Professional Documents
Culture Documents
The Oswestry Risk Index
The Oswestry Risk Index
B. Balain, The revised Tokuhashi, Tomita and modified Bauer scores are commonly used to make
A. Jaiswal, difficult decisions in the management of patients presenting with spinal metastases. A
J. M. Trivedi, prospective cohort study of 199 consecutive patients presenting with spinal metastases,
S. M. Eisenstein, treated with either surgery and/or radiotherapy, was used to compare the three systems.
J. H. Kuiper, Cox regression, Nagelkerke’s R2 and Harrell’s concordance were used to compare the
D. C. Jaffray systems and find their best predictive items. The three systems were equally good in terms
of overall prognostic performance. Their most predictive items were used to develop the
From The Centre for Oswestry Spinal Risk Index (OSRI), which has a similar concordance, but a larger coefficient
Spinal Studies, of determination than any of these three scores. A bootstrap procedure was used to
Robert Jones & internally validate this score and determine its prediction optimism.
Agnes Hunt The OSRI is a simple summation of two elements: primary tumour pathology (PTP) and
Orthopaedic and general condition (GC): OSRI = PTP + (2 – GC).
District Hospital This simple score can predict life expectancy accurately in patients presenting with spinal
NHS Trust, metastases. It will be helpful in making difficult clinical decisions without the delay of
Oswestry, United extensive investigations.
Kingdom
Cite this article: Bone Joint J 2013;95-B:210–16.
The decision-making process in the treatment derived for surgically treated patients, the
of patients with metastatic disease of the spine modified Bauer score,4 could distinguish
is seldom easy. The two main concerns in a significantly between patients with good, mod-
patient who might become paralysed are the erate or poor survival prospects.8,9 Three fur-
B. Balain, MS, FRCS, FRCS(Tr
patient’s life expectancy when they present and ther studies assessed the validity of the revised
& Orth), Consultant Spinal the potential for added morbidity from treat- Tokuhashi score.10-12 Two found it could pre-
Surgeon
A. Jaiswal, MS, MRCS,
ment. Various scoring systems have been pro- dict survival,10,11 but the third found it only
Speciality Trainee posed to predict life expectancy and help poorly predicted survival in a group of Euro-
J. M. Trivedi, FRCS, MCh,
FRCS(Tr & Orth), Consultant
decision making in these patients. Some are pean patients.12 The two original studies and
Spinal Surgeon applicable to all patients1,2 and others to spe- the five evaluative studies had a predominance
S. M. Eisenstein, FRCS,
Consultant Spinal Surgeon
cific subgroups, for example those treated sur- of surgically treated patients.
D. C. Jaffray, FRCS, gically,3,4 conservatively,5 or those with Various researchers have suggested modifica-
Consultant Spinal Surgeon
Robert Jones & Agnes Hunt
epidural metastases.6 A scoring system valid tions to these scores,13,14 or reported on specific
Orthopaedic and District for all patients, rather than a specific subset, conditions where they do not help.15,16 Internal
Hospital NHS Trust, Oswestry,
Shropshire SY10 7AG, UK.
would be most useful for deciding between and external validation can address such prob-
available treatments. It is therefore not surpris- lems, ideally using a measure of overall perfor-
J. H. Kuiper, PhD, Lecturer
Keele University, Institute for ing that the two scoring systems applicable to mance of the scoring system against individual
Science and Technology, all patients, the revised Tokuhashi1 and the elements that make up that system.7 To our
Staffordshire ST5 5BG, UK.
Tomita2 scores, are widely cited and used. knowledge, such methods have not been used
Correspondence should be sent
to Mr A. Jaiswal; e-mail:
The items that make up these scores and so far. In addition, the scoring systems could be
[email protected] their weightings were obtained from single improved by combining their best elements in a
©2013 The British Editorial
cohort studies, which means that they may not manner similar to the combined staging system
Society of Bone & Joint perform as well in a new group of patients.7 for trophoblastic disease.17
Surgery
doi:10.1302/0301-620X.95B2.
