Operative treatment of spine metastases aims to maintain quality of life1-5. The decision for a surgical procedure and the corresponding operative strategy are often based on the patients’ estimated life expectancies4,5. Current prognostication models are not built for estimation of short-term survival (30 days or 90 days)6-11, and some studies suggest a lack of accuracy3,4,6,7. The identification of new risk factors and the development of more sophisticated algorithms might improve the accuracy of prognostication model estimates12,13.
The traditional method of constructing a survival algorithm is to round the effect estimates (e.g., hazard ratios [HRs]) of prognostic factors, to sum up the factors that are present in a patient, and to relate these to survival estimates8-11,14,15. A nomogram is a more advanced method that more fully describes the precision of effect estimates on prognostic factors; it provides a user-friendly figure that includes prognostic factors set to a common point scale16,17. This scale translates into an individualized survival probability. Another, more complex computer-based method is boosting (a form of machine learning), which assesses the combination of a multitude of factors into regression trees to optimize predictive power18,19.
Our study aim was to assess factors associated with survival; to use these factors to create a classic scoring algorithm, nomogram, and boosting algorithm; and to test the predictive accuracy of the three created algorithms at estimating survival.
Materials and Methods
Study Design and Patient Selection
This retrospective study was approved by our institutional review board and a waiver of consent was obtained. Patients who were ≥18 years of age and who had undergone a surgical procedure for spine metastases between January 2002 and January 2014 at two tertiary care centers were included. The spine encompassed the cervical, thoracic, and lumbar vertebrae. Lymphoma and multiple myeloma were included4. We excluded patients treated by kyphoplasty or vertebroplasty only (rarely done and only in the most palliative setting), radiosurgery, or revision procedures. Only the first procedure was included in patients who underwent multiple procedures, to avoid violating the statistical rule of independence. Choice of treatment was decided by mutual agreement between the surgeon and the patient.
We identified 1,330 patients with an International Classification of Diseases, Ninth Revision (ICD-9) diagnosis of pathologic fracture of vertebrae (733.13). Additionally, we used a word-based query to search operative reports in our orthopaedic oncology database and identified 796 new patients. Medical records of 2,126 patients were screened, and 649 patients were included in this study.
The outcome was survival, defined as the time from the surgical procedure until death from any cause. We set the latest follow-up moment for survival on March 18, 2015. This was the most recent date that a death had occurred in our cohort according to the Social Security Death Index (SSDI).
Factors known or suggested to be associated with survival were included as explanatory variables4,9-11,13-15,20,21. Medical records were manually searched for primary cancer type, location, operative technique, fracture type, symptoms, preoperative American Spinal Injury Association (ASIA) impairment scale, Eastern Cooperative Oncology Group (ECOG) performance status, number of skeletal and visceral metastases, date of primary cancer diagnosis, local radiation therapy, and any systemic therapy prior to the surgical procedure (all forms of nonsurgical and non-radiotherapeutic adjuvants [chemotherapy, immunotherapy, hormone therapy, and metabolic therapy], regardless of the dose, response, and period before the surgical procedure). Two researchers performed manual data collection, and another researcher cross-checked a random 10% sample of the data to ensure a robust database. No systematic inconsistencies were noticed, except for establishing the ECOG performance status; dichotomizing the ECOG performance status into a relatively good score (0, 1, or 2) or a poor score (3 to 4) resolved discrepancies4,22.
