Clinical prediction models for malignant bone tumors

Abstract

Malignant bone tumors, consisting of primary and secondary types, present significant clinical challenges due to high morbidity and mortality rates, where prognosis is closely tied to recurrence and metastasis. Current diagnostic processes for bone malignancies face difficulties, including high rates of misdiagnosis and significant delays, particularly for conditions like osteosarcoma and Ewing’s sarcoma. Clinical prediction models (CPMs) serve as essential mathematical tools that integrate multiple risk factors to provide individualized estimates for outcomes such as recurrence, treatment efficacy, or prognosis. These models are typically implemented as scoring scales, nomograms, or web applications to enhance clinical utility. In spinal oncology, CPMs have been developed for rare primary tumors using machine learning and inflammatory markers, as well as for more common spinal metastases. Established systems, including the Tokuhashi, Tomita, and Bauer scores, along with the recent Skeletal Oncology Research Group (SORG) nomogram, are utilized to predict survival and guide treatment strategies. Furthermore, models such as SPRING, OPTIModel, PATHfx, and the Istituto Ortopedico Rizzoli (IOR) score have demonstrated clinical utility in prognosticating extra-spinal bone tumors. Although CPMs show promise as routine auxiliary tools in bone oncology, most current models are limited by retrospective designs and small sample sizes, requiring extensive multicenter validation to ensure accuracy and reliability before widespread clinical implementation.

Graphical abstract

Keywords

clinical prediction models / malignant bone tumors / nomograms / prognosis / spinal tumors

1 Introduction to malignant bone tumors

Malignant bone tumors are categorized into primary bone tumors and secondary bone tumors. Primary bone tumors mainly consist of three types: osteosarcoma, Ewing’s sarcoma, and chondrosarcoma. Among these, osteosarcoma is the most prevalent, accounting for nearly two-thirds of all cases.[1] Osteosarcoma predominantly affects adolescents and children, ranking as the third most common childhood malignancy.[2] Ewing’s sarcoma also frequently occurs in adolescents and children; statistics indicate an incidence of 1 in 100,000 among individuals aged 10 to 19 years.[3] The cellular origin of Ewing’s sarcoma remains unclear. It is hypothesized that these tumors arise from undifferentiated primitive neuroectodermal or neural crest cells. Some researchers suggest that Ewing’s sarcoma originates from primitive stem cells, with the degree of malignancy depending on the stage at which the stem cells are arrested during differentiation.[4] Chondrosarcoma, conversely, is a malignant bone tumor originating from cartilage. Unlike osteosarcoma and Ewing’s sarcoma, chondrosarcoma occurs primarily in adults aged 40 to 75 years.[5,6] It is more frequently found in the axial skeleton or proximal bones, typically originating from the pelvic girdle, vertebrae, and proximal long bones. Although primary malignant bone tumors account for less than 1% of all confirmed cancer cases annually, they carry high morbidity and mortality rates. Furthermore, the prognosis is usually closely associated with recurrence and metastasis.[7]

Secondary bone tumors result from the metastasis of malignancies from other organs. It is reported that bone is one of the organs most frequently affected by metastatic cancer.[8,9] Although the exact mechanism of bone metastasis has not been fully elucidated, the bidirectional interaction between tumor cells and the bone microenvironment, along with genetic alterations in primary tumor cells, is considered a critical factor in the occurrence and progression of bone metastasis.[10–13] It is well established that bone metastasis is associated with various skeletal-related events, such as spinal cord compression, pathological fractures, or hypercalcemia, which lead to a decline in quality of life and increased mortality.[12,14]

2 Introduction to clinical prediction models (CPMs)

In current clinical practice, pathology, immunohistochemistry, and imaging hold vital positions in the diagnosis and differential diagnosis of bone-related malignancies; however, many difficulties still persist in the diagnostic process. Research indicates that, to date, the rates of misdiagnosis and missed diagnosis for osteosarcoma based on imaging and medical history are high, particularly in elderly patients, with an incidence of approximately 23% to 43%.[15–17] Wurtz et al.[18] reported that the average diagnostic delay for Ewing’s sarcoma patients is about 10 months, while another study found a misdiagnosis rate of 80.77%.[19] The diagnosis of bone metastases with an unknown primary site is also typically challenging, and confirming the diagnosis for such patients usually requires a combination of positron emission tomography-computed tomography (PET-CT) and other auxiliary examinations.[20]

