[ 5 , 6 ] However, those with very painful, impending or already fractured metastases frequently require additional interventions by means of surgical stabilisation.[ 5 ] Considering that any in-hospital treatment should be kept at a minimum in the palliative setting, and improvement in quality of life should be strived for, the extent of (invasive) treatment has to be chosen wisely.[ 7 – 10 ].

Available prognostic models predicting all-cause mortality in patients with symptomatic bone metastases to the extremities and spine all aspire to provide a guidance for physicians as how invasive treatments should be, given the expected individual prognosis.[ 7 , 11 – 21 ] Some models have already been validated on external cohorts[ 22 – 25 ] to assess their performance in clinical practice, usually on cohorts with similar characteristics in terms of bone metastasis location and performed treatment. Regardless of their usage in spine or extremity metastasis patients, most prediction models are based on identical variables, including primary histological subtype, presence of impending or pathological fracture, number of bone metastases, and presence of organ metastases.[ 7 , 11 – 21 ] However, it remains to be determined how well models perform in a selected cohort of patients undergoing surgery for their bone metastases. Further, it is unclear whether prediction models developed in patients with metastases to the extremity perform equally well when tested in patients with spine metastases – and the other way round.

We therefore asked the following questions: 1) How is the prognostic performance of established prediction models for mortality in patients with spine and/or extremity bone metastases undergoing surgery? 2) How well is the performance of models in clinically relevant subgroups? 3) Which patient characteristics are of importance for mortality prediction?

Methods

The present retrospective single-centre study included all consecutive patients undergoing surgical treatment for spine or extremity bone metastases at a single university hospital between 29th May, 2000 and 21 st April, 2022. Overall, 526 patients were retrospectively included. Ten of these patients underwent surgery for both spine and extremity metastases during follow-up. They were only included once in the dataset, depending on which metastasis location had been treated first (6 patients with spine metastasis undergoing surgery first; Fig. 1 ).

We chose 9 distinct prognostic models developed in patients with spine metastases [ 14 , 17 , 19 , 20 ], extremity metastases [ 7 , 11 , 15 , 18 ], or both.[ 16 ] Models were selected based on the possibility to (theoretically) apply them both in patients with extremity and spine metastases, and on availability of variables in our dataset.

Variables included in the prediction models herein investigated were collected using medical records, fever charts, histopathological and surgical reports, and adhered to a pre-specified data ascertainment protocol. Parameters included patient age, sex (male/female), type of treatment, primary tumour histology, presence of multiple bone metastases (no/yes), presence of visceral metastases (no/yes), presence of brain metastases (no/yes), Karnofski index at time of surgery (0–100), haemoglobin levels 7 to 1 days prior to surgery (mmol/L or g/dL), tumour location, presence of pathological fracture (no/impending/yes – definition no vs. impending derived from medical records as documented by physician in charge), time from primary tumour diagnosis to surgery (in days), disease-free interval (no/yes), American Society of Anaesthesiologists score (ASA; 0 to 4), Eastern Cooperative Oncology Group performance status (ECOG; 0 to 4), body mass index (BMI), presence of neurological deficit (no/yes), history of previous chemotherapy (CTX; no/yes), surgical strategy (intralesional/marginal/wide), remaining primary disease (no/yes), and the updated version of the Charlson Comorbidity Index (Suppl. table 1 ).[ 26 ] Patient outcome (i.e. death and time to death) was assessed as documented in the medical documentation system or official obituary columns. Tumour profile based on primary tumour histology adhered to definitions proposed by the authors of the respective prediction models. Yet, some histologies had not been defined in detail in the original studies, wherefore they were included in tumour profile categories based on classifications used in other scores and medical reasoning (Suppl. table 2 ).

After surgery, patients were followed up on a regular basis, with clinical appointments usually scheduled at 2 and 6 weeks, as well as 1 year postoperatively. In parallel, patients had regular appointments with oncologists (in case of active treatment) and/or outpatient palliative care teams.

The manuscript was prepared according to the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) guidelines.[ 27 ] The study was approved by the institutional review board (IRB-number 36–030 ex 23/24) and carried out according to the Declaration of Helsinki.