Five studies have validated the revised The aims of the current study were: 1) to
29323 $2.00 Tokuhashi and the Tomita scoring systems in a compare the variation and discriminative abil-
Bone Joint J
mixed group of patients with spinal metasta- ity of three prognostic scoring systems for
2013;95-B:210–16. ses.8-12 Two studies compared seven scoring patients with spinal metastases, the revised
Received 5 February 2012;
Accepted after revision 5
systems, including the revised Tokuhashi and Tokuhashi, the Tomita and the modified Bauer
December 2012 Tomita systems, but found that only a score score; and 2) to develop an improved,
Table I. Overview of patient characteristics formulated. Robustness of the new index was investigated
Characteristic and its parameters optimised using leave-one-out cross val-
Patients (n) 199 idation and a shrinkage procedure.22 The new index was
Female (n, %) 81 (41) then internally validated by a bootstrapping method23
Male (n, %) 118 (59) using 150 bootstrap replications.
Mean (SD) age (yrs) 61.6 (12.5) All statistical analyses were performed with R version
Primary site (n, %)
2.13.0 (R Foundation for Statistical Computing; University
Breast 33 (17)
Prostate 31 (16) of Vienna, Vienna, Austria) using the packages Survival,
Bronchus 20 (10) Penalized, rms and boot (all R software). A p-value of
Kidney 18 (9) < 0.05 was assumed to denote statistical significance.
Myeloma 18 (9)
Adenocarcinoma (unknown origin) 13 (7)
Lymphoma 12 (6)
Results
Others (< 10 each) 54 (27) Of the 199 patients, 118 were male and 81 female, with a
Treatment (n, %) mean age of 62 years (23 to 86). The most common site of
Surgical 104 (52) the primary tumour was the breast (n = 33) and over half
Conservative 95 (48) the patients were treated surgically (Table I). At the time of
evaluation, only 18 patients were still alive, with a maxi-
mum survival time of 14 years. The median survival time
for the whole group of patients was six months.
internally validated scoring system by combining the most Each of the three scoring systems was significantly asso-
predictive items of these scores. Although the modified ciated with survival time (Table II, Models 1 to 3). All sys-
Bauer system has been derived to predict survival of surgi- tems explained approximately the same amount of
cally treated patients, we included it in this study because it variation in survival times (Nagelkerke’s R2 = 0.15 to 0.17),
proved the most predictive in comparative studies.8,9 and were equally good at comparing survival between
patients (c = 0.64 to 0.67) (Table II).
Patients and Methods We then analysed each individual scoring item in a uni-
After obtaining ethical approval, a cohort of 199 consecu- variable Cox regression. The Tomita and modified Bauer’s
tive patients with spinal metastases was used for this study. scoring items for primary tumour were highly associated
Survival of patients was evaluated in September 2010. with survival (Nagelkerke’s R2 = 0.18 and 0.19, respec-
The clinical parameters recorded included age at presen- tively). The scores for general condition and visceral metas-
tation, gender, primary tumour, staging information and tases were also highly associated with survival. Age and
individual items used in the three scoring systems. The deci- gender were not associated with survival (Table III). When
sions in relation to treatment were also recorded. using gender, age and all twelve dissimilar items of the three
Patient survival was analysed using Cox regression and scoring systems as independent variables in a multivariable
age, gender, the individual items of the three scoring systems, Cox regression analysis, we found five parameters signifi-
and their total scores as independent variables. Two overall cantly associated with survival, namely Tomita’s primary
measures of performance were used to compare the various tumour score, the general condition score, the modified
parameters. The first was a generalisation of the coefficient Bauer’s score for absence of lung cancer, revised
of determination (Nagelkerke’s R2), a measure of the Tokuhashi’s extra-spinal bone metastasis score, and any of
explained variation in survival times.7,18 It captures aspects the major internal organ metastasis scores (Table II,
of calibration such as how well predicted risk agrees with Model 4). These five parameters explained a larger propor-
actual risk, and discrimination such as how well the pre- tion of the variance in survival times than any of the three
dicted ranking of patients in terms of risk agrees with their original scores (Nagelkerke’s R2 = 0.31; Table II). The com-
actual ranking.19 The second was the concordance ‘c’, a bination of these five parameters represents the best-fit full
measure of discriminative ability of the individual predictive model for predicting survival.