As we expected, there was a nonlinear association between body mass index (BMI) and survival; therefore, we categorized BMI as underweight (patients who had a BMI of <18.5 kg/m2), normal weight (patients who had a BMI of 18.5 to 30 kg/m2), and overweight (patients who had a BMI of >30 kg/m2)23. The modified Charlson Comorbidity Index was used to determine comorbidity status using an ICD-9 code-based algorithm classifying 12 comorbidities (e.g., chronic pulmonary disease, diabetes) (see Appendix)24,25. We dichotomized comorbidity status into any additional comorbidity (in addition to the metastases) or none. The presence of a pathologic fracture was based on radiology reports. The time between the start of neurological symptoms and the surgical procedure was categorized into no neurological symptoms, acute to subacute spinal cord injury (<14 days), and chronic spinal cord injury (≥14 days)26. We used the preoperative ASIA impairment scale to determine if a patient had any neurological deficits (score of A, B, C, or D) or none (score of E); we believed that this cutoff point was most reliable. Patients with prior but no current deficits were classified as ASIA E27. We arbitrarily categorized time between primary cancer diagnosis and the surgical procedure into recent (<30 days) and not recent (≥30 days). Based on a study by Katagiri et al.28, we dichotomized primary cancer types into cancers with a relatively good prognosis (referred to as the good prognosis group and including patients with lymphoma, multiple myeloma, breast cancer, kidney cancer, prostate cancer, or thyroid cancer) and cancers with a relatively poor prognosis (referred to as the poor prognosis group and including patients with lung cancer, colon cancer, rectal cancer, bladder cancer, esophageal cancer, liver cancer, melanoma, gastric cancer, or other cancers). On the basis of the study by Nathan et al.4, we categorized metastases to those in other bones (outside the spine) and to those in other organs (none, liver and/or lung, and brain). We believe that laboratory values closest to the day of the surgical procedure, with a maximum range of 7 days before the surgical procedure, reflect routine preoperative assessment.
Bivariate Cox regression was performed to assess what factors were marginally associated with survival (p < 0.10). Log-rank curves were visualized for each explanatory variable and the proportional hazards assumption on the basis of Schoenfeld residuals was tested. We accepted departure from the proportional hazards assumption as long as log-rank curves did not cross for the first 3 years29. All variables marginally associated with survival were included in a stepwise, backward, multivariate Cox proportional hazards model to assess independent association with survival (variables with p values of <0.05 were retained). We used multiple chained imputation to estimate missing values (40 imputations).
On the basis of factors independently associated with survival, we developed three prognostic scoring algorithms: a classic scoring algorithm, a nomogram, and a boosting algorithm. The accuracy for predicting 30-day, 90-day, and 365-day survival in each of the three scoring algorithms was tested using receiver operating characteristic (ROC) analysis with the corresponding area under the curve (AUC).
We used fivefold cross-validation to assess overfitting of the developed algorithms for every imputed data set19,30. By this technique, the algorithms were constructed using a random 80% of the data set and were tested on the remaining 20% of the data set, which was repeated 5 times per imputed data set; we averaged the AUCs, 95% confidence intervals (CIs), and p values to demonstrate and to compare the accuracy of the three algorithms.
Two-tailed p values of <0.05 were considered significant.
Building the Classic Scoring Algorithm, Nomogram, and Boosted Algorithm
The classic scoring algorithm was developed by summing the points for each prognostic factor, based on the nearest integer of the HR (Table I)14,28. To allow scoring for the classic scoring algorithm, we categorized age (<65 years compared with ≥65 years) and spine metastases (1 compared with >1) on the basis of cutoff points from Nathan et al.4. We categorized white blood-cell count (<11,000/μL compared with ≥11,000/μL) and hemoglobin levels (≤10 g/dL compared with >10 g/dL) on the basis of our institutional reference values. The total number of points was categorized into three prognostic groups on the basis of the distribution of points in our population: those with a good prognosis (0 to 2 points), intermediate prognosis (3 to 4 points), and poor prognosis (5 to 12 points) (Fig. 1). Subsequently, we calculated the survival probabilities per prognostic group (Table II)14,28.
The nomogram was constructed using the β regression coefficient of every factor independently associated with survival and transforming these to a scale from 0 to 10016. The eventual nomogram was averaged over the 40 imputed nomograms. The total sum of points corresponded to the 30-day, 90-day, and 365-day survival probability (Fig. 2).
Third, we developed a boosting algorithm. Boosting is a flexible regression method and has emerged as one of the most powerful methods for predictive data mining19. The general idea is to compute a regression tree, whereafter consecutive trees are built for the prediction residuals of the prior tree. We performed a boosted regression on the outcome of death 30 days, 90 days, and 365 days after the surgical procedure. Parameters for the boosted regression were based on a study by Schonlau19 and were set at 0.5 for bagging, 3 for interactions, 0.01 for shrinkage, and 10,000 for the number of iterations (Appendix).