Statistical analysis

Means and medians are provided with corresponding standard deviations and interquartile ranges (IQRs), respectively for normally and non-normally distributed variables. Categorical data are reported as counts and valid percentages. Variable distribution was assessed using the Shapiro Wilk test. Chi-squared tests, Fisher’s exact tests and Wilcoxon rank-sum tests were used to assess differences between groups, as appropriate. Given that for most models[ 7 , 14 , 17 – 20 ], no equation or nomogram of the involved prognostic scores were published, we used each predictor variable in its specified form to fit respective Cox proportional hazards regression and logistic regression models ad-hoc. Performance of prediction models was assessed in the entire cohort (n = 526), as well as separately in the spine (n = 224) and extremity subgroups (n = 302; Suppl. tables 3 – 5 ). The prediction model proposed by Bollen et al .[ 14 ] was analysed two times, once adhering to the study population the model was developed on, i.e. excluding patients with tumours from haematological origin (multiple myeloma, lymphoma, myeloic sarcoma, histiocytic sarcoma), and once on the entire patient cohort (with the tumour profile considering malignancies of haematological origin as “moderate”). Similarly, the model by van der Linden et al .[ 20 ] was once studied in the entire cohort, and once by excluding patients with melanoma or kidney cancer (as in the original publication). The models by Ratasvouri et al .[ 18 ], Bollen et al . [ 14 ] , Janssen et al . [ 15 ] , Katagiri et al . [ 16 ] , Leithner et al . [ 17 ] , Willeumier et al . [ 7 ] , Tomita et al . [ 19 ] , and van der Linden et al.[ 20 ] are based on Cox-regression models with death as primary endpoint. The model published by Sorensen et al .[ 11 ] is based on logistic regression with 3-, 6- and 12-month mortality as outcomes. Therefore, model performance was assessed using Harrell’s c-statistic (between 0 and 1, higher values indicate better prediction) for Cox-regression models, area under the receiver operating characteristic curve (AUC ROC; values between 0 and 1, higher values indicate better accuracy) for logistic regression models, as well as the Brier score, representing the average prediction error (values between 0 and 1, lower values indicative of better prediction).[ 28 ] Differences in c statistics between individual models deriving from the entire cohort (excluding models by Ratasvouri et al . due to low case numbers and the model by Sorensen et al . since being based on logistic regression analysis) were analysed using R command compare (Suppl. table 6 ). Calibration of models in all datasets was assessed with calibration plots for observed vs. predicted survival at 6 months (Suppl. Figures 1 – 3 ).

To study the importance of all patient characteristics included in any of the prognostic models for the prediction of mortality, variable importance plots were constructed based on the partial chi-square minus degrees of freedom of the respective variable in a Cox proportional hazards regression model containing all patient characteristics as predictors, as reported previously (Fig. 2 ).[ 29 ] These included all candidate variables used in either of the prediction models (sex, age, BMI, ASA, ECOG performance status, Karnofski index, Charlson Comorbidity Index, clinical profile [risk group as defined by the most recently published scoring model[ 7 ]], haemoglobin levels [g/dL], presence of visceral metastases, presence of brain metastases, presence of multiple bone metastases, prior CTX, remaining primary disease, presence of impending/pathological fracture, location of metastases [omitted for spine and extremity subgroups], neurological deficit, surgical strategy, disease-free interval, and time from primary tumour diagnosis to surgery > 6 months). Separate models were fitted for the entire cohort, the spine subgroup, and the extremity subgroup (Fig. 2 ; Suppl. Figure 4 ). We carried out complete case analyses (Suppl. table 7 ). A sensitivity analysis only including patients with intralesional resection was performed (Suppl. Figure 4 ). A p-value of < 0.05 was considered statistically significant. Statistical analyses were carried out with Stata Version 16.1 for Mac (StataCorp, College Station, TX, US) and R version 4.3.