parameter.7,20 The concordance estimates the probability Since the score components for metastasis to viscera/
that, of two randomly chosen patients, the patient with the internal organs or the bony skeleton had smaller individual
higher prognostic score will outlive the patient with the coefficients of determination (Nagelkerke’s R2 ? 0.05) than
lower score, and is a generalisation of the area under the the other three (Nagelkerke’s R2 > 0.05), we also fitted a
receiver operating characteristic curve (AUC).7 Since calibra- model using only these last three predictors. A model with
tion and discrimination are both important in prognostic these predictors would not rely on extra investigations to
risk scores,21 we mainly used Nagelkerke’s R2 to compare predict survival. This model, based only on Tomita’s pri-
the scoring systems. Confidence limits for these two meas- mary tumour score, the general condition score and the
ures were determined using 150 bootstrap replications. absence of lung cancer in the modified Bauer score,
By combining the best-performing individual items as explained 27% of the variation in survival times
judged from the Cox regression, a new risk index was (Nagelkerke’s R2 = 0·27; Table II, Model 5). Shrinking
Table II. Cox regression analysis of the three existing prognostic models (Models 1 to 3) and three new models that are based on items from the
three existing models (Models 4 to 6) (CI, confidence interval)
Model* Coefficient Hazard ratio (95% CI) p-value Nagelkerke’s R2 (95% CI) c (95% CI)
2
1. Revised Tokuhashi score 0.83 (0.79 to 0.88) < 0.0001 0.18 (0.08 to 0.26) 0.67 (0.65 to 0.73)
2. Tomita score1 1.22 (1.15 to 1.30) < 0.0001 0.17 (0.08 to 0.26) 0.65 (0.62 to 0.71)
3. Modified Bauer score4 0.63 (0.54 to 0.74) < 0.0001 0.15 (0.06 to 0.23) 0.64 (0.61 to 0.70)
5. Primary tumour and general condition 0.27 (0.17 to 0.34) 0.68 (0.64 to 0.72)
- Primary tumour (Tom.) 0.33 1.38 (1.21 to 1.58) < 0.0001
- General condition (Tok.) -0.38 0.68 (0.56 to 0.83) 0.0002
- No lung cancer (Bauer) -0.69 0.50 (0.29 to 0.87) 0.01
6. Oswestry Spinal Metastasis Risk Index (OSRI) 1.91 (1.64 to 2.23) < 0.0001 0.28 (0.20 to 0.37) 0.67 (0.64 to 0.72)
* Tom., taken from Tomita system1; Tok., taken from revised Tokuhashi system2; Bauer, taken from Bauer system4; any, taken from any of the three
existing scoring systems; extrasp., extra-spinal; mets, metastases
Table III. Univariable analysis of individual items in the scoring systems (CI, confidence interval)
Tomita
Primary tumour grade 1.50 (1.33 to 1.68) < 0.001 0.19 0.63
Metastases to vital organs 1.14 (1.06 to 1.24) < 0.001 0.05 0.58
Bone metastases including spine 1.32 (0.95 to 1.85) 0.10 0.01 0.53
Modified Bauer
No vital organ metastases 0.76 (0.65 to 0.89) < 0.001 0.05 0.58
No lung cancer 0.23 (0.14 to 0.38) < 0.001 0.12† 0.56‡
Primary tumour breast, kidney, lymphoma, multiple myeloma 0.49 (0.36 to 0.64) < 0.001 0.10† 0.60‡
One solitary skeletal metastasis 0.76 (0.54 to 1.06) 0.09 0.01 0.53
Others
Age 1.01 (1.00 to 1.02) 0.12 0.01 0.55
Gender (male) 1.25 (0.92 to 1.68) 0.15 0.01 0.54
* KPS, Karnofsky Performance Status
† the combined R2 for the two Bauer primary tumour items is 0.18
‡ the combined c for the two Bauer primary tumour items is 0.63
the coefficients such that they maximised the fit using constructed a simplified Oswestry Spinal Metastasis Risk
leave-one-out cross validation mainly affected the coeffi- Index (OSRI) ranging from one to seven by adding the three
cient associated with the absence of lung cancer (optimised parameters, after inverting the general condition score. The