The mean age of the 649 patients in our study group was 60 years, 377 (58%) were men, and the mean BMI was 27 kg/m2 (Table III). The primary cancers were lung cancer (n = 115), kidney cancer (n = 82), breast cancer (n = 77), multiple myeloma (n = 72), prostate cancer (n = 57), and other cancers (n = 246) (Table IV). The median follow-up was 11 months (interquartile range, 3 to 33 months), and all patients who were alive at 30 days (n = 597), 90 days (n = 495), and 365 days (n = 321) were included in the follow-up (Fig. 3).
Factors Associated with Survival
In bivariate analysis, older age (HR, 1.01 [95% CI, 1.00 to 1.02]; p = 0.016), additional comorbidities (HR, 1.31 [95% CI, 1.10 to 1.56]; p = 0.002), acute to subacute spinal cord injury (HR, 1.41 [95% CI, 1.15 to 1.72]; p = 0.001), neurological deficits (HR, 1.29 [95% CI, 1.08 to 1.54]; p = 0.004), poor performance status (HR, 1.93 [95% CI, 1.50 to 2.47]; p < 0.001), primary cancers in the poor prognosis group (HR, 2.01 [95% CI, 1.67 to 2.41]; p < 0.001), >1 spine metastasis (HR, 1.57 [95% CI, 1.28 to 1.92]; p < 0.001), other bone metastases outside the spine (HR, 1.34 [95% CI, 1.13 to 1.60]; p = 0.001), lung and/or liver metastasis (HR, 1.59 [95% CI, 1.30 to 1.94]; p < 0.001), brain metastasis (HR, 2.55 [95% CI, 1.95 to 3.33]; p < 0.001), previous systemic therapy (HR, 1.68 [95% CI, 1.40 to 2.01]; p < 0.001), higher white blood-cell count (HR, 1.02 [95% CI, 1.01 to 1.04]; p = 0.006), and lower hemoglobin levels (HR, 0.92 [95% CI, 0.87 to 0.97]; p = 0.002) were associated with decreased survival (Table V).
In multivariate analysis, the following factors were independently associated with decreased survival: older age (HR, 1.01 [95% CI, 1.00 to 1.02]; p = 0.009), poor performance status (HR, 1.54 [95% CI, 1.19 to 2.00]; p = 0.001), primary cancers in the poor prognosis group (HR, 1.68 [95% CI, 1.40 to 2.04]; p < 0.001), >1 spine metastasis (HR, 1.32 [95% CI, 1.07 to 1.63]; p = 0.009), lung and/or liver metastasis (HR, 1.35 [95% CI, 1.09 to 1.66]; p = 0.005), brain metastasis (HR, 1.90 [95% CI, 1.43 to 2.51]; p < 0.001), previous systemic therapy (HR, 1.65 [95% CI, 1.36 to 2.00]; p < 0.001), higher white blood-cell count (HR, 1.03 [95% CI, 1.01 to 1.04]; p = 0.002), and low hemoglobin levels (HR, 0.92 [95% CI, 0.87 to 0.98]; p = 0.009) (Table VI).
Accuracy of the Prediction Algorithms
The accuracy (AUC) of the survival estimates of the classic scoring algorithm was 0.70 for 30 days, 0.69 for 90 days, and 0.73 for 365 days for both the training (in-sample) and the test data sets (out-of-sample) (Table VII). The nomogram performed better than the classic scoring algorithm for all time points, with an accuracy of 0.76 for 30 days, 0.74 for 90 days, and 0.77 for 365 days with relatively reliable out-of-sample (test data set) estimations. Notably, the boosting algorithm performed better at all time points for in-sample (training data sets) survival estimation compared with the other two algorithms, yet out-of-sample (test data sets) survival estimation proved slightly worse compared with the nomogram.
Older age; poor performance status; primary cancers in the poor prognosis group (lung and other); >1 spine metastasis; presence of a lung, liver, and/or brain metastasis; previous systemic therapy; increased white blood-cell count; and decreased hemoglobin levels were independently associated with worse survival in patients with spine metastases. The boosting algorithm was most accurate for the training data set at all three time points (p < 0.001); however, the accuracy of the boosting algorithm decreased to the level of the nomogram on the testing data sets. Although there was no significant difference among the three algorithms on the test data sets, we suggest using the nomogram to aid surgical decision-making in clinical practice, as AUCs were highest and because it appreciates continuous predictors without categorization.