Given the aforementioned issues, it is clinically crucial to combine multiple types of useful information to achieve a more precise estimation method. A CPM is a mathematical tool used to predict a specific future event. In disease research, CPMs can integrate multiple risk factors into a unified calculation tool through various statistical methods, thereby enabling individualized prediction and evaluation for each specific patient. The construction of any model follows a similar basic principle: identifying factors that may cause or influence a specific outcome (such as recurrence, metastasis, treatment efficacy, or prognosis), and then utilizing this information to predict the probability of that outcome occurring in an individual in the future. Depending on the clinical outcome under study, CPMs are categorized into two types: diagnostic models and prognostic models. Diagnostic models are used to estimate the probability of a specific individual having a certain disease, whereas prognostic models primarily evaluate the probability of a certain outcome occurring within a specific future timeframe.[21]

The primary expression formats of CPMs include the following three:

● Scoring scales: A traditional prediction model presented in the form of text and options, frequently seen in clinical guidelines for various diseases.[22]

● Nomograms: A graphical format primarily composed of several lines marked proportionally. By reading the scores for each factor line, the probability of the outcome can be predicted on the outcome line. Nomograms have been widely studied in recent years and are the most utilized clinical models in bone tumor modeling.[23,24]

● Web applications: A relatively new model format, usually based on nomograms, which further visualizes calculation formulas using programming languages (such as R, Python, and C++).[25,26] Compared with traditional nomograms, web applications offer more intuitive and effective outcome predictions. Since results are calculated by computer algorithms, manual calculation errors are minimized by automating the complex mathematics involved in nomogram interpretation, although the accuracy of the prediction still depends on the user for correct data entry. While web applications minimize calculation errors, their deployment requires robust data encryption and privacy protocols to protect sensitive patient information in accordance with medical data regulations.

In addition to the three common types above, CPMs also have some less common expression formats, such as tables, other graphical formats, and mobile software.

3 Development and application of CPMs in spinal tumors

3.1 Primary spinal tumors

Primary spinal tumors are relatively rare, accounting for only about 10% of all spinal tumors.[27,28] Due to their low incidence, the development of CPMs for primary spinal tumors is relatively limited and is mainly based on multicenter collaborations or public databases. A summary of representative CPMs for primary spinal tumors, including their target populations and key predictors, is presented in Table 1.

A retrospective study evaluated the prognostic value of inflammatory markers and preoperative D-dimer levels in the Ewing’s sarcoma family of tumors. The results showed that preoperative D-dimer levels and the C-reactive protein/albumin ratio are independent prognostic factors for overall survival (OS).[29] By combining age, surgical approach, and metastasis information, the authors successfully constructed a CPM for the Ewing’s sarcoma tumor family. Based on four machine learning algorithms—Cox, Random Survival Forest, CoxBoost, and DeepCox—Fan et al. developed a survival prediction model for spinal and pelvic Ewing’s sarcoma. This model outperformed the traditional American Joint Committee on Cancer (AJCC) staging system in predicting OS and cancer-specific survival (CSS).[30] Huang et al.[31] established a specific prediction model for the survival of primary spinal osteosarcoma patients, utilizing age, tumor grade, surgical status, and tumor size to determine 1-, 3-, and 5-year OS. Research on chordoma found that age, disease progression, tumor size, and surgical treatment are independent risk factors for the prognosis of classic spinal chordoma patients; the researchers subsequently constructed a nomogram incorporating these four characteristics.[32]

To date, surgery remains the optimal treatment for primary spinal tumors; however, the prognosis for elderly patients who cannot undergo surgery is often difficult to assess.[33] To address this, Huang et al.[34] developed a prognostic nomogram for elderly primary spinal tumor patients who are unable to tolerate or refuse surgery. The model consists of age, histological type, and tumor stage. Validation results indicated that the model effectively identifies individuals at high risk for survival within the elderly population.