coefficients 0.32, -0.35 and -0.43 respectively for the OSRI is a simple summation of two elements: primary
three items). tumour pathology (PTP) and general condition (GC): OSRI
Based on the coefficients of the optimised Cox model, = PTP + (2 – GC). The reversal of the general condition
which were essentially equal for the three parameters, we score was needed because the Tomita score is a risk index
Survival
Moderate growth Kidney, uterus, tonsils, epipharynx, 2
synovial cell sarcoma, metastatic
thymoma 0.4
Rapid growth Stomach, colon, liver, melanoma, 4
teratoma, sigmoid colon, pancreas,
rectum, unknown origin
0.2
Very rapid growth Lung 5
influenced by sample size, whereas statistical significance on The visceral metastases item was nearly identical for all
its own is considerably influenced by sample size. Another scoring systems, and significantly associated with survival
difference may be the nature of the studies. Our study was a in our study, albeit with a low coefficient of determination
prognostic study of 199 patients. One earlier study was also (Nagelkerke’s R2 = 0.05). Although skeletal metastases
prognostic but smaller (69 patients),8 making it more diffi- items differed between the systems, its association with
cult to generalise their findings. The other comparative study survival was weak for all (Nagelkerke’s R2 ≤ 0.02). The
was larger (254 patients) but retrospective.9 From the larger importance of visceral metastases compared to
451 patients eligible for that study, almost half (197) were bony metastases in predicting survival in this study con-
excluded as a result of loss to follow-up or missing data.9 In curs with other studies.4-6,28
addition, our study has a minimal treatment bias (surgical Clearly from the above, the similar overall ability of the
treatment ≈ 50%), and a more homogenous spread of data scoring systems to predict survival hides major differences in
across various types of primary tumour. performance between individual items within each system. It
The differences between scoring systems become clear is therefore not surprising that prediction of survival
when comparing the variation and discriminative ability of improved after combining the most predictive items
their individual items. For each system, only one or two (Nagelkerke’s R2 = 0.31 vs Nagelkerke’s R2 = 0.15 to 0.18
items had a strong relationship to survival times for the original systems). Two specific items from the revised
(Nagelkerke’s R2 > 0.05). The type of the primary tumour Tokuhashi system, the number of affected vertebrae and the
had a strong correlation with survival times for all scores. neurological condition, were not significant in multivariate
This is not surprising given the wide differences in survival analysis, yet their importance has been stressed.1,3,29 The
between various cancers,24 which is most likely to be related most probable explanation for the disappearance of these
to factors such as different probabilities for lethal spread of items on multivariate analysis is their link to better predic-
cells from various tumours.25,26 Clear differences do, how- tors. Any patient with a neurological deficit is also likely to
ever, exist between the systems in assigning scoring values to score poorly on general condition. Likewise, the presence of
different tumours. As a result of these differences the associ- multiple spinal metastases is likely to correlate with the pres-
ation between the type of tumour and survival varied from ence of extra-spinal bony metastases. Nevertheless, this
Nagelkerke’s R2 = 0.04 (revised Tokuhashi) to 0.19 (Tomita). information is important for surgical planning.