Results

Between 29th May, 2000 and 21 st April, 2022, a total of 526 patients were enrolled. In the entire cohort, 271 patients were male (52%), and median age was 67.0 [IQR: 58.1–74.0] years at time of surgery (Table 1 ). The most common histologies were hormone dependent breast cancer ( n = 108; 21%), lung cancer ( n = 93; 18%), kidney cancer ( n = 89; 17%), prostate cancer ( n = 63; 12%), and myeloma ( n = 35; 7%; Table 2 ). Median follow-up after surgery for bone metastases of the entire cohort was 267 [IQR: 86—856] days. During this period, 457 patients died of their disease (87%). Surgical procedures performed for bone metastases are visible in Suppl. Figure 5 . Patients with extremity metastases were older than patients with spine metastases ( p < 0.001), rather female ( p = 0.016), more likely to present with multiple bone ( p = 0.001), visceral ( p = 0.001) and brain metastases ( p < 0.001), had a lower Karnofski score ( p < 0.001), ASA ( p = 0.011) and ECOG ( p < 0.001), lower haemoglobin levels ( p < 0.001), lower Charlson Comorbidity Index ( p = 0.007), presented more often with pathological fractures ( p < 0.001) and no neurological deficit ( p < 0.001), were more likely to undergo intralesional or marginal resection ( P = 0.042), and more often had remaining primary disease following surgery ( p = 0.029; Table 1 ).

Prognostic performance of established prediction models

When assessing the prognostic performance in the entire cohort, the models by Sorensen et al . (AUC at 3, 6, 12 months 0.807, 0.834, 0.821), Janssen et al . (c-index 0.711) and Katagiri et al . (c-index 0.699) achieved highest discriminatory performance (Fig. 3 a). Notably, both Cox-models ( Janssen et al . , Katagiri et al . ) had significantly higher c-indices as compared to all other models (except for the model by Leithner et al . ; Suppl. table 6 ). Corresponding Brier scores pointed into the same direction (Fig. 3 a). Similar observations were made in a sensitivity analysis only including patients with intralesional resection (Suppl. Figure 4 a). There was good general agreement between predicted and observed survival at 6 months (Suppl. Figure 1 ).

Performance of models in clinically relevant subgroups

Similar to the entire cohort, the models by Sorensen et al . (AUC at 3, 6, 12 months 0.826, 0.837, 0.826), Janssen et al . (c-index 0.723) and Katagiri et al . (c-index 0.717) achieved the highest discriminatory performance when tested on the spine subgroup (n = 224; Fig. 3 b). Notably, the models by Sorensen et al . and Janssen et al . has been developed for patients with extremity metastases only, while the one by Katagiri et al . had been created for patients with either extremity or spine metastases (Fig. 3 b). Similar findings were made for the Brier score (Fig. 3 b). Sensitivity analysis only including patients with intralesional resection yielded similar results (Suppl. Figure 4 b). Good general agreement between observed and predicted survival at 6 months was present (Suppl. Figure 2 ).

When studying the models in the extremity subgroup (n = 302), performance of all models tended to be lower compared to the spine metastasis subgroup (Fig. 3 c). Similarly, the models developed by Sorensen et al . (AUC at 3, 6, 12 months 0.782, 0.813, 0.814), Janssen et al . (c-index 0.697) and Katagiri et al . (c-index 0.681) achieved the highest performance, followed by the model proposed by Leithner et al . (c-index 0.675; Fig. 3 c). Consistent results were obtained using the Brier score (Fig. 3 c). Similar results were obtained in a sensitivity analysis only involving patients undergoing intralesional resection (Suppl. Figure 4 c). Further, we observed good general agreement between observed and predicted survival at 6 months (Suppl. Figure 3 ).

Importance of patient characteristics for mortality prediction

In the entire cohort, low haemoglobin levels, unfavourable tumour profile (based on histology) and presence of visceral metastases ranked among the most important predictors (Fig. 2 a). Unfavourable tumour profile, low haemoglobin levels and presence of visceral metastases were the most important predictors in the spine subgroup (Fig. 2 b). Further, unfavourable tumour profile, low haemoglobin levels and presence of an impending or pathological fracture were the most relevant predictors in the extremity subgroup (Fig. 2 c).

Discussion

Our study suggests helpful predictive performance of established prediction models in both patients with spine metastases and patients with extremity metastases. The models with the highest discriminatory performance, regardless of the study cohort, were the ones by Sorensen et al . and Janssen et al . . Discriminatory performance of models was generally better when tested in the spine subgroup in comparison to the extremity subgroup. Consistent across subgroups, the most important predictors for mortality included unfavourable tumour profile (based on histology) and low haemoglobin levels.