This study had limitations. First, no uniform criteria were used to determine treatment. This may have resulted in a study population with relatively good prognosis, as some patients might have been determined to be too sick to undergo a surgical procedure. Therefore, our model is most applicable for patients at least considered for a surgical procedure. Second, we did not include variables with >35% missing values (e.g., albumin). These factors might have increased the accuracy of our algorithms. Third, our ICD-9 and word-based search might have missed eligible patients. However, we expect this number to be low and not influence our results. Fourth, algorithms are always more accurate for the sample on which they are developed; external validation is necessary to demonstrate generalizability31. Fifth, we did not exploit the potential power of the boosting algorithm, as we included only those factors independently associated with survival. Inclusion of all possible explanatory variables might have improved the accuracy, but at the cost of clinical practicality and external validity18. Sixth, we did not determine the cause of death; non-cancer-related deaths (e.g., car accident) might have compromised model accuracy and prognostic factors. However, we believe that the majority of patients died from cancer. Seventh, we might have missed outcomes because of potential inconsistencies in the SSDI. However, the SSDI has been shown to accurately reflect survival for older oncology patients32. Eighth, some explanatory variables (e.g., ASIA impairment scale) might suffer interobserver variability; however, we chose cutoff points that we believed were most reliable.
Adding factors associated with outcome to prediction models improves their accuracies12,21. We identified three new factors associated with survival, specifically for patients with metastatic spine disease: previous systemic therapy, white blood-cell count, and hemoglobin levels. Previous systemic therapy regimens varied among patients (e.g., based on cancer type, health status, tolerance), and these variations might have influenced survival; further breakdown of systemic therapy could improve accuracy of prognostication, and we see this as a limitation. However, systemic therapy in general was associated with worse survival in our study. We believe that this is a reflection of a more developed cancer stage, rather than the therapy itself4. Increased lymphocyte counts and decreased hemoglobin concentration have been associated with survival in patients with bone metastases, including extremity metastases12. Inflammation in the cancer microenvironment has been shown to increase proliferation and survival in malignant cells, promoting angiogenesis and (further) metastasis; this might explain why higher white blood-cell counts were associated with decreased survival12,33. Lower red blood-cell counts are thought to reflect worse chronic disease status12.
Previous studies have demonstrated that age20, performance status8,11,12,15,34, pathologic fracture12,14, neurological compromise8,9,12,15, primary cancer type8-12,14,15,34, >1 spine metastasis8-10,14,15,34, extraspinal bone metastases8,10,12,14,15, visceral metastases8,10,11,14,15,34, additional comorbidities35, and BMI36 were associated with decreased survival (see Appendix). In our study, pathologic fracture, neurological compromise, extraspinal bone metastases, additional comorbidities, and BMI were not independently associated with survival. Different definitions of variables, statistical approaches, and selection biases might explain these differences. When added to the multivariate regression model, extraspinal bone metastases were marginally associated with survival (p = 0.064). Patients with a low BMI less frequently underwent a surgical procedure (3%); this could have neutralized BMI as a possible predictor.
As the performance status has been associated with quality of life34,37, it would be interesting to add the score of a questionnaire (e.g., the European Quality of Life-5 Dimensions [EQ-5D]) to prognostic models.
Compared with our models, other models were based on smaller sample sizes and therefore were potentially missing parameters that were only weakly associated with survival (see Appendix). A future study should externally validate and compare existing models.
Cancer-specific prognostic models might result in more accurate predictions and merit further study; however, treatment shifts occur that affect survival, and one would need large numbers to obtain robust models. We compared the performance of the nomogram between patients with good prognosis cancers (n = 431) and those with poor prognosis cancers (n = 218) and found no substantial differences.
The suggestion that expert clinicians predict survival more accurately compared with scoring models has been described4. This shows that prognostic models are not capable of including all factors (such as psychosocial factors) and should be used as a tool to assist the surgeon in his or her decision-making.
In conclusion, we identified risk factors associated with survival that should be considered in prognostication. Any systemic therapy for cancer prior to a surgical procedure, white blood-cell count, and hemoglobin levels are newly found prognostic factors and should, in addition to those already known, be taken into consideration when estimating survival in patients with spine metastases. A nomogram proved to be an accurate tool to predict survival and can be made available on web-based applications to assist the surgeon in decision-making. Currently, we are working on external validation of the developed algorithms and aim to develop an online tool to estimate survival for use in clinical practice.