3.2 Spinal metastases

Spinal metastases are more common than primary spinal tumors. Statistics show that approximately 20% to 40% of cancer patients will develop spinal metastases, with up to 20% of these patients experiencing symptoms due to spinal cord compression.[35] The treatment goal for spinal metastasis patients is palliative, focusing on maintaining or restoring neurological function, controlling pain, improving quality of life, and maintaining spinal stability. The development and application of CPMs in spinal metastatic disease primarily focus on predicting patient survival rates.

Tokuhashi et al.[36,37] introduced the first preoperative prognostic scoring system for spinal metastases in 1990, which was later modified and released as the Revised Tokuhashi Score in 2005. This system provides a comprehensive evaluation based on general condition, primary site, metastasis status (spinal, extra-spinal bone, and major visceral metastases), and the degree of spinal cord–derived paralysis, serving as a reliable basis for determining survival and treatment options.[35]

The Tomita Score, developed by Professor Tomita’s team at Kanazawa University in 2001, is another widely used prognostic model for spinal metastases. Prognostic factors in this model include the biological behavior of the primary cancer (growth rate), visceral metastasis status (treatability), and bone metastasis status (isolated, single, or multiple).[38] Summing these three factors yields a prognostic score, which determines the treatment strategy: a score of 2 to 3 suggests wide or marginal resection for long-term control; 4 to 5 suggests marginal or intralesional resection; 6 to 7 suggests palliative surgery for short-term relief; and 8 to 10 suggests nonsurgical treatment.[38]

Bauer and Wedin[39] evaluated the postoperative survival of 153 extremity bone metastasis patients and 88 spinal metastasis patients, identifying pathological fractures, visceral metastases, brain metastases, and bronchial lung cancer as risk factors, while isolated bone metastasis, breast cancer, renal cancer, myeloma, and lymphoma were protective factors. Based on these variables, the Bauer prediction model was developed to categorize patients into prognostic groups. Because pathological fracture was never proven to be a prognostic factor for spinal metastasis, the Medical University of Graz modified the Bauer score for spinal applications. The Revised Bauer Scoring System includes four factors: visceral metastasis, isolated bone metastasis, primary lung cancer, and primary breast/renal cancer, lymphoma, or myeloma. Studies show this system effectively assists in determining treatment strategies for spinal metastasis patients.[40]

Notably, the Skeletal Oncology Research Group (SORG) recently developed a prognostic algorithm and nomogram for 30-day, 90-day, and 365-day survival rates in spinal metastatic disease.[41,42] The SORG algorithm indicates that advanced age, poor performance status, primary tumor type, multiple spinal metastases, lung/liver metastasis, brain metastasis, any preoperative systemic therapy (chemotherapy, immunotherapy, hormone therapy, etc), higher white blood cell counts, and lower hemoglobin levels are significantly associated with decreased survival. The SORG nomogram is shown in Figure 1.

To validate the clinical efficacy of spinal metastasis models, a US-based study compared the value of multiple models, finding that the SORG nomogram had the highest accuracy in predicting postoperative survival.[43] Similarly, Li et al.[44] applied these models to a cohort of 268 Chinese patients, where the SORG model demonstrated superior predictive performance compared with other scoring systems. The success of these site-specific models in spinal oncology underscores the necessity of having similarly tailored prognostic tools for malignant involvement in other skeletal regions.

4 Development and application of CPMs in extra-spinal bone tumors

Expanding beyond the axial skeleton, research on CPMs for extra-spinal bone tumors has primarily focused on the prognosis of extremity bone metastases. In 2016, Sørensen et al.[45] published the 2008-SPRING model, a survival prediction nomogram for patients undergoing bone resection and reconstruction for extremity metastases. The authors utilized seven variables: hemoglobin, visceral metastasis, multiple bone metastases, fracture status (impending/actual), Karnofsky score, American Society of Anaesthesiologist’s score (ASA), and primary cancer group, selecting 3, 6, and 12 months as survival milestones. In 2018, this was updated to the 2013-SPRING model. Comparisons of area under the curve and Brier scores showed that the 2013-SPRING model performed better, with strong calibration and discrimination for 3-, 6-, and 12-month OS.