The models with the highest discriminatory performance when assessed in the entire cohort were the ones by Sorensen et al . and Janssen et al . . Both models had been developed for patients with extremity metastases undergoing surgery. Also, the model by Katagiri et al . showed good discriminatory performance when tested in the entire cohort, yet c-index was lower. Of note, in all analyses we focused on patients undergoing surgical stabilisation of their bone metastases. Five of the nine validated models had been developed for surgically treated patients[ 11 , 15 , 17 – 19 ], while three had been developed for patients either undergoing RTX or surgery[ 7 , 14 , 16 ], and one for patients with RTX only.[ 20 ] While models developed for patients treated surgically achieved higher discriminatory performance in comparison to models developed for patients undergoing RTX only, also the model for patients with RTX performed well. This eventually mirrors the clinical reality, with close margins between the decision for RTX or surgery.

Further, we studied the question whether models developed for patients with extremity metastases or spine metastases perform equally well when tested vice versa. Of the prediction models analysed in the current study, 4 each had been developed for extremity [ 7 , 11 , 15 , 18 ] or spine metastasis patients only [ 14 , 17 , 19 , 20 ], while one had been established for patients with both extremity and spine metastases.[ 16 ] Intriguingly, models’ discriminatory performance was all but comparable when tested in extremity or spine subgroups, regardless of the original developmental population. Even more, models with good performance in the entire cohort also performed better when tested in the two subgroups. This indicates that specific variables included in the prediction models are of prognostic relevance in both patients with spine metastases and patients with extremity metastases. Although being hypothesis generating only, scores containing variables representing systemic tumour burden (e.g. presence of visceral metastases, presence of brain metastases) seem to show generally higher discriminatory performance.

When looking at the importance of patient characteristics for mortality prediction, low haemoglobin levels and unfavourable tumour profile (based on histology) ranked among the most important predictors, likewise in the entire cohort, and in the spine and extremity metastasis subgroups. These findings are largely in line with literature.[ 22 , 30 , 31 ] Furthermore, presence of an impending/pathological fracture was a top-ranking predictor in the extremity, but not spine metastasis subgroup, again aligning with literature.[ 32 – 34 ].

Some limitations apply to the current study that should be considered. First, the study population in which models were tested constitutes a retrospective one, with patients included over a prolonged time. As treatment strategies, especially those involving systemic therapies, have improved over time, the discriminatory performance of tested models may have been influenced by this modulating factor. In addition, information on some variables could not be ascertained given lack of primary documentation, posing a potential bias to our findings. Second, all patients in this validation cohort had undergone surgical treatment with or without additional RTX. Therefore, conclusions towards models’ performance in patients treated with RTX only, or analgesics, bisphosphonates and denosumab cannot be drawn from the present analysis. With our approach, mixing of patients suffering from impending or pathological fractures (that are indicative of a fast progressing underlying malignancy and thus of worse prognosis [ 35 , 36 ]), with patients who present with small, stable, or slowly progressing bony lesions (that can be managed by means of analgesics, bisphosphonates, denosumab or RTX[ 5 ]) was avoided, though. Third, tumour histology—being a main prognostic factor in patients with bone metastases—was classified differently depending on the models used (i.e. tumour profile) or had not been defined at all in the original publications. Thus, some histological subtypes were allocated to specific tumour profile categories based on classifications described in other scores and medical reasoning (Suppl. table 2 ). Notably, the number of available prediction models is not limited to the nine prediction models herein investigated (e.g. model by Rades et al.[ 37 ], Tokuhashi score[ 13 ], PathFX model[ 38 ]), thus results obtained are not definite. Further, prediction models were either based on Cox-regression or logistic regression analysis. Consequently, the statistical metrics (AUC, c-index) used to assess discriminatory performance varied. In addition, owing to the study’s design, we could not evaluate whether patients’ decision in favour or against surgery had potentially been influenced by the results of the prognostic models.