Tables showing the conditions of the modified Charlson Comorbidity Index, number of patients per condition, and ICD-9 codes used for the algorithm; the computational parameters for boosted regression, explanation, recommended settings, and settings chosen for our study; and the most commonly used prognostication models to estimate survival in patients with spine metastases, factors included in their models, and comparison with our prognostication model are available with the online version of this article as a data supplement at jbjs.org.
NOTE: We thank Harvard Catalyst (Boston, Massachusetts) for the statistical support during this study.
Investigation performed at Massachusetts General Hospital, Boston, and Brigham and Women’s Hospital, Boston, Massachusetts
Disclosure: There was no external funding for this study. On the Disclosure of Potential Conflicts of Interest forms, which are provided with the online version of the article, one or more of the authors checked “yes” to indicate that the author had a relevant financial relationship in the biomedical arena outside the submitted work.
1. Coleman RE. Clinical features of metastatic bone disease and risk of skeletal morbidity. Clin Cancer Res. 2006 ;12(20 Pt 2):6243s–9s.
2. Kim CH, Chung CK, Sohn S, Lee S, Park SB. Less invasive palliative surgery for spinal metastases. J Surg Oncol. 2013 ;108(7):499–503. Epub 2013 Aug 27.
3. Leithner A, Radl R, Gruber G, Hochegger M, Leithner K, Welkerling H, Rehak P, Windhager R. Predictive value of seven preoperative prognostic scoring systems for spinal metastases. Eur Spine J. 2008 ;17(11):1488–95. Epub 2008 Sep 12.
4. Nathan SS, Healey JH, Mellano D, Hoang B, Lewis I, Morris CD, Athanasian EA, Boland PJ. Survival in patients operated on for pathologic fracture: implications for end-of-life orthopedic care. J Clin Oncol. 2005 ;23(25):6072–82.
5. Quinn RH, Randall RL, Benevenia J, Berven SH, Raskin KA. Contemporary management of metastatic bone disease: tips and tools of the trade for general practitioners. J Bone Joint Surg Am. 2013 ;95(20):1887–95.
6. Ulmar B, Naumann U, Catalkaya S, Muche R, Cakir B, Schmidt R, Reichel H, Huch K. Prognosis scores of Tokuhashi and Tomita for patients with spinal metastases of renal cancer. Ann Surg Oncol. 2007 ;14(2):998–1004. Epub 2006 Nov 3.
7. Zoccali C, Skoch J, Walter CM, Torabi M, Borgstrom M, Baaj AA. The Tokuhashi score: effectiveness and pitfalls. Eur Spine J. 2016 ;25(3):673–8. Epub 2015 Dec 1.
8. 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(11):1110–3.
9. Sioutos PJ, Arbit E, Meshulam CF, Galicich JH. Spinal metastases from solid tumors. Analysis of factors affecting survival. Cancer. 1995 ;76(8):1453–9.
10. 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.
11. van der Linden YM, Dijkstra SP, Vonk EJ, Marijnen CA, Leer JW; Dutch Bone Metastasis Study Group. 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.
12. 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. Epub 2011 May 13.
13. Ghori AK, Leonard DA, Schoenfeld AJ, Saadat E, Scott N, Ferrone ML, Pearson AM, Harris MB. Modeling 1-year survival after surgery on the metastatic spine. Spine J. 2015 ;15(11):2345–50. Epub 2015 Jul 6.
14. Bauer HC, Wedin R. Survival after surgery for spinal and extremity metastases. Prognostication in 241 patients. Acta Orthop Scand. 1995 ;66(2):143–6.
15. 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.
16. Iasonos A, Schrag D, Raj GV, Panageas KS. How to build and interpret a nomogram for cancer prognosis. J Clin Oncol. 2008 ;26(8):1364–70.
17. Oberije C, De Ruysscher D, Houben R, van de Heuvel M, Uyterlinde W, Deasy JO, Belderbos J, Dingemans AM, Rimner A, Din S, Lambin P. A validated prediction model for overall survival from stage III non-small cell lung cancer: toward survival prediction for individual patients. Int J Radiat Oncol Biol Phys. 2015 ;92(4):935–44. Epub 2015 Apr 30.