There may be two main reasons for the different perfor- Determining visceral and skeletal metastases is impor-
mances of primary tumour items, namely the number of tant for staging, but requires extensive investigations that
tumour groups and the weighting of each group. The two may not be available in urgent situations. With radiographs
best performing items (Tomita and modified Bauer) distin- and mammograms, lesions smaller than two millimetres are
guished between a small number of tumour groups and had likely to be missed.30,31 The OSRI excluded information on
weightings derived from a Cox regression analysis.2,4 The these metastases and predicted survival almost as well as
worse performing revised Tokuhashi item distinguishes six the best-fit full model (Nagelkerke’s R2 = 0.28 vs 0.31),
groups,1 increasing the risk of misranking specific tumours. meaning the hardest-won information had the smallest
Other authors have indeed pointed out that ranking of spe- yield in terms of prediction of survival.
cific tumours in the revised Tokuhashi system should be Tumour cells from metastases contribute identically to
changed, especially unknown primaries, lung and breast the risk of dying as those from the site of the primary
cancers.13-15 Moreover, it is not clear how the weightings in tumour.32 Their exact contribution depends on the type of
the revised Tokuhashi tumour item were derived. Since primary tumour, and is, for instance, larger for melanoma
Tomita’s primary tumour item formed the main basis of the than for breast carcinoma.32 This explains the importance
primary tumour item in the OSRI, we retained Tomita’s of the site of the primary tumour in predicting survival.
labels based on tumour growth (Table IV) even though Nevertheless, why does information on metastases help so
other factors such as the probability of lethal spread of cells little in the prediction of survival for these patients? It is
from the tumour are considered more important determi- most likely that the effect of the extra volume of tumour
nants of the aggressiveness of a tumour.25,26 represented by metastases is overwhelmed by the large dif-
The second item that strongly correlated with survival ferences in survival from different types of tumour.
(Nagelkerke’s R2 = 0.11 is the patient’s general condition, Although tumour load is important in predicting survival,33
and is part of the revised Tokuhashi system. Earlier com- it is probably less helpful in a group with a wide variety of
parative studies of scoring systems for spinal metastases primary pathology, as seen in this study. In addition, all
found that a poorer general condition predicted better patients in this study had spinal metastases, and precise
survival.8,9 However, many other studies have found the information on the exact size of further metastases may be
same positive correlation between general condition and less relevant.
survival as our study.6,10,12,27,28 Tomita et al2 believed that The survival times of patients in adjacent risk groups in
the general condition was reflective of the status of the vis- the OSRI differ significantly, making predictions clinically
ceral metastasis, although they did not investigate this relevant. Although the difference in median survival times
proposed link. was small between some of the five risk groups, the time at
which 85% of patients had died doubled with each drop in quality of life in these patients. Similarly, when the progno-
category of risk. In our experience and that of others,34 sis is good, even radical surgery should not be withheld.
knowledge of this higher quantile is useful in discussions The OSRI is a useful tool, and the rates of survival quoted
with patients and decisions about management. We have in this series will help surgeons hold meaningful discussions
refrained from using the OSRI as a basis for specific with patients and relatives.
surgical recommendations, because treatment for each No benefits in any form have been received or will be received from a commer-
patient is highly individualised with due consideration to cial party related directly or indirectly to the subject of this article.
spinal stability, pain, neurological deficit and expected local
tumour control. Medical co-morbidities, personal, and References
social circumstances of the patients all contribute to the 1. 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)
decision making process. 2005;30:2186–2191.
Our study includes 40 patients (20%) with myeloma or 2. Tomita K, Kawahara N, Kobayashi T, et al. Surgical strategy for spinal metasta-
lymphoma. These patients have been excluded in two of the ses. Spine (Phila Pa 1976) 2001;26:298–306.
scoring systems1,2 but included in the modified Bauer 3. Tokuhashi Y, Matsuzaki H, Toriyama S, Kawano H, Ohsaka S. Scoring system
for the preoperative evaluation of metastatic spine tumor prognosis. Spine (Phila Pa
system4 and in studies comparing the scores.8,9 Although 1976) 1990;15:1110–1113.
the results presented in this study included these patients, 4. Bauer HC, Wedin R. Survival after surgery for spinal and extremity metastases:
the statistical analysis was also done after excluding them prognostication in 241 patients. Acta Orthop Scand 1995;66:143–146.