Leiden University developed the OptiModel score to predict survival in patients with symptomatic long bone metastases by analyzing 1,520 patients across 6 Dutch hospitals.[46] This model includes only three variables: clinical condition classification (based on Bollen et al.),[47] Karnofsky score, and the presence of visceral/brain metastasis. External validation in 250 surgical patients showed high similarity between actual and expected survival. The model was subsequently launched as a web application.[46] In 2019, Meares et al.[48] found OptiModel to be the most accurate for 12- and 24-month survival predictions in femoral metastasis patients.

The PATHfx model, based on a Bayesian network, was introduced by Forsberg et al.[49] in 2011 to estimate 3- and 12-month survival in systemic bone metastasis patients. It originally used 10 factors: age, sex, surgical indication, number of metastases, surgeon’s estimate, visceral metastasis, lymph node metastasis, hemoglobin, lymphocyte count, and Katagiri score.[50] Validation in Scandinavian, Italian, and Japanese cohorts confirmed its accuracy and clinical utility.[51,52]

Errani et al.[53] published the IOR scoring system in 2021 based on a prospective cohort of patients with symptomatic long bone metastases treated with minor or major surgery. The IOR score showed good discrimination for 12-month survival. Alfaro et al.[54] compared multiple models (Revised Katagiri, PATHfx, OptiModel, and IOR) in a Chilean population, concluding that the IOR score was the most accurate for predicting 12-month OS.

5 Summary

Numerous prediction models have been successfully developed and applied in bone tumor research, with the potential to become routine auxiliary tools. However, these models are not perfect. Most current constructions are limited by retrospective, single-center designs and small sample sizes. Furthermore, while advanced machine learning algorithms offer enhanced predictive power, their “black box” nature often lacks the transparency and interpretability essential for clinical trust and decision-making. Therefore, extensive multicenter validation and a greater emphasis on model explainability are required before clinical implementation to ensure reliability, accuracy, and practical utility.

References

[1] Casali PG, Bielack S, Abecassis N, et al.; ESMO Guidelines Committee, PaedCan and ERN EURACAN. Bone sarcomas: ESMO-PaedCan-EURACAN Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol. 2018;29(Suppl 4):iv79–95.

[2] Valery PC, Laversanne M, Bray F. Bone cancer incidence by morphological subtype: a global assessment. Cancer Causes Control. 2015;26:1127–39.

[3] Balamuth NJ, Womer RB. Ewing’s sarcoma. Lancet Oncol. 2010;11:184–92.

[4] Rajwanshi A, Srinivas R, Upasana G. Malignant small round cell tumors. J Cytol. 2009;26:1–10.

[5] Biermann JS, Chow W, Reed DR, et al. NCCN guidelines insights: bone cancer, version 2.2017. J Natl Compr Canc Netw. 2017;15:155–67.

[6] Hoang TM, Nguyen MT, Chen W, et al. Establishment and validation of an individualized nomogram to predict distant metastasis in chondrosarcoma patients: a population-based study. Transl Cancer Res. 2022;11:327–38.

[7] Ferguson JL, Turner SP. Bone cancer: diagnosis and treatment principles. Am Fam Physician. 2018;98:205–13.

[8] Coleman RE. Metastatic bone disease: clinical features, pathophysiology and treatment strategies. Cancer Treat Rev. 2001;27:165–76.

[9] Mundy GR. Metastasis to bone: causes, consequences and therapeutic opportunities. Nat Rev Cancer. 2002;2:584–93.

[10] Kang Y, Siegel PM, Shu W, et al. A multigenic program mediating breast cancer metastasis to bone. Cancer Cell. 2003;3:537–49.

[11] Yin JJ, Pollock CB, Kelly K. Mechanisms of cancer metastasis to the bone. Cell Res. 2005;15:57–62.