Conclusions

The present study suggests a particularly helpful predictive performance of the models by Sorensen et al . and Janssen et al . in both patients with spine and extremity bone metastases undergoing surgery. Our results point towards a consistent utility of these prediction models in both extremity and spine bone metastasis patients. Of note, both models incorporate haemoglobin levels and tumour profile, which ranked among the patient characteristics with the highest predictive importance. As we only included patients undergoing surgery for their bone metastases, additional validation studies involving conservatively treated patients are warranted to confirm the generalisability of our findings.

Supplementary Information

Supplementary Material 1.

Acknowledgements

N/A.

Abbreviations

American Society of Anaesthesiologists

Area under the receiver operating characteristic curve

Body Mass Index

Chemotherapy

Eastern Cooperative Oncology Group

Interquartile range

Radiotherapy

Strengthening the Reporting of Observational studies in Epidemiology

Authors’ contributions

MAS: Formal analysis, Investigation, Methodology, Visualization, Writing – Original Draft; PE: Data curation, Investigation, Validation, Writing – Original Draft; FAW: Formal analysis, Validation, Writing – Review & Editing; LL: Data curation, Supervision, Validation, Writing – Review & Editing; PF: Supervision, Validation, Writing – Review & Editing; SS: Supervision, Validation, Writing – Review & Editing; PS: Supervision, Validation, Writing – Review & Editing; AL: Conceptualization, Investigation, Project administration, Validation, Writing – Review & Editing.

Funding

None.

Data availability

The datasets generated and/or analysed during the current study are not publicly available due data protection policies but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the local institutional review board (IRB-number: 36–030 ex 23/24). Given the retrospective design of the study, no written patient consent was obtained.

Consent for publication

N/A.

Competing interests

MAS reports support for attending scientific meetings by implantcast, Alphamed Fischer, ImplanTec and PharmaMar, outside of the submitted work. FAW reports research support by the Foundation for Cardiovascular Research— Zurich Heart House, Kurt und Senta Herrmann Foundation, Theodor und Ida Herzog-Egli Stiftung, 4TEEN4 Pharmaceuticals GmbH, PAM Theragnostics GmbH, and Sphingotec GmbH, financial support by the Research Funding for excellent research by the University of Zurich, Fonds zur Förderung des akademischen Nachwuchses of the University of Zurich, Amgen Switzerland, European Society of Cardiology, Swiss Society of Cardiology, Swiss Heart Foundation, and Medical University of Graz, support for attending scientific meetings by 4TEEN4 Pharmaceuticals GmbH, PAM Theragnostics GmbH, Sphingotec GmbH, European Society of Cardiology, Swiss Society of Cardiology, Foundation for Cardiovascular Research— Zurich Heart House, and Critical Care Clinical Trialists Consortium, support in biomarker measurements by Roche Diagnostics, 4TEEN4 Pharmaceuticals GmbH, PAM Theragnostics GmbH, and Sphingotec GmbH, as well as membership in the Steering Committee of the National Registry of Acute Myocardial Infarction in Switzerland, outside of the submitted work. AL reports institutional educational grants by Alphamed, Medacta, and Johnson&Johnson, outside of the submitted work. The other authors report no conflicts of interest.

Footnotes

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Maria A. Smolle and Philip Elmer these authors share first authorship.

References

1. Siegel RL, Miller KD, Wagle NS, Jemal A. Cancer statistics, 2023. CA Cancer J Clin. 2023;73(1):17–48.

2. Storm HH, Kejs AM, Engholm G, Tryggvadóttir L, Klint A, Bray F, et al. Trends in the overall survival of cancer patients diagnosed 1964–2003 in the Nordic countries followed up to the end of 2006: the importance of case-mix. Acta Oncol. 2010;49(5):713–24.

3. Hernandez RK, Wade SW, Reich A, Pirolli M, Liede A, Lyman GH. Incidence of bone metastases in patients with solid tumors: analysis of oncology electronic medical records in the United States. BMC Cancer. 2018;18(1):44.

4. Migliorini F, Maffulli N, Trivellas A, Eschweiler J, Tingart M, Driessen A. Bone metastases: a comprehensive review of the literature. Mol Biol Rep. 2020;47(8):6337–45.

5. Fallon M, Giusti R, Aielli F, Hoskin P, Rolke R, Sharma M, et al. Management of cancer pain in adult patients: ESMO Clinical Practice Guidelines. Ann Oncol. 2018;29(Suppl 4):iv166–91.