(data not shown). No noticeable difference in results was 5. van der Linden YM, Dijkstra SP, Vonk EJ, et al. Prediction of survival in patients
with metastases in the spinal column: results based on a randomized trial of radio-
found, in line with the initial studies.8,9 therapy. Cancer 2005;103:320–328.
A strength of the OSRI is its use of existing knowledge 6. Bartels RH, Feuth T, van der Maazen R, et al. Development of a model with which
from other scoring systems.7 This also represents a weak- to predict the life expectancy of patients with spinal epidural metastasis. Cancer
2007;110:2042–2049.
ness, as potential improvements have been omitted. For
7. Steyerberg EW. Clinical prediction models: a practical approach to development,
instance, we could have devised our own primary tumour validation, and updating. New York: Springer, 2009.
item that would outperform Tomita’s item in our patients. 8. Leithner A, Radl R, Gruber G, et al. Predictive value of seven preoperative prog-
However, doing so would increase the risk of poor generali- nostic scoring systems for spinal metastases. Eur Spine J 2008;17:1488–1495.
sation to new situations. Another way to improve the system 9. Wibmer C, Leithner A, Hofmann G, et al. Survival analysis of 254 patients after
manifestation of spinal metastases: evaluation of seven preoperative scoring sys-
would have been to measure other biological parameters tems. Spine (Phila Pa 1976) 2011;36:1977–1986.
such as the mitotic index.35 However, at the start of the study 10. Yamashita T, Siemionow KB, Mroz TE, Podichetty V, Lieberman IH. A prospec-
we were not in a position to decide on these. tive analysis of prognostic factors in patients with spinal metastases: use of the
revised tokuhashi score. Spine (Phila Pa 1976) 2011;36:910–917.
A second weakness of our new index is its lack of external
11. Tokuhashi Y, Ajiro Y, Umezawa N. Outcome of treatment for spinal metastases
validation. However, basing it on existing scoring systems using scoring system for preoperative evaluation of prognosis. Spine (Phila Pa 1976)
will reduce the risk of devising a scoring system purely for a 2009;34:69–73.
local situation. In addition, we have performed an internal 12. Pointillart V, Vital JM, Salmi R, Diallo A, Quan GM. Survival prognostic factors
and clinical outcomes in patients with spinal metastases. J Cancer Res Clin Oncol
validation of the OSRI using a bootstrapping procedure to 2011;137:849–856.
guard against over-optimism.7 None of the other scores have 13. Enkaoua EA, Doursounian L, Chatellier G, et al. Vertebral metastases: a critical
used such methods, which increases the chance that their appreciation of the preoperative prognostic tokuhashi score in a series of 71 cases.
Spine (Phila Pa 1976) 1997;22:2293–2298.
parameter values are over-optimistic and over-emphasise the
14. Ulmar B, Richter M, Cakir B, et al. The Tokuhashi score: significant predictive
contribution of less generalisable components. value for the life expectancy of patients with breast cancer with spinal metastases.
In conclusion, we have compared the ability of three dif- Spine (Phila Pa 1976) 2005;30:2222–2226.
ferent scoring systems for spinal metastases to predict and 15. Ogihara S, Seichi A, Hozumi T, et al. Prognostic factors for patients with spinal
metastases from lung cancer. Spine (Phila Pa 1976) 2006;31:1585–1590.
distinguish patient survival. Based on the most informative 16. Ulmar B, Naumann U, Catalkaya S, et al. Prognosis scores of Tokuhashi and Tom-
scoring items from these systems, we propose the OSRI. This ita for patients with spinal metastases of renal cancer. Ann Surg Oncol 2007;14:998–
is easy to calculate without reliance on expensive and time 1004.
consuming investigations, and outperforms the previous 17. Kohorn EI, Goldstein DP, Hancock BW, et al. Workshop Report: Combining the
staging system of the International Federation of Gynecology and Obstetrics with the
scores. For example: 1. A patient presenting with lung pri- scoring system of the World Health Organization for Trophoblastic Neoplasia. Report
mary (very rapid growth tumour) with a poor general condi- of the Working Committee of the International Society for the Study of Trophoblastic
Disease and the International Gynecologic Cancer Society. Int J Gynecol Cancer
tion score would have OSRI = 5 + (2-0) = 7, implying a 2000;10:84–88.
median survival of one month (85% chance of dying within 18. Nagelkerke NJD. A note on a general definition of the coefficient of determination.