18. Mayr A, Binder H, Gefeller O, Schmid M. The evolution of boosting algorithms. From machine learning to statistical modelling. Methods Inf Med. 2014;53(6):419–27. Epub 2014 Aug 12.
19. Schonlau M. Boosted regression (boosting): an introductory tutorial and a Stata plugin. Stata J. 2005;5(3):330–54.
20. Chi JH, Gokaslan Z, McCormick P, Tibbs PA, Kryscio RJ, Patchell RA. Selecting treatment for patients with malignant epidural spinal cord compression-does age matter?: results from a randomized clinical trial. Spine (Phila Pa 1976). 2009 ;34(5):431–5.
21. Tokuhashi Y, Ajiro Y, Umezawa N. Outcome of treatment for spinal metastases using scoring system for preoperative evaluation of prognosis. Spine (Phila Pa 1976). 2009 ;34(1):69–73.
22. Oken MM, Creech RH, Tormey DC, Horton J, Davis TE, McFadden ET, Carbone PP. Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol. 1982 ;5(6):649–55.
23. Goodwin PJ, Ennis M, Pritchard KI, Koo J, Trudeau ME, Hood N. Diet and breast cancer: evidence that extremes in diet are associated with poor survival. J Clin Oncol. 2003 ;21(13):2500–7.
24. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373–83.
25. Quan H, Li B, Couris CM, Fushimi K, Graham P, Hider P, Januel JM, Sundararajan V. 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. Epub 2011 Feb 17.
26. Karimi-Abdolrezaee S, Eftekharpour E, Wang J, Morshead CM, Fehlings MG. Delayed transplantation of adult neural precursor cells promotes remyelination and functional neurological recovery after spinal cord injury. J Neurosci. 2006 ;26(13):3377–89.
27. Kirshblum SC, Burns SP, Biering-Sorensen F, Donovan W, Graves DE, Jha A, Johansen M, Jones L, Krassioukov A, Mulcahey MJ, Schmidt-Read M, Waring W. International standards for neurological classification of spinal cord injury (revised 2011). J Spinal Cord Med. 2011 ;34(6):535–46.
28. 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.
29. Hosmer DLS. Applied survival analysis: regression modeling of time to event data. 1st ed. New York: John Wiley & Sons; 1999.
30. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996 ;15(4):361–87.
31. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015 ;350:g7594.
32. Huntington JT, Butterfield M, Fisher J, Torrent D, Bloomston M. The Social Security Death Index (SSDI) most accurately reflects true survival for older oncology patients. Am J Cancer Res. 2013 ;3(5):518–22.
33. Mantovani A, Allavena P, Sica A, Balkwill F. Cancer-related inflammation. Nature. 2008 ;454(7203):436–44.
34. Choi D, Fox Z, Albert T, Arts M, Balabaud L, Bunger C, Buchowski JM, Coppes MH, Depreitere B, Fehlings MG, Harrop J, Kawahara N, Martin-Benlloch JA, Massicotte EM, Mazel C, Oner FC, Peul W, Quraishi N, Tokuhashi Y, Tomita K, Verlaan JJ, Wang M, Crockard HA. Prediction of quality of life and survival after surgery for symptomatic spinal metastases: a multicenter cohort study to determine suitability for surgical treatment. Neurosurgery. 2015 ;77(5):698–708; discussion 708.
35. Piccirillo JF, Tierney RM, Costas I, Grove L, Spitznagel EL Jr. Prognostic importance of comorbidity in a hospital-based cancer registry. JAMA. 2004 ;291(20):2441–7.
36. Jung SH, Yang DH, Ahn JS, Lee SS, Ahn SY, Kim YK, Kim HJ, Lee JJ. Decreased body mass index is associated with poor prognosis in patients with multiple myeloma. Ann Hematol. 2014 ;93(5):835–40. Epub 2013 Dec 6.
37. Chong S, Shin SH, Yoo H, Lee SH, Kim KJ, Jahng TA, Gwak HS. Single-stage posterior decompression and stabilization for metastasis of the thoracic spine: prognostic factors for functional outcome and patients’ survival. Spine J. 2012 ;12(12):1083–92. Epub 2012 Nov 15.