6. Migliorini F, Eschweiler J, Trivellas A, Driessen A, Knobe M, Tingart M, et al. Better pain control with 8-gray single fraction palliative radiotherapy for skeletal metastases: a bayesian network meta-analysis. Clin Exp Metastasis. 2021;38(2):197–208.

7. Willeumier JJ, van der Linden YM, van der Wal C, Jutte PC, van der Velden JM, Smolle MA, et al. An easy-to-use prognostic model for survival estimation for patients with symptomatic long bone metastases. J Bone Joint Surg Am. 2018;100(3):196–204.

8. Bickels J, Dadia S, Lidar Z. Surgical management of metastatic bone disease. J Bone Joint Surg Am. 2009;91(6):1503–16.

9. Frassica DA. General principles of external beam radiation therapy for skeletal metastases. Clin Orthop Relat Res. 2003;415(SUPPL):S158–64.

10. Nathan SS, Healey JH, Mellano D, Hoang B, Lewis I, Morris CD, et al. Survival in patients operated on for pathologic fracture: implications for end-of-life orthopedic care. J Clin Oncol. 2005;23(25):6072–82.

11. Sorensen MS, Gerds TA, Hindso K, Petersen MM. Prediction of survival after surgery due to skeletal metastases in the extremities. Bone Joint J. 2016;98-B(2):271–7.

12. Piccioli A, Spinelli MS, Forsberg JA, Wedin R, Healey JH, Ippolito V, et al. How do we estimate survival? External validation of a tool for survival estimation in patients with metastatic bone disease-decision analysis and comparison of three international patient populations. BMC Cancer. 2015;15:424.

13. Tokuhashi Y, Matsuzaki H, Oda H, Oshima M, Ryu J. A revised scoring system for preoperative evaluation of metastatic spine tumor prognosis. Spine (Phila Pa 1976). 2005;30(19):2186–91.

14. Bollen L, van der Linden YM, Pondaag W, Fiocco M, Pattynama BP, Marijnen CA, et al. Prognostic factors associated with survival in patients with symptomatic spinal bone metastases: a retrospective cohort study of 1,043 patients. Neuro Oncol. 2014;16(7):991–8.

15. Janssen SJ, van der Heijden AS, van Dijke M, Ready JE, Raskin KA, Ferrone ML, et al. 2015 Marshall Urist Young Investigator Award: prognostication in patients with long bone metastases: does a boosting algorithm improve survival estimates? Clin Orthop Relat Res. 2015;473(10):3112–21.

16. Katagiri H, Takahashi M, Wakai K, Sugiura H, Kataoka T, Nakanishi K. Prognostic factors and a scoring system for patients with skeletal metastasis. J Bone Joint Surg Br. 2005;87(5):698–703.

17. Leithner A, Radl R, Gruber G, Hochegger M, Leithner K, Welkerling H, et al. Predictive value of seven preoperative prognostic scoring systems for spinal metastases. Eur Spine J. 2008;17(11):1488–95.

18. Ratasvuori M, Wedin R, Hansen BH, Keller J, Trovik C, Zaikova O, et al. Prognostic role of en-bloc resection and late onset of bone metastasis in patients with bone-seeking carcinomas of the kidney, breast, lung, and prostate: SSG study on 672 operated skeletal metastases. J Surg Oncol. 2014;110(4):360–5.

19. Tomita K, Kawahara N, Kobayashi T, Yoshida A, Murakami H, Akamaru T. Surgical strategy for spinal metastases. Spine (Phila Pa 1976). 2001;26(3):298–306.

20. van der Linden YM, Dijkstra SP, Vonk EJ, Marijnen CA, Leer JW. Prediction of survival in patients with metastases in the spinal column: results based on a randomized trial of radiotherapy. Cancer. 2005;103(2):320–8.

21. Salim X, D’Alessandro P, Little J, Mudhar K, Murray K, Carey Smith R, et al. A novel scoring system to guide prognosis in patients with pathological fractures. J Orthop Surg Res. 2018;13(1): 228.