[12] Jin JK, Dayyani F, Gallick GE. Steps in prostate cancer progression that lead to bone metastasis. Int J Cancer. 2011;128:2545–61.

[13] Patel LR, Camacho DF, Shiozawa Y, Pienta KJ, Taichman RS. Mechanisms of cancer cell metastasis to the bone: a multistep process. Future Oncol. 2011;7:1285–97.

[14] Decroisette C, Monnet I, Berard H, et al.; Groupe Français de Pneumo-Cancérologie 0601 Team. Epidemiology and treatment costs of bone metastases from lung cancer: a French prospective, observational, multicenter study (GFPC 0601). J Thorac Oncol. 2011;6:576–82.

[15] Chen DS, Mellman I. Oncology meets immunology: the cancer-immunity cycle. Immunity. 2013;39:1–10.

[16] Wu T, Dai Y. Tumor microenvironment and therapeutic response. Cancer Lett. 2017;387:61–8.

[17] Clara JA, Monge C, Yang Y, Takebe N. Targeting signalling pathways and the immune microenvironment of cancer stem cells - a clinical update. Nat Rev Clin Oncol. 2020;17:204–32.

[18] Wurtz LD, Peabody TD, Simon MA. Delay in the diagnosis and treatment of primary bone sarcoma of the pelvis. J Bone Joint Surg Am. 1999;81:317–25.

[19] Widhe B, Widhe T, Bauer HC. Ewing sarcoma of the rib--initial symptoms and clinical features: tumor missed at the first visit in 21 of 26 patients. Acta Orthop. 2007;78:840–4.

[20] Guzik G, Barańska B. Metastasis of adenocarcinoma of an unknown primary site to the right crus simulating sarcoma - a case report. Ortop Traumatol Rehabil. 2013;15:273–9.

[21] Gu HQ, Zhou ZR, Zhang ZH, et al. Clinical prediction models: basic concepts, application scenarios, and research strategies. Chin J Evid-based Cardiovasc Med. 2018;10:1454–1456, 1462.

[22] Strauss SJ, Frezza AM, Abecassis N, et al.; ESMO Guidelines Committee, EURACAN, GENTURIS and ERN PaedCan. Electronic address: [e-mail]. Bone sarcomas: ESMO-EURACAN-GENTURIS-ERN PaedCan clinical practice guideline for diagnosis, treatment and follow-up. Ann Oncol. 2021;32:1520–36.

[23] Iasonos A, Schrag D, Raj GV, Panageas KS. How to build and interpret a nomogram for cancer prognosis. J Clin Oncol. 2008;26:1364–70.

[24] Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16:e173–80.

[25] Yang S, Zhou H, Feng C, et al. Web-based nomograms for overall survival and cancer-specific survival of bladder cancer patients with bone metastasis: a retrospective cohort study from SEER database. J Clin Med. 2023;12:726.

[26] Li W, Xu C, Hu Z, et al. A visualized dynamic prediction model for lymphatic metastasis in Ewing’s sarcoma for smart medical services. Front Public Health. 2022;10:877736.

[27] Kerr DL, Dial BL, Lazarides AL, et al. Epidemiologic and survival trends in adult primary bone tumors of the spine. Spine J. 2019;19:1941–9.

[28] Li DH, Chen X, Hao B, et al. Advances in surgical treatment of primary spinal tumors. Chin J Neurosurg Dis Res. 2018;17:564–6.

[29] Xu K, Lou Y, Sun R, et al. Establishment of a nomogram-based model for predicting the prognostic value of inflammatory biomarkers and preoperative D-dimer level in spinal Ewing’s sarcoma family tumors: a retrospective study of 83 patients. World Neurosurg. 2019;121:e104–12.

[30] Fan G, Yang S, Qin J, et al. Machine learning predict survivals of spinal and pelvic Ewing’s sarcoma with the SEER database. Global Spine J. 2022;14:1125–36.

[31] Huang Z, Huang C, Wang Y, et al. Clinical features, risk factors, and prediction nomogram for primary spinal osteosarcoma: a large-cohort retrospective study. Global Spine J. 2022;14:930–40.