2 months). 2. A patient presenting with myeloma (slow Biometrika 1991;78:691–692.
growing tumour), with a good general condition score would 19. Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of pre-
diction models: a framework for traditional and novel measures. Epidemiology
have OSRI = 1 + (2-2) = 1, implying a median survival of 23 2010;21:128–138.
months (85% chance of dying within 69 months). 20. Harrell FE Jr, Califf RM, Pryor DB, Lee KL, Rosati RA. Evaluating the yield of
Decisions regarding surgery in terminally ill patients pre- medical tests. JAMA 1982;247:2543–2546.
senting with spinal metastases should take into account the 21. Cook NR. Use and misuse of the receiver operating characteristic curve in risk pre-
diction. Circulation 2007;115:928–935.
morbidity of the proposed treatment as well as life expec- 22. Goeman JJ. L1 penalized estimation in the Cox proportional hazards model. Biom J
tancy. Unnecessary surgery should not jeopardise the 2010;52:70–84.
23. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in devel- 30. Spratt JS, Ter-Pogossian M, Long RT. The detection and growth of intrathoracic
oping models, evaluating assumptions and adequacy, and measuring and reducing neoplasms: the lower limits of radiographic distinction, the antemortum size, the
errors. Stat Med 1996;15:361–387. duration, and the pattern of growth as determined by direct mensuration of tumor
24. Jemal A, Clegg LX, Ward E, et al. Annual report to the nation on the status of can- diameters from random thoracic roentgenograms. Arch Surg 1963;86:283–288.
cer, 1975-2001, with a special feature regarding survival. Cancer 2004;101:3–27. 31. Spratt JS, Greenberg RA, Heuser LS. Geometry, growth rates, and duration of can-
25. Chen LL, Nolan ME, Silverstein MJ, et al. The impact of primary tumor size, lymph cer and carcinoma in situ of the breast before detection by screening. Cancer Res
node status, and other prognostic factors on the risk of cancer death. Cancer 1986;46:970–974.
2009;115:5071–5083. 32. Michaelson JS, Chen LL, Silverstein MJ, et al. Why cancer at the primary site
26. Rew DA, Wilson GD. Cell production rates in human tissues and tumours and their and in the lymph nodes contributes to the risk of cancer death. Cancer
significance. Part 1: an introduction to the techniques of measurement and their lim- 2009;115:5084–5094.
itations. Eur J Surg Oncol 2000;26:227–238. 33. Michaelson JS, Chen LL, Silverstein MJ, et al. How cancer at the primary site
27. Maltoni M, Caraceni A, Brunelli C, et al. Prognostic factors in advanced cancer and in the lymph nodes contributes to the risk of cancer death. Cancer
patients: evidence-based clinical recommendation: a study by the Steering Commit- 2009;115:5095–5107.
tee of the European Association for Palliative Care. J Clin Oncol 2005;23:6240–6248. 34. Kiely BE, Tattersall MHN, Stockler MR. Certain death in uncertain time: informing
28. Bartels RH, Feuth T, Rades D, et al. External validation of a model to predict the hope by quantifying a best case scenario. J Clin Oncol 2010;28:2802–2804.
survival of patients presenting with a spinal epidural metastasis. Cancer Metastasis 35. Gold JS, Gönen M, Gutiérrez A, et al. Development and validation of a prognostic
Rev 2011;30:153–159. nomogram for recurrence-free survival after complete surgical resection of localised
29. Sioutos PJ, Arbit E, Meshulam CF, Galicich JH. Spinal metastases from solid primary gastrointestinal stromal tumour: a retrospective analysis. Lancet Oncol
tumors: analysis of factors affecting survival. Cancer 1995;76:1453–1459. 2009;10:1045–1052.