22. Bollen L, Wibmer C, Van der Linden YM, Pondaag W, Fiocco M, Peul WC, et al. Predictive Value of Six Prognostic Scoring Systems for Spinal Bone Metastases: An Analysis Based on 1379 Patients. Spine (Phila Pa 1976). 2016;41(3):E155-62.

23. Sorensen MS, Gerds TA, Hindso K, Petersen MM. External validation and optimization of the SPRING model for prediction of survival after surgical treatment of bone metastases of the extremities. Clin Orthop Relat Res. 2018;476(8):1591–9.

24. Smolle MA, Musser E, Bergovec M, Friesenbichler J, Wibmer CL, Leitner L, et al. Survival prediction in patients treated surgically for metastases of the appendicular skeleton-an external validation of 2013-SPRING model. Cancers (Basel). 2022. 10.3390/cancers14143521.

25. Ben Gal O, Soh TCF, Vaughan S, Jayasanker V, Mahendra A, Gupta S. The prediction of survival after surgical management of bone metastases of the extremities-a comparison of prognostic models. Curr Oncol. 2022;29(7):4703–16.

26. Quan H, Li B, Couris CM, Fushimi K, Graham P, Hider P, et al. Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. Am J Epidemiol. 2011;173(6):676–82.

27. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies. J Clin Epidemiol. 2008;61(4):344–9.

28. Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology. 2010;21(1):128–38.

29. Hijazi Z, Oldgren J, Lindbäck J, Alexander JH, Connolly SJ, Eikelboom JW, et al. The novel biomarker-based ABC (age, biomarkers, clinical history)-bleeding risk score for patients with atrial fibrillation: a derivation and validation study. Lancet. 2016;387(10035):2302–11.

30. Ahmed AK, Goodwin CR, Heravi A, Kim R, Abu-Bonsrah N, Sankey E, et al. Predicting survival for metastatic spine disease: a comparison of nine scoring systems. Spine J. 2018;18(10):1804–14.

31. Meares C, Badran A, Dewar D. Prediction of survival after surgical management of femoral metastatic bone disease - a comparison of prognostic models. J Bone Oncol. 2019;15: 100225.

32. Kirkinis MN, Lyne CJ, Wilson MD, Choong PF. Metastatic bone disease: a review of survival, prognostic factors and outcomes following surgical treatment of the appendicular skeleton. Eur J Surg Oncol. 2016;42(12):1787–97.

33. Hansen BH, Keller J, Laitinen M, Berg P, Skjeldal S, Trovik C, et al. The Scandinavian Sarcoma Group Skeletal Metastasis Register. Survival after surgery for bone metastases in the pelvis and extremities. Acta Orthop Scand Suppl. 2004;75(311):11–5.

34. Weiss RJ, Tullberg E, Forsberg JA, Bauer HC, Wedin R. Skeletal metastases in 301 breast cancer patients: patient survival and complications after surgery. Breast. 2014;23(3):286–90.

35. Ulas A, Bilici A, Durnali A, Tokluoglu S, Akinci S, Silay K, et al. Risk factors for skeletal-related events (SREs) and factors affecting SRE-free survival for nonsmall cell lung cancer patients with bone metastases. Tumour Biol. 2016;37(1):1131–40.

36. Sathiakumar N, Delzell E, Morrisey MA, Falkson C, Yong M, Chia V, et al. Mortality following bone metastasis and skeletal-related events among men with prostate cancer: a population-based analysis of US Medicare beneficiaries, 1999–2006. Prostate Cancer Prostatic Dis. 2011;14(2):177–83.

37. Rades D, Fehlauer F, Schulte R, Veninga T, Stalpers LJ, Basic H, et al. Prognostic factors for local control and survival after radiotherapy of metastatic spinal cord compression. J Clin Oncol. 2006;24(21):3388–93.

38. Forsberg JA, Eberhardt J, Boland PJ, Wedin R, Healey JH. Estimating survival in patients with operable skeletal metastases: an application of a bayesian belief network. PLoS One. 2011;6(5): e19956.

Associated Data

Supplementary Materials

Supplementary Material 1.

Data Availability Statement

The datasets generated and/or analysed during the current study are not publicly available due data protection policies but are available from the corresponding author on reasonable request.