[32] Lin K, Song K, Wang S, Jiang L, Wang H, Dong J. Predict overall survival of spinal conventional chordoma: development and assessment of a new predictive nomogram. Clin Neurol Neurosurg. 2020;197:106174.

[33] Missenard G, Bouthors C, Fadel E, Court C. Surgical strategies for primary malignant tumors of the thoracic and lumbar spine. Orthop Traumatol Surg Res. 2020;106:S53–62.

[34] Huang Z, Zhao Z, Wang Y, Wu Y, Guo C, Kong Q. Clinical characteristics, prognostic factors, and predictive model for elderly primary spinal tumor patients who are difficult to tolerate surgery or refuse surgery. Front Oncol. 2022;12:991599.

[35] Barzilai O, Fisher CG, Bilsky MH. State of the art treatment of spinal metastatic disease. Neurosurgery. 2018;82:757–69.

[36] Tokuhashi Y, Matsuzaki H, Toriyama S, Kawano H, Ohsaka S. Scoring system for the preoperative evaluation of metastatic spine tumor prognosis. Spine (Phila Pa 1976). 1990;15:1110–3.

[37] 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;30:2186–91.

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

[39] Bauer HC, Wedin R. Survival after surgery for spinal and extremity metastases. Prognostication in 241 patients. Acta Orthop Scand. 1995;66:143–6.

[40] Leithner A, Radl R, Gruber G, et al. Predictive value of seven preoperative prognostic scoring systems for spinal metastases. Eur Spine J. 2008;17:1488–95.

[41] Paulino Pereira NR, Mclaughlin L, Janssen SJ, et al. The SORG nomogram accurately predicts 3- and 12-months survival for operable spine metastatic disease: external validation. J Surg Oncol. 2017;115:1019–27.

[42] Paulino Pereira NR, Janssen SJ, Van Dijk E, et al. Development of a prognostic survival algorithm for patients with metastatic spine disease. J Bone Joint Surg Am. 2016;98:1767–76.

[43] Ahmed AK, Goodwin CR, Heravi A, et al. Predicting survival for metastatic spine disease: a comparison of nine scoring systems. Spine J. 2018;18:1804–14.

[44] Li Z, Guo L, Guo B, et al. Evaluation of different scoring systems for spinal metastases based on a Chinese cohort. Cancer Med. 2022;12:4125–36.

[45] Sørensen MS, Gerds TA, Hindsø K, Petersen MM. Prediction of survival after surgery due to skeletal metastases in the extremities. Bone Joint J. 2016;98-B:271–7.

[46] Willeumier JJ, van der Linden YM, van der Wal CWPG, et al. An easy-to-use prognostic model for survival estimation for patients with symptomatic long bone metastases. J Bone Joint Surg Am. 2018;100:196–204.

[47] Bollen L, van der Linden YM, Pondaag W, 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;16:991–8.

[48] 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.

[49] 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:e19956.

[50] Forsberg JA, Wedin R, Bauer HC, et al. External validation of the Bayesian Estimated Tools for Survival (BETS) models in patients with surgically treated skeletal metastases. BMC Cancer. 2012;12:493.

[51] Piccioli A, Spinelli MS, Forsberg JA, 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.

[52] Ogura K, Gokita T, Shinoda Y, et al. Can a multivariate model for survival estimation in skeletal metastases (PATHFx) be externally validated using Japanese patients? Clin Orthop Relat Res. 2017;475:2263–70.

[53] Errani C, Cosentino M, Ciani G, et al. C-reactive protein and tumour diagnosis predict survival in patients treated surgically for long bone metastases. Int Orthop. 2021;45:1337–46.

[54] Alfaro PA, Delgado J, Dumas A, et al. Comparison between different prognostic models to be used for metastatic bone disease on appendicular skeleton in a Chilean population. Eur J Orthop Surg Traumatol. 2021;31:1657–62.

RIGHTS & PERMISSIONS

The Author(s) 2026. Published by Wolters Kluwer Health, LLC. on behalf of Higher Education